RESEARCH DOSSIER
Insurance Comparison Website Ranking Fairness and Disclosure
Abstract. Insurance comparison websites organize access to heterogeneous contracts whose economic value depends on contingent payments, exclusions, eligibility, contractual obligations, and service quality. A comparison service reduces the procedural cost of obtaining quotes while also determining which products become visible, how their characteristics are represented, and which alternatives receive attention. This monograph examines the fairness of those determinations through insurance economics, platform incentives, consumer search, the structure of insurance indemnity, and comparative conduct regulation. It distinguishes market representation, risk eligibility, quote completion, coverage comparability, organic order, sponsorship, and purchase handoff as separately auditable operations. It develops a documentary synthesis of official supervisory findings and legal texts from the European Union, United Kingdom, Australia, and the United States, informed by empirical scholarship on insurance search and experimental ranking research. Worked examples demonstrate how a plausible ranking by premium may reverse under coverage restrictions or household-specific loss scenarios. Proposed audit procedures identify the evidence required to distinguish legitimate underwriting differences, technical failure, commercially motivated exclusion, and ranking-induced detriment. The conclusion is that ranking fairness requires fidelity between the platform's represented purpose and its operational decisions, substantiated by comparability and outcome testing throughout the insurance distribution chain. The research does not estimate the current prevalence of biased rankings, establish misconduct by any particular active platform, or claim a newly established scientific theory.
Keywords: Insurance comparison websites; insurance aggregators; ranking fairness; sponsored placement; insurance distribution; market conduct; price comparison; search costs; consumer protection; coverage comparability; quote eligibility; platform competition; commissions; disclosure; algorithmic audits; consumer understanding; insurance governance; fair value.
Introduction
The scientific object
An insurance comparison website is a digital intermediary that gathers, transforms, and presents information about insurance contracts from multiple providers in a form intended to facilitate comparison. An aggregator may retrieve individual quotes through technical interfaces, display general tariffs, refer customers to providers, transmit application data, or support conclusion of contracts. These activities occupy different positions in a distribution chain and have different informational and regulatory consequences. The term comparison website describes a functional setting, while the formal legal status of an operator depends on its jurisdiction and actual activities.
The ranking is an ordered presentation of a selected set of offers conditional on consumer characteristics, platform participation agreements, technical availability, underwriting decisions, contractual attributes, and the ordering rule. An order is meaningful only relative to its purpose. Ordering by annual premium answers a different question from ordering by expected financial protection, convenience, contractual breadth, or preferences entered by the customer. A placement purchased by a provider answers a commercial allocation question. Each can exist within a comparison service, provided the relationship between the displayed result and its meaning is clear and the underlying practice meets applicable conduct obligations.
Fairness concerns consumers, providers, and the integrity of competition. Consumer-facing fairness addresses understandable comparisons, consistency with expressed needs, commercial influence, and avoidable harm. Supplier-facing fairness concerns access and treatment according to relevant criteria and the service promised. Competition concerns the consequences of platform power and contracts across the market. Separating these questions prevents two common analytical errors: inferring wrongful discrimination from every differential rank, and interpreting competitive provider access as proof that a consumer has obtained suitable protection.
Insurance makes these questions particularly consequential. A quoted product is a contract over future uncertain states. Consumers typically cannot observe the claims response of each alternative before selecting one. A low premium can reflect a legitimate cost advantage, narrower insured events, a larger deductible, lower limits, different settlement conditions, or risk classification. The comparative signal is therefore necessarily multidimensional. A ranking which presents prices accurately while concealing differences in the protection purchased can still mischaracterize the decision.
Research questions and method
This study asks how remuneration, ownership, exclusivity, affiliate relationships, and platform optimization affect availability and ordering; how coverage matching and disclosure support informed choice; and which empirical designs and governance checks distinguish harmful steering from legitimate differentiation. Retail motor, home, travel, and selected health and term life applications serve as examples. The core object is comparison and ranking, rather than general sales-funnel attrition, all forms of digital persuasion, or affinity-group distribution.
The method combines comparative documentary analysis, a structured narrative review of empirical research, economic reasoning, insurance contract analysis, and hypothetical worked examples. Official statutes, consolidated directive text, handbook provisions, regulatory announcements, supervisory reports, and a tribunal docket have priority for institutional claims. Academic results are used at the level of material inspected. Where a full article was unavailable, the published abstract and bibliographic record constrain the claim. Regulatory studies are distinguished from legal requirements and industry observations. The source ledger records what was inspected and the scope of each result.
The research is an analytical synthesis, not a systematic review with a prespecified exhaustive search protocol and not an empirical audit of proprietary ranking logs. No new applications for coverage, consumer experiments, transactions, or insurer queries were executed. The proposed audit designs are prospective procedures; numerical examples specify synthetic contractual terms and inputs. Jurisdictional treatment is illustrative and version-bound rather than a substitute for national legal determinations. These boundaries permit explicit conclusions about mechanisms and evidentiary requirements without implying that a population estimate has been established.
Insurance Aggregator Market Structure and Commercial Incentives
Search economics and intermediation
A fragmented insurance market imposes several costs on a prospective policyholder. Finding providers, completing underwriting questionnaires, interpreting terms, negotiating or accepting a quote, and moving from an existing contract each consume resources. A comparison platform aggregates many of these tasks around a single interface. Standardized inputs can make additional quotes cheaper to obtain and can reduce the informational advantage enjoyed by a familiar insurer. However, the cost of interpreting contractual differences remains even when the cost of obtaining an extra price approaches zero.
Brown and Goolsbee's (2002) study of Internet comparison shopping relates the growth of Internet use in the 1990s to reductions in US term life prices using policy-level data and hedonic regressions. That result supports the economic plausibility of lower search friction intensifying competition under a particular institutional setting. It does not identify the present effect of any paid placement practice, and term life contracts and retail motor renewal behavior have different relevant attributes. The study's comparison with periods before websites entered the market and with other insurance types is particularly valuable as a reminder that market-wide technological change alone is an insufficient explanatory variable.
Honka's (2014) search and switching cost analysis models both consideration and purchase decisions in US auto insurance. A consumer can incur costs even after obtaining information, because changing an insurer is a separate decision. The monetization of a platform's ability to produce a completed switch therefore cannot be understood purely as payment for an additional quote. A ranking that favors providers with rapid digital completion may lower a real transaction cost, but that same incentive may disadvantage an appropriate policy whose referral journey is less immediate.
Economically, the intermediary supplies access and comparison to consumers and supplies qualified demand to providers. Both sides influence the value of participation. More participating insurers may make a service attractive to buyers; greater buyer traffic may make participation attractive to insurers. The value of each additional insurer depends on the distinct offers, underwriting capacity, geography, and needs served. Ten labels drawn from the same underwriting facility are not equivalent to ten independent carriers. The use of network effects as an explanation of platform consolidation requires evidence about the degree of substitution and entry, rather than a presumption that high traffic establishes market power.
