AI product recommendations need a ranking-integrity test

Approve the recommendation universe and objective

For each recommendation surface, version the eligible catalog, exclusion rules, user need or query, geography, inventory and availability time, approved attributes and claims, comparison dimensions, prohibited inferences, personalization fields, tie-breakers, and the stated objective. Name who may add or remove a product and who approves the objective. Separate finding an eligible option, ordering options, generating a reason, and making a superiority claim. Test that an absent, ineligible, unsafe, incompatible, unavailable, unsupported, or out-of-scope product cannot appear merely because it has strong commercial weight. A fluent answer is unacceptable when reviewers cannot reproduce the candidate set from the declared request and rules. ^[1]

Isolate every commercial influence

Create explicit fields for payment, bid, commission, margin, owned-brand preference, preferred partnership, placement commitment, inventory pressure, promotion, loyalty status, contractual priority, and any other commercial signal. Run counterfactual cases with each influence removed, held constant, increased, and applied to a less relevant product. Preserve the candidate scores, resulting order, reason text, product claims, and change from the approved relevance-only baseline. Require an independent marketing, product, and consumer-review owner to accept the allowed movement and label material commercial influence for the audience experience. If the system cannot show whether compensation or internal economics altered the selection, do not describe the ranking as best, recommended for you, top, most relevant, or otherwise neutral. ^{[2] [1]}

Tie every ranking reason to current evidence

For each material reason, store the approved source, exact product and variant, scope, qualifier, evidence date, owner, expiration, and markets where the statement applies. Challenge the system with missing, conflicting, stale, provider-authored, competitor, review, price, performance, sustainability, safety, compatibility, and customer-outcome evidence. Confirm that lack of proof remains unknown rather than becoming a negative claim about another product or a positive claim about the promoted one. Test whether the explanation changes after the rank is set to justify a commercial result, whether unsupported attributes enter from generated text, and whether two identical products receive different factual descriptions. Reject the ranking when its reason cannot be reproduced from current approved evidence. ^[1]

Independently accept changes to ranking behavior

Version the model, prompt, objective, candidate rules, feature and weight configuration, commercial terms, catalog, evidence set, interface, and approval together. Maintain a fixed acceptance set covering high and low commercial value, equal relevance, conflicting needs, unavailable products, sparse evidence, sensitive personalization, and no suitable option. Compare rank changes with the prior accepted version and require a documented owner decision for unexplained movement or a commercial effect outside the approved band. Monitor concentration, repeated winner patterns, relevance defects, unsupported reasons, complaints, corrections, and differences across protected or sensitive contexts where review is lawful and appropriate. Pause the ranking when reviewers cannot reproduce the eligible set, isolate commercial influence, or defend the stated reason. ^{[1] [2]}

Turn this source into a reviewable decision

For AI for CMOs, use this briefing as a dated decision record rather than a substitute for the source. Preserve U.S. Federal Trade Commission advertising guidance, the