Platform payment and selection
A comparison platform may receive a flat referral fee, per-click payment, qualified lead payment, commission per completed contract, performance bonus, advertising payment, or a combination of arrangements. Some platform owners have related providers or intermediaries. A public or membership-based comparison service may have other funding. Each arrangement changes which observable events have financial value.
Consider two prospective providers. Provider A pays a hypothetical 32 units for a bound policy and has a 25 percent probability of converting a qualified referral. Provider B pays 20 units and has a 50 percent probability. Expected gross revenue per otherwise equivalent referral is 8 for A and 10 for B. A platform optimizing paid conversions may consequently favor B even though A pays the larger unit fee. The example illustrates why commission level alone cannot establish the direction of commercial steering. Acquisition cost, reversals, contractual targets, retention payments, and the joint probability of selection and binding must also be considered.
These revenue calculations describe incentives rather than evidence of actual conduct. Performance-based remuneration can sustain a search service that consumers use without an upfront charge. The CMA's (2017) digital comparison tools final report recognizes commission-funded investment and acquisition efficiencies while examining the competitive pressure on commissions. A distribution channel can reduce insurer acquisition expense and simultaneously create dependency on its flow of potential policyholders. The sign and distribution of the efficiency gain depend on how provider competition and platform competition operate.
An economically meaningful ranking audit includes the choice of participants before it studies their ordering. A commercial agreement can restrict participation to suppliers that pay for leads. Network entry may instead be conditioned on licensing, solvency documentation, implementation cost, service standards, or the capacity to provide reliable quotes. A complete causal explanation distinguishes partnership exclusions, technical filtering, underwriting refusals, and post-quote ranking. It also distinguishes the consumer's universe of offers from the platform's contractual universe.
The feedback from ranking to insurance product design
Visibility can alter what insurers find profitable to offer. If competitive pressure is concentrated on the initially displayed premium, providers may redesign product bundles to meet a displayed price threshold. Reductions in operational expense and actuarially justified risk segmentation can benefit customers. Removing coverage that customers value, shifting charges into mandatory extras, increasing deductibles, or presenting optional features late in the purchase can transform apparent price competition into competition over presentation.
A provider's response may occur even if the ordering algorithm never uses remuneration. A platform's published comparison fields establish the attributes that compete for attention. When a consequential clause has no visible field, improving that clause may carry little competitive reward. When a quality score is decisive but opaque, providers may optimize toward the score while neglecting service dimensions that it poorly measures. The object of product governance thus extends to the information design used by distributors.
The platform's influence need not be ownership control of the policy. The insurer determines contract terms and underwriting; the intermediary controls conditions of access and representation within its service. An insurer that supplies incomplete data may be responsible for the incompleteness, while a platform that knowingly maps an unknown field to included coverage creates another source of harm. Effective governance allocates these responsibilities jointly through data specifications, monitoring, correction, and escalation.
Ranking Algorithms, Quote Eligibility and Coverage Matching
The sequence from market to visible result
A consumer profile contains declared risk facts and decision preferences. The profile enters a quote process where insurers apply eligibility and underwriting rules, and participating intermediaries respond with offers or status information. The platform normalizes returned data, checks validity, filters by product requirements, creates an order, and presents a subset in the available interface. A separate handoff may cause the provider to revise or withdraw a quote after additional information.
Each operation requires observability. Suppose a panel contains 22 nominal insurance brands, 17 respond to a particular query, 12 return a valid offer, 8 meet the customer's recorded coverage constraints, and 5 appear on the first screen. The counts refer to different populations. The 5 visible results cannot be interpreted as the entire panel and the 8 feasible offers cannot be interpreted as all offers in the market. Distinct underwriting groups must be counted separately if diversity is a stated objective. A quote request timeout belongs to a third category besides rejection and valid offer: it is an unknown response and should remain unknown until resolved.
The result set also changes through time. An insurer may amend a rate, remove a product, change eligibility, or correct a data field. A comparison should maintain a quote timestamp, validity period, jurisdiction, version of contract wording, and the exact input basis. If quotes arrive asynchronously, the system can initially show results as provisional or wait for a stated period. Reordering already displayed rows while a consumer tries to select one introduces a distinct interaction problem. An audit should capture initial and settled orders, the delay until a quote arrives, and whether late quotes were visible before a selection was made.
Ordering and relevance
An organic ordering is an ordering represented as arising from price, product characteristics, or another declared comparison criterion without preferential purchase of its displayed position. A sponsored ordering or placement allocates prominence for consideration. An organic order can nevertheless be indirectly influenced by commercial participation restrictions or by a learned conversion predictor. Representational accuracy requires checking both the narrow rank rule and the upstream candidate set.
Price ordering is an exact operation once the comparable price definition and candidate set are established. A sorting rule should reproduce the advertised ordering for all eligible offers, including tie-breaking and handling of missing prices. A multiattribute ranking has greater discretion because it must specify how different product dimensions are assessed and combined. In an ordinary weighted score, differences in scale and missing fields can dominate the intended weights. A deductible adjustment in dollars, a five-point service rating, and a coverage flag cannot be added meaningfully without normalization and an explicitly defended trade-off.
Predictive ranking can estimate click-through, purchase likelihood, predicted preference, or a target linked to consumer benefit. Training labels are outcomes of past exposure and prior ranking decisions. Products that previously appeared first received more opportunities to generate positive labels. A model learned from such observations may reproduce exposure patterns even when no commission variable is present. Position-aware experiments, logging of candidate sets, and methods that correct for known exposure probabilities can reduce this problem. They cannot recover effects for products or populations with no support in the data.
In an insurance setting, there is a substantive distinction between a consumer's expressed interest in a feature and an underwriter's determination that a risk is eligible. A retail customer may prefer cover for an event that a given insurer will not cover. A ranker should not promote the unsuitable contract merely because a predictive model assigns a high probability of selection. Hard product requirements, such as a legally required coverage or a binding customer constraint, belong in qualification and validation before optimization. Preferences permit trade-offs; eligibility and mandatory requirements constrain the available alternatives.
Contractual matching and expected financial exposure
A policy representation should retain the dimensions material to the risk. For motor insurance these may include third-party limits, own-damage coverage, applicable deductibles, use restrictions, and settlement basis. For home insurance, insured perils, replacement valuation, sublimits, catastrophe exclusions, and occupancy conditions matter. For travel insurance, territorial scope, covered trip duration, medical exclusions, cancellation triggers, and pre-existing condition treatment can be decisive. Specifications differ by geography and product. Normalization should preserve both the comparable label and source wording.
The value of an indemnity policy can be examined using the ordinary notion of expected premium plus retained loss. Let a consumer face loss L in a period, and let B_j(L) denote the monetary amount payable by policy j in the given scenario, after applying applicable deductibles, exclusions, and limits. If annual premium plus compulsory charges is P_j, gross financial outlay including retained loss is P_j + L - B_j(L). The expectation over stated loss scenarios compares mean outlay; the full distribution of retained loss matters because an insurance contract transfers risk, and households have different resources and risk preferences.
Consider three hypothetical annual offers for a homeowner and a specified peril. No actual rates or wording are represented.
| Offer | Annual premium | Deductible for peril | Limit | Peril covered |
|---|---|---|---|---|
| A | 320 | 500 | 20,000 | No |
| B | 390 | 1,000 | 20,000 | Yes |
| C | 450 | 250 | 20,000 | Yes |
Assume probability 0.05 of one loss of 4,000 and probability 0.95 of no loss. No other risks, loads, add-ons, contractual provisions, or expenses enter the illustration. With A the retained loss in the event is 4,000; with B it is 1,000; with C it is 250. Mean outlay is respectively 520, 440, and 462.50. The premium-only order is A, B, C. Mean outlay reverses the first two positions and places C ahead of A. For a household unable to absorb a deductible of 1,000, C can be preferable to B despite the higher mean outlay. If the specified peril is a hard requirement, A should not qualify. The numbers show why a universal cheapest or best label cannot be inferred from a single price scalar.
The reasoning also identifies the limits of coverage adjustment. Expected loss inputs are uncertain and can be household-specific. A severe tail event may dominate financial vulnerability without dominating its expected value. Service delays, bodily injury, legal expenses, nonfinancial loss, and risk preferences require separate treatment. Practical comparison should display contractual differences, preserve uncertainty and allow explicit trade-offs rather than manufacturing a false precision score.
Dominance provides a narrower and stronger comparison when two genuinely equivalent contractual offers are available to the same consumer at the same time. If one costs less and is at least as favorable on every decision-relevant dimension, with a strict improvement in at least one, it dominates the other for consumers with monotonic preferences over those dimensions. In practice, unobserved service quality or differences in wording can make a claim of complete dominance unsafe. Conditional dominance under explicitly stated attributes is a useful audit flag, not proof that every rational consumer would choose the cheaper row.
The historical FCA (2014) thematic review of comparison websites examined issues including core policy information, excesses, add-ons, and travel insurance medical conditions. The report's later cover labels it historical. Its continuing analytical relevance is the distinction between a price signal and the underlying promise; its historical regulatory citations do not substitute for present applicable rules.
Sponsored Placement and Conflicts of Interest
Commercial influence and representational harm
The central sponsorship distinction is between paying for access to a platform's audience and paying for a position that consumers interpret as a comparative judgment. A labeled advertisement can finance a service and provide relevant information. A paid entry represented as the best independent result can cause consumers to attribute significance to its position that the ordering does not possess. Where a payment conditions candidate inclusion rather than rank, a statement about the basis of organic sorting can be accurate and still leave the represented breadth of comparison misleading.
Commercial influence can have several forms: a fixed sponsored slot, a bid-weighted ordering, commission as a scoring feature, revenue-based tie-breaking, exclusive supplier participation, preferential data latency treatment, and optimization of predicted commissions via proxy variables. A technically neutral sorting function after commercially biased filtering is insufficient evidence of an impartial comparative process. The converse also deserves attention: a related insurer may achieve top position under genuinely comparable premium sorting. Ownership does not establish that the order was manipulated.
The FCA (2014) announcement of its historical review documents incomplete conflict disclosure by some group-owned sites while reporting no evidence that those firms used the relationships to gain commercial advantage. This distinction is essential. A disclosure deficiency establishes a problem with representation within the investigated facts, not an evidentiary basis for a broader allegation of manipulation. Auditors must determine whether commercial exposure and outcome differences survive controls for eligibility, price, product relevance, quote latency, and availability.
Related-party contracts deserve close attention where platform management has compensation tied to the group's aggregate earnings. A platform may be commercially rewarded for favoring related insurers even without a transaction-specific payment. Data access, prominent defaults, faster integration, and a greater opportunity to return a modified quote can shift visibility. Review should identify the economic benefits attached to each decision, the decision-makers and their oversight, and the actual code or configuration implementing those decisions.
Disclosure as one component of governance
Disclosure should be timely, comprehensible, proximate to the decision, and sufficiently specific to explain the practical influence at issue. One statement that providers may compensate the site fails to communicate whether a particular offer is an advertisement, a paid ordering, or an organic result from a limited panel. A clear description of general funding is useful alongside result-specific labeling and an explanation of ranking factors. The user should be able to distinguish paid visibility from the advertised comparison logic without opening contractual terms.
The FTC's (2015) native advertising business guidance distinguishes the deception inherent in disguised advertising from the accuracy of the advertisement's product description. Its reasoning supports evaluating the visual and semantic cues surrounding a listing, with jurisdiction and legal perimeter determined separately. The fact that a promoted policy really costs the displayed price does not establish that a consumer recognizes its commercial placement.
Sponsor disclosures should be evaluated by comprehension, not only formal presence. A placement label can be noticed yet misunderstood if surrounding icons or descriptors suggest superior value. The interface may use a separate labeled sponsor region and a stable organic order. Where a sponsored offer also qualifies organically, a consumer can be shown its organic position or the implementation can avoid duplicate representations. The choice depends on preserving the comparative meaning and the ability to locate feasible alternatives.
Conflict mitigation may require removing remuneration from organic order, restricting paid products to declared placements, documenting limits on commercial overrides, and independently reviewing candidate-inclusion rules. Transparency alone does not resolve a prohibited conflict or an unsuitable distribution practice. A ranker can behave exactly as disclosed yet place a person in a foreseeable situation in which essential coverage disappears. Conversely, categorical prohibition of revenue-supported intermediation without analysis of its effects could reduce useful information and intensify acquisition costs.
The appropriate unit of evidence is a versioned decision. A defensible sponsorship log identifies the placement type, payment basis, contractual relation, product and quote IDs, organic rank absent the placement, display rank, label presented, and conditions of the user's query. These fields enable reconciliation between the technical decision and what the user was told, while limiting unnecessary retention of personal risk details.
Disclosure of Comparison Scope and Product Features
Market coverage and the denominator problem
A comparison claim has a numerator and a denominator. A platform can count the providers on its partnership panel, brands that sent an offer during a period, unique licensed underwriting companies, or products accessible to a profile. Each is a different quantity. An assertion about the whole market requires an independently defensible market definition, reference date, and method for enumerating available providers and products. Counting nominal labels while omitting direct-only distribution can create the impression of broad coverage without demonstrating it.
The CMA (2017) CARE principles for comparison tools address the business model, scope, ownership, ordering, effects of commercial relationships, and compulsory costs. These areas represent complementary information rather than interchangeable disclosure categories. An explanation that rank is based on price does not tell the consumer how much of the market has been searched. A market-coverage statement does not tell them whether sponsored position is mixed into the organic ranking.
A historical Australian example demonstrates the independent importance of the coverage claim. The ACCC (2014) reported an infringement notice and payment of AUD 10,200 concerning advertising of a health insurance comparison service's breadth, in circumstances where other websites compared policies from more health funds. The payment was not an admission of contravention. The case concerns a representation about the comparator service, regardless of whether any individual row was accurately priced.
A study of breadth should record the market scope (jurisdiction, class, distribution channels, relevant date), the partner panel, unique carriers, eligible conditional panel, successfully queried offers, and displayed offers. A sample of direct insurer quotes can estimate missed opportunities under defined assumptions but does not establish completeness. The exact denominator may be unknown where rates are personalized and products are restricted. A qualified statement about panel coverage is preferable to a generalized market-wide percentage when the independently observable universe is incomplete.
Product features, total price, and temporal validity
The comparative representation should distinguish a quote from a binding contract. An indicative quote may be conditioned on the truth and completeness of the entered information, risk assessment, evidence, and the time of completion. The comparison platform's schema must accommodate not only numeric prices but also conditions and quality states. A field with missing data is not the same as an exclusion. A field labeled included must refer to a verified entitlement under the applicable wording.
Annual premium, installment amounts, compulsory fees, tax where applicable, payment-plan charges, and separately selected add-ons should have consistent treatment. A cashback may reduce the effective cash price under specified conditions while leaving the contractual premium unchanged. A retail voucher is not identical to cash if it carries restrictions or differs in personal value. Farghly, Spohn, and Ng (2021), in an FCA-hosted experiment on insurance promotions, report that certain promotions impaired deal selection and premium comprehension in simulated purchase decisions. The result supports testing the representation of promotions; it does not quantify their effect within any particular ranking algorithm.
The FCA's (2014) general insurance add-ons market study used behavioral experiments alongside market information and reported consumer detriment associated with add-on sales. In a comparator, a product excluding a necessary feature can look cheaper before the feature is repurchased, especially if an optional price is revealed after the consumer invests time in a selected offer. The appropriate comparison basis is stated explicitly: equivalent required cover, total cost of selected features, and a separate view of additional optional benefits.
A hypothetical pair illustrates the distinction. Offer D advertises 240 per year, excluding a required legal-expense option costing 45 and imposing a mandatory payment fee of 20. Offer E costs 295 including the same option and fee. On a like-for-like assumed basis the totals are 305 and 295 respectively. D has the lower initial figure and the higher total cost. If legal-expense cover is only a preference rather than a requirement, both the base and configured totals can be shown without implying that the unconfigured contract is equivalent.
The standardized insurance product information document under the EU distribution regime assists in making key information accessible before concluding non-life contracts. A comparison schema should link to the corresponding document, preserve its version, and record how compressed fields derive from its content or full policy wording. Standardization facilitates comparison but cannot substitute for contract-specific qualification of eligibility, limitations, and exclusions.
Human understanding and accessibility
The platform's service description matters because users may attribute a degree of personalization to automated results that the service has not performed. Which?'s (2025) research on consumer confusion includes UK survey and qualitative evidence of users' trust in comparison services and simplified understandings of insurance. These observations describe beliefs, not actual proof that the trusted products met individual needs. The implication is an evaluation question: can a customer accurately explain the scope of the search, the role of the ranking, and at least the differences material to the stated insurance task?
Consider a mobile user who sees only a brand, premium, and purchase button. Relevant deductible and coverage details might be accessible in a secondary view. If the service's advertised ordering implies that the initial card is the preferred option, the user may have already formed a choice before reaching the detail. A comprehension study should vary presentation timing, space constraints, label language, and navigation friction; it should assess the ability to identify limitations as well as whether users complete the task.
Accessibility requires usable results for consumers with diverse needs, including adequate contrast and readable text, keyboard access, screen-reader labeling, understandable terminology, alternative assisted pathways where appropriate, and stable control states. In the 2025 fair-treatment paper announcement, the IAIS links consumer diversity to the insurance product lifecycle and recognizes the continuing role of risk-based pricing and insurer autonomy. For an aggregator, diverse consumer circumstances make testing the disclosure and search process an essential complement to average efficiency metrics.
Information should be layered according to the decision: a short basis of comparison at the ranking, prominent limitations, a consistent comparison detail, and authoritative contract documents. Excessive undifferentiated data can increase search costs, so the objective is actionable understanding rather than the maximal number of fields on a screen. The platform can offer side-by-side comparison without reproducing every clause, provided critical differences are represented and the underlying materials remain accessible.
Consumer Choice Experiments and Ranking Fairness Metrics
What empirical research identifies
An observed click is generated by both product attributes and its exposure. A premium difference, a familiar brand, position, interface presentation, and a campaign may influence the same observation. If high-quality offers tend to appear first, a simple correlation between rank and click-through combines quality selection and positional effects. If low-premium offers tend to appear first, the correlation combines price and exposure. The research task is to separate these factors rather than to label the entire association a rank effect.
Ursu (2018), in The Power of Rankings, uses experimental rank variation from an online travel intermediary. The published abstract describes a causal analysis of search and a structural explanation through lower search costs. The finding supports an identification strategy for ranking experiments, not numerical extrapolation to insurance. An insurance product's quality is partly realized after a contingent event; this changes the relevance of observed clicks, conversion, and immediate satisfaction.
The CMA (2022) online choice architecture evidence review discusses position, limited reordering, and commercial rank incentives, synthesizing several settings. Position can help consumers by placing relevant products near the top. A positional effect becomes a concern when the allocation of attention conflicts with the represented comparison or steers selection toward worse options. Randomization, where ethically and legally permissible, can distinguish the value of rank from the value of the product.
The key insurance outcome is not simply choice of the lowest premium. Evaluation can measure understanding of terms, satisfaction of hard requirements, comparable total expenditure, selected deductible and limits, unexpected handoff changes, cancellation, and appropriately interpreted claims outcomes. Different time horizons address different harms. A rapid purchase can be efficient; a later surprise at claim time can reveal a deficiency invisible to checkout measures.
Counterfactual experiments and statistical inference
A proposed consumer experiment should define a representative product class and a task with explicitly specified coverage needs. Participants receive a common set of hypothetical or properly validated offers. Random assignment can vary the ordering, sponsorship label, or disclosure design while holding the underlying contracts fixed. The dependent variables include the selected contract, comprehension questions, time spent, and cost or coverage errors. Research participants should be protected from real unsuitable purchases, deceptive inducements, and unnecessary collection of sensitive information.
A factorial design can vary two binary treatments: whether the default order is premium-only or applies stated hard requirements before sorting; and whether sponsorship labeling is minimally present or prominently explained. The four groups estimate separate treatment effects and their interaction. Ethical design should use simulated decisions where a deliberately inferior disclosure is under study. If a real-world randomized rollout assesses improved disclosures, the control must meet applicable legal and consumer-protection requirements. The causal estimand is defined for the assigned policy, not only among consumers who clicked.
For illustration, suppose a simulated experiment randomly assigns 400 participants per arm to two lawful candidate interfaces. In the first arm, 128 select a contract that violates the task's explicit coverage condition; in the second, 92 do. The proportions are 0.32 and 0.23. The unadjusted difference (second minus first) is -0.09, or nine percentage points. Under independent binomial sampling, the estimated standard error is the square root of (0.32 × 0.68 / 400 + 0.23 × 0.77 / 400), approximately 0.0314. A simple normal-approximation 95 percent interval is about -0.152 to -0.028. All data are invented for methodological illustration; the interval is not a scientific estimate of any real comparison service.
In an actual trial, assignment may be by user, session, household, provider, or time block. Repeated sessions and shared underwriting responses can invalidate naive independence assumptions. Clustered standard errors or randomization-based inference should reflect the assignment unit. Prespecification of primary outcomes, treatment changes, exclusion criteria, and multiple-comparison handling protects against selective inference. Recruitment should include customers with varied financial literacy and insurance needs, and subgroup analyses should report denominators, uncertainty, and whether analyses were prespecified.
A counterfactual quote audit differs from a human trial. Researchers submit equivalent specified profiles across configurations, recording results without proceeding to bind. Safe, authorized sandbox or test accounts are preferable to fictitious live applications. The same consumer inputs, timestamp window, policy class, and provider quote versions should be used for comparisons. When an external insurer reprices during the experiment, that change must either be identified and modeled or the affected observation treated as noncomparable. A difference between an organic rank with and without a commission feature is meaningful only if the offer set and other scores are held constant.
Observational data can complement experiments. An audit might regress top-position exposure on commission, price, coverage attributes, provider, time, and risk profile. Such a coefficient is conditional association unless strong identification assumptions are justified. Provider remuneration can correlate with operational efficiency or the selection of participants. An unobserved distribution contract may influence both the returned quote and commission. A sudden contractual change can motivate a quasi-experiment, but parallel trends, spillovers, concurrent changes and anticipation must be examined before giving causal interpretations.
Measures and their interpretation
Ranking fairness has multiple operational questions. The following measures are conventional audit diagnostics applied to insurance rather than a proposed new theory. Let N be a stated collection of quote sessions. Let the feasible set in a session consist of valid quotes meeting the specified mandatory coverage and eligibility conditions. A consumer-benefit measure should be computed only where this feasible set and the attributes required by the measure are known.
| Diagnostic | Operational definition | Key limitation |
|---|---|---|
| Quote response rate | Valid responses divided by attempted provider requests | Nonresponse and underwriting rejection differ |
| Feasible offer coverage | Displayed distinct feasible offers divided by observed feasible offers | Observed universe may omit external market |
| Hard-requirement violation rate | Displayed offers violating a declared constraint divided by displayed offers subject to it | Unknown terms are a separate state |
| Top-position commercial exposure | Share of covered sessions with paid or related-party first placement | Does not identify manipulation |
| Like-for-like price regret | Selected total comparable cost minus minimum feasible comparable cost | Requires true equivalence on relevant terms |
| Rank displacement | Change in position of identical eligible offers under specified alternative ranking | Different candidate sets confound it |
| Disclosure comprehension | Proportion correctly explaining the ranking and commercial influence | Survey measurement and sample effects |
| Handoff discrepancy rate | Material unmatched price or contract changes divided by linked handoffs | Distinguish legitimate new underwriting information |
Denominators must be reported. If 1,000 sessions begin, 900 return valid offers, and only 600 have complete matched coverage and follow-through, comparable-cost regret over the last 600 does not measure harm across all 1,000. Missing outcomes may concentrate among vulnerable customers or providers with poor integration. Report the inclusion rate, patterns of missingness, sensitivity bounds when justified, and separate reasons for ineligibility or lost follow-up.
Exposure can be summarized with weights for position, provided the weights reflect a stated purpose. For example, in a synthetic audit assign weights 1, 0.5 and 0.25 to first, second and third positions. In three sessions a sponsor appears at ranks 1, 2 and 3, so its total weighted exposure is 1.75; the total available weight over three sessions is 5.25, giving one-third of weighted top-three exposure. These arbitrary weights illustrate calculation, not a universal model of attention. They should be calibrated to actual interface exposure or experimentally observed attention if used as empirical evidence.
Group comparisons require care. Insurance underwriting legitimately differentiates on some actuarially relevant factors within applicable legal constraints; regulated attributes and proxies require jurisdiction-specific treatment. Different groups may have different eligibility sets and product needs. Equality of raw top-provider outcomes is not a sufficient fairness criterion. The audit should compare similarly situated profiles for a particular question, track both unconditional access and conditional ranking, and assess whether disparities arise from underwriting, participation, technical errors, or ordering. Where sensitive attributes are necessary for testing, authorized privacy-preserving processes and appropriate legal grounds must be established.
An individual profile may receive no feasible offer. A metric requiring a best alternative should be undefined for that profile, with a separately reported no-match rate. Substituting zero regret for undefined regret would create false evidence of good performance. An appropriate platform response may communicate the gap in its panel, allow users to revise preferences that are genuinely optional, or direct them to other suitable channels without silently changing mandatory coverage.
Triangulation through later outcomes
Claims data can reveal mismatches between expected and contracted cover. Complaints may identify misleading order labels, misunderstood excesses, unexpected charges, inaccurate coverage fields, and problems with handoff. These outcomes should be linked to the version of the platform and policy shown at purchase, not just the most recent product page. Distinguish complaints about the insurance carrier's contractual claims decision from complaints about a comparator's representation; both may be involved in one experience.
Claims acceptance rates and complaint frequencies require appropriate exposure, product mix, policy duration, and reporting definitions. A low claims ratio can reflect lower-risk customers, exclusions, pricing, claim propensity, or service practices; it is not a direct measure of insurance suitability. The audit should combine indicators with targeted case review. A complaint sample can establish mechanisms of harm even when the number of complainants does not estimate population incidence. Feedback mechanisms should permit correction of quote and product metadata before another customer encounters the same defect.
Competition, Insurance Distribution and Consumer Law Comparison
International supervisory context
The IAIS (2024) Insurance Core Principles and ComFrame provide an international supervisory framework. Their structure differentiates principle statements, standards, and guidance. Conduct expectations need implementation through appropriate jurisdictional powers, institutions, and legal obligations. An application paper or supervisory framework does not itself make a particular website legally liable under every national system. The value of convergence lies in common supervisory questions about professional intermediary conduct, consumer communication, conflicts, and risk-based oversight.
Comparative law requires separation of the platform's actual conduct from its self-description; the legal instruments applying to that conduct; and the powers available to the competent authority. The same commercial website can operate differently in different countries or across different insurance classes. A platform that merely publishes generic information can fall within different rules from one that solicits individualized applications, transfers encoded data, and links to the conclusion of a particular contract. Market conduct and advertising controls can coexist with distribution licensing and competition rules.
European Union
The Insurance Distribution Directive (Directive (EU) 2016/97, consolidated text of October 9, 2024) is particularly relevant because Article 2 addresses online comparison and ranking linked to the possibility of concluding an insurance contract. Application depends on the activities and conditions described in the definition and the exclusions. EIOPA's (2024) comparison website perimeter answer discusses facilitation of specific contracts and the mere-information boundary. A formal characterization as a lead generator, directory, comparison tool, or technology provider does not resolve the classification without examining what the service actually does.
Article 17 sets a general conduct obligation and addresses misleading communications and incentives that conflict with customers' interests. Article 19 specifies intermediary disclosures including the nature of remuneration, with distinct treatment where a fee is paid directly by the consumer. Article 20 distinguishes demands-and-needs assessment and the additional explanation associated with advice, and provides for standardized information for relevant non-life products. Article 25 concerns product oversight and governance within its scope. These provisions show why a disclosure that a platform receives payment does not decide the separate issue of the suitability of a recommendation or incentive design.
Article 20's fair-and-personal-analysis rule applies where an intermediary represents that its advice is based on that standard; it is not a blanket mandate that every comparison website survey every product. A non-advised comparative service must nonetheless accurately describe its basis and fulfill applicable distribution obligations. Chapter VI contains additional requirements for insurance-based investment products. Retail general insurance should not be subjected by analogy to that chapter's precise duties without scope analysis.
Consumer-law transparency introduces a distinct issue. The European Commission's (2021) unfair commercial practices guidance explains ranking parameters, their relative importance, accessible presentation, and the prohibition on concealed payment specifically for better search position. The general description of key parameters need not disclose detailed algorithmic mechanisms. The rules' application depends on the instrument's scope and national implementation, including relevant financial-services provisions and later amendments. The guidance is an interpretive source, not a replacement for the operative legislation.
The practical synthesis is that the consumer's decision should reveal the service's legal and commercial nature, the basis of qualification and order, and the essential contractual features. A public ranking explanation can identify decisive factors and their relationship without publishing proprietary model weights or training data; a competent supervisor may require more granular evidence under its own powers.
United Kingdom
The FCA's historical thematic work supplies evidence of comparison-service risks, while the current Consumer Duty handbook provisions provide a different, operative regulatory perspective for firms and activities within scope. PRIN 2A sets retail customer outcomes concerning products and services, price and value, understanding, and support. The 2026 guidance on the distribution chain relates responsibility to the firm's role and capacity to influence outcomes. Value relates to benefits and payment, rather than a rule that only the cheapest product can be distributed.
The interaction of a platform and an insurer is salient where one controls ranking while the other controls terms, pricing, and underwriting. Distribution responsibilities should be allocated according to the role actually exercised. A platform's non-advised status does not eliminate applicable requirements concerning its communications and influence on outcomes. Conversely, the Consumer Duty does not create a fiduciary relationship or an advice obligation solely by its existence. Correct legal classification remains necessary.
The CMA's 2017 work is a market-study source; the principles articulated there guide interpretation and policy design, not a standalone legislative code. Competition proceedings require a different analysis from consumer disclosure. Wide retail price-parity clauses may restrict prices on rival platforms; narrow clauses generally relate to a supplier's own channel. Their consequences depend on actual terms, market conditions, entry, substitution, and efficiencies. A platform may argue that protection against free-riding supports investment in comparison services, while rivals may argue that parity prevents price competition. The net effect requires proof.
The CMA's 2023 concurrency report records that its 2020 Compare the Market home-insurance parity decision was set aside. The official tribunal case record identifies BGL (Holdings) Limited and Others v Competition and Markets Authority, [2022] CAT 36, judgment of August 8, 2022. The CMA's retrospective account states that the tribunal found the market definition materially wrong and proof of anticompetitive effect insufficient. The initial decision therefore cannot be cited as an operative finding of infringement. Nor does the reversal establish universal legality of wide parity agreements. The case illustrates why economic theory of harm, relevant market definition, and factual effects must be distinguished.
The CMA final report's distinction between visits to multiple tools and fully comparing obtained quotes matters for competition. A consumer who opens several platforms but enters extensive underwriting information on only one does not exert the same purchasing pressure as a consumer who obtains comparable offers from several. Evidence of traffic patterns cannot automatically establish effective platform substitution.
Australia and the United States
The ACCC's (2015) guide for comparator operators addresses operators and suppliers in the context of fair trading and the Competition and Consumer Act 2010. Its consumer information on comparison sites distinguishes limited commercial panels and ownership or monetary relationships from broader public services and emphasizes nonprice attributes. These sources support examination of misleading market-scope claims and commercial representation under Australian law. The relevant financial-services and state or sector-specific allocation of responsibilities should be determined separately for the actual conduct; this study does not treat the guide as the exhaustive regulatory framework for every insurer and product.
In the United States, insurance regulation is substantially jurisdiction-specific, and the federal role varies by activity. General federal advertising guidance is relevant to genuinely applicable activities, but the division of insurance authority, state producer requirements, product regulation, and possible limitations on federal jurisdiction prevent mechanically importing European distribution rules. This monograph therefore limits its American legal analysis to identified federal advertising materials and US insurance-search research, without purporting to determine the law of each state.
The FTC's October 2026 health-plan marketing warning addresses plan representations, including whether offerings constitute comprehensive health insurance, and the depiction of benefits and costs. The warning letters did not allege wrongdoing by their recipients. For a comparison platform that redirects individuals toward health-plan marketers, the boundary between a regulated insurance contract and a differently structured health product may itself be material to the consumer's selection. An apparent ranking among comparable plans can be misleading if the underlying offerings belong to different product categories and the distinction is obscured.
Comparative implications
| Setting | Identified instrument or source | Main issue for comparison |
|---|---|---|
| International | IAIS framework and 2025 paper announcement | Conduct and diverse consumer needs |
| European Union | IDD and Commission consumer-law guidance | Distribution perimeter, incentives, demands and needs, ranking disclosure |
| United Kingdom | FCA PRIN 2A; historic FCA and CMA reviews | Retail outcomes, fair value, transparency, competitive effects |
| Australia | ACCC comparator guidance and historical notice | Misleading scope and commercial representation |
| United States | FTC advertising guidance and health-plan warning | Identifiable commercial presentation and plan claims; state-specific perimeter |
The recurring concerns are comparable even where legal obligations differ. Classification precedes compliance analysis; disclosure, incentives, product suitability, and competitive effects remain distinct legal and economic questions. A supervisor should establish the instrument and version it relies on before prescribing a particular remedy. A scientific comparative text should report the origin, legal force, and temporal limits of each source, avoiding the appearance of a uniform worldwide code.
Supervisory Audit Design and Corrective Governance
Establishing an auditable record
A comprehensive audit begins by defining the service and claims being examined: the legal entity operating it, underwriting and intermediary counterparties, insurance class, region, target market, displayed representation, supported devices, funding arrangements, and relevant legal obligations. A precise scope can distinguish primary problems of comparability from adjacent risks in the sales funnel. The audit must be able to reconstruct decisions on a specified version of the service.
The minimum useful record of a quote session includes a pseudonymous session identifier; profile version and permitted test attributes; quote and product IDs; participation status; request and response time; validity timestamp; eligibility status and reason category; contract wording version; price components; coverage fields; declared user constraints; filter result; ranking model or rule version; organic and displayed order; sponsored status; relevant commercial category; and handoff outcome. Additional personal information should be minimized and protected according to the applicable privacy regime. The objective is decision reconstruction rather than unrestricted collection.
Data matching deserves a dedicated control. For example, a carrier's feed might use null for unknown coverage, a second feed might use zero for an excluded benefit, and a third might bundle the benefit into a coverage category. A mapping that treats all three as zero creates a false comparability; a mapping that treats all as included creates misleading results. Contract experts should review data dictionaries, gray cases, amendments, and samples of live wording. A changed coverage definition requires compatibility review, not simply technical confirmation that the feed parses.
A robust validation sample includes ordinary high-frequency risks, boundary conditions, profiles with few available quotes, unusually expensive risks, consumers who require assistance, and combinations of coverage needs. Error and timeout states are included. Test accounts and insurer sandboxes should be authorized and separated from live production applications. Repeated queries that could burden providers or create artificial market signals require permission, rate controls, and a documented sampling protocol.
Replaying ranking and testing claims
A supervisory replay needs the original feasible set, rule version, and material input values. For a stated price-only order, recompute each comparable total, apply eligibility and declared tie-breaking, and compare with stored output. Evaluate sponsorship separately. A test can fail even if the first result is the cheapest when other eligible offers have been improperly hidden. Conversely, an apparent discrepancy may reflect a valid late change to a price or contract. Evidence should include the cause, exact time and version, and reproduction steps.
For a multiattribute ranking, review how the objective relates to the represented service, mandatory constraints, sources of ratings, handling of uncertainty, commercial proxies and personalized inputs. A counterfactual disables specified commercial factors while preserving the same valid quote set and all legitimate features. Compare changes in top positions, displacement and conditional selection rates; if weights are re-normalized, record the alteration explicitly. The counterfactual answers the effect of the chosen perturbation in the recorded architecture, not the net welfare effect of removing all commercial relationships from the market.
The audit should inspect the perception of labels alongside technical truth. Independent consumers are given realistic tasks and asked, before selecting a policy, which offers are advertisements, whether the results cover the market, what the rank means, what changed if they revise a filter, and which costs and limitations remain. A disclosure is assessed in the context of the actual visual order rather than as isolated legal text. Mobile and assistive technology tests should reproduce the decision task. Results should be stratified where feasible and should include misunderstandings rather than simply task completion.
A governance review reconciles the ranking and its commercial contract. It asks which benefits are paid, which contractual terms control participation, how conflicts are recorded, who approves rank changes, whether the operator can override positions manually, what evidence establishes the stated basis, and how discrepancies are escalated. Access to detailed commercial information may be restricted to authorized auditors and supervisors; public explanations can use general parameters and clear sponsorship identifications.
Decision, remediation and verification
A finding should identify the relevant representation or obligation, the versioned evidence, the affected consumer population, expected consequence, cause, and uncertainty. A failed claim of independent order, an omitted mandatory cost, an unknown coverage field shown as included, and a misleading panel claim require different corrective actions. The evidence is stronger when replay and consumer testing converge, although a clear misrepresentation can justify prompt correction without waiting for a population estimate.
Corrective controls follow the mechanism. Product data errors require correction of feed specifications, validation against wording, and review of affected historical quotes. A sponsorship problem may require separating paid elements, removing remuneration from the represented organic criterion, correcting labels, or revising commercial terms. Inaccurate scope claims require precise denominator correction and possibly revision of marketing. Rank-derived unsuitable selections can require stronger eligibility filters and reconsideration of displayed recommendations. Remedies must remain proportionate to applicable law and actual harm.
Consider a hypothetical review of a travel insurance comparator. A test reveals that a particular excluded medical condition is represented as included because an unknown data field is mapped incorrectly. The comparator first stops displaying the erroneous representation, identifies the affected product versions and sessions, informs appropriate partners, corrects the mapping, tests eligible and ineligible profiles, and assesses the need for customer communication or redress under applicable obligations. A general change to the order alone would leave the original defect unresolved.
Corrective verification should compare the same test cases before and after the change; inspect adjacent cases; validate actual user-facing representations; and monitor subsequent production behavior within authorized procedures. Change control should capture who approved the remediation, exact code and data versions, effective time, and residual risk. If a patch resolves one coverage type but removes valid results for another, the correction has introduced a new issue requiring reassessment.
Governance of repeated rank changes should align experimentation with consumer outcomes. A conversion increase is insufficient evidence of customer benefit. Management information should include feasibility, disclosure understanding, price or coverage discrepancies, no-match rates, complaints, and risks from new remuneration terms. Periodic independent sampling of the full quote path can uncover failures absent from aggregate dashboards. Where the platform has a limited role in a distribution chain, arrangements with carriers and intermediaries should make the information necessary for appropriate monitoring available within the lawful boundaries of each party's responsibilities.
Research agenda and evidence thresholds
Several empirical questions remain open across insurance markets: how position affects selection conditional on contractual quality; how disclosed commercial placement changes comprehension and welfare; when broader panels materially improve options for underserved risk groups; and how product design changes in response to the dominance of headline price. These questions require insurance-specific field or simulated experiments, matched quote datasets, and longitudinal observations linked to policy terms and outcomes.
A study of coverage comparability should use a reproducible data dictionary and separately measure expert agreement on categorical mappings. A study of paid ranking should preregister the payment features, treatment, eligible offer universe and estimands. A study of contract outcomes should record changes in wording and the possibility that claim severity distributions differ across insurance products and consumers. A study of market power must define the relevant platform and insurer alternatives and account for direct channels and multi-homing. Each inquiry has a different valid denominator and distinct threats to inference.
The evidence threshold increases with the specificity of the allegation. An observed missing disclosure can establish an interface deficiency in the captured version. A claim that the operator caused consumers to purchase worse policies requires credible linkage between display, decisions and the defined harm. A claim of unlawful supplier discrimination requires the applicable legal standard, relevant comparator and evidence of differential treatment. A market-wide assertion of anticompetitive effects requires appropriate market definition, data, and identification. This hierarchy preserves proportionality between evidence and conclusion.
Conclusion
Insurance comparison platforms create value by reducing the cost of acquiring and organizing offers. Their scientific importance lies in the simultaneous organization of market access, information, and attention. Insurance contracts vary in the risks they transfer, the losses they leave with the customer, applicable conditions, and the quality of service. Ranking by an accurately reported premium can be a useful and transparent operation when the basis and limitations of the comparison are understood. It cannot by itself establish that a policy meets a person's needs.
The analysis establishes three closely connected findings. First, fairness must be examined throughout the quote sequence, beginning with market and conditional panel coverage and ending with the represented order and actual handoff. A technically accurate sort can coexist with exclusion, missing offers, or divergent contractual terms. Second, commercial influence requires evidence about the actual allocation of attention and the platform's stated purpose. Remuneration and ownership create incentives, but commercial relations alone do not prove that a particular ranking was improperly altered. Third, product comparability requires consistent total costs and meaningful contract attributes, with unknowns distinguished from exclusions and consumer preferences distinguished from hard requirements.
The empirical literature demonstrates that search frictions and ranking can influence economic behavior. The magnitude, direction, and welfare implications of effects in insurance require insurance-specific identification and careful measures of risk transfer. Comparative supervisory material supports evaluating consumer understanding, commercial representation and product fit, while the historical competition case demonstrates the necessity of market definition and proof of effects. The proposed audit procedures connect these bodies of knowledge to specific operational evidence without creating new legal duties or claiming a new ranking theory.
A fair comparison service earns confidence through correspondence between its claims, contractual data, and ordering operations. An auditable service makes participation, qualification, commercial placement, coverage, price, and changes in quote status visible at appropriate levels. Governance that verifies those operations and corrects consequential discrepancies makes reduced search friction a meaningful improvement in the consumer's ability to obtain adequate protection. This conclusion is bounded to the mechanisms and evidence examined in this monograph; it is a foundation for subsequent empirical work, not a verdict on unexamined operators.
References
Australian Competition and Consumer Commission (ACCC). 2014. Compare The Market Pays Penalty for Alleged Misleading Health Insurance Advertising. Media release. Official source
Australian Competition and Consumer Commission (ACCC). 2015. A Guide to Comparator Websites for Website Operators and Suppliers. August 3. Official source
Australian Competition and Consumer Commission (ACCC). n.d. Buying Online. Section: Comparison Websites and Apps. Accessed November 19, 2026. Official source
Brown, Jeffrey R., and Austan Goolsbee. 2002. Does the Internet Make Markets More Competitive? Evidence from the Life Insurance Industry. Journal of Political Economy 110(3): 481–507. DOI: 10.1086/339714. Bibliographic record and abstract
Competition and Markets Authority (CMA). 2017. Digital Comparison Tools Market Study: Final Report. September 26. Official source
Competition and Markets Authority (CMA). 2017. Digital Comparison Tools: Summary of Final Report. September. Official source
Competition and Markets Authority (CMA). 2022. Evidence Review of Online Choice Architecture and Consumer and Competition Harm. April. Official source
Competition and Markets Authority (CMA). 2023. Promoting Competition in Services We Rely On: Annual Concurrency Report 2023. Section on Chapter I investigation of price comparison website parity clauses. Official source
Competition Appeal Tribunal. 2022. BGL (Holdings) Limited and Others v Competition and Markets Authority. Case 1380/1/12/21, [2022] CAT 36, judgment August 8. Official case record. Official source
European Commission. 2021. Commission Notice—Guidance on the Interpretation and Application of Directive 2005/29/EC concerning Unfair Business-to-Consumer Commercial Practices in the Internal Market. 2021/C 526/01. OJ C 526, December 29, 2021, pp. 1–129. Official source
European Insurance and Occupational Pensions Authority (EIOPA). 2024. Insurance Comparison Website. Questions and Answers Database, question 3168. Official source
European Parliament and Council. 2016. Directive (EU) 2016/97 of January 20, 2016 on Insurance Distribution (Recast). OJ L 26, February 2, 2016, pp. 19–59. Consolidated text, October 9, 2024. Official source
Farghly, Flo, Max Spohn, and Cherryl Ng. 2021. Discounts, Cashbacks, and Soft Toys: The Impact of Promotions on Consumer Decisions in the General Insurance Markets. FCA Research Note, May 28. Official source
Federal Trade Commission (FTC). 2015. Native Advertising: A Guide for Businesses. December. Official source
Federal Trade Commission (FTC). 2026. FTC Staff Warn Marketers of Healthcare Plans and Lead Generators Not to Deceive Consumers about the Terms of the Advertised Plans. October. Official source
Financial Conduct Authority (FCA). 2014. General Insurance Add-ons Market Study. Final findings, July 24; web page updated April 1, 2016. Official source
Financial Conduct Authority (FCA). 2014. Price Comparison Websites Failing to Meet FCA Expectations. Press release, July 16. Official source
Financial Conduct Authority (FCA). 2014. Price Comparison Websites in the General Insurance Sector. Thematic Review TR14/11. July. Cover updated August 2025 to mark historical status. Official source
Financial Conduct Authority (FCA). 2026. FCA Handbook, PRIN 2A: The Consumer Duty. Provisions consulted November 19, 2026, including dated June 26, 2026 guidance and July 31, 2023 rules. Official source
Honka, Elisabeth. 2014. Quantifying Search and Switching Costs in the US Auto Insurance Industry. RAND Journal of Economics 45(4): 847–884. DOI: 10.1111/1756-2171.12073. Bibliographic record and abstract
International Association of Insurance Supervisors (IAIS). 2024. Insurance Core Principles and Common Framework for the Supervision of Internationally Active Insurance Groups. December 2024 revision; official standard-setting overview. Official source
International Association of Insurance Supervisors (IAIS). 2025. IAIS Publishes Application Paper on Fair Treatment of a Wide Range of Consumers. July 7. Official source
Ursu, Raluca M. 2018. The Power of Rankings: Quantifying the Effect of Rankings on Online Consumer Search and Purchase Decisions. Marketing Science 37(4): 530–552. DOI: 10.1287/mksc.2017.1072. Bibliographic record and abstract
Which? 2025. Risky Business: Consumer Confusion around General Insurance. Policy and insight report, drawing on October 2024 insurance buying survey. Official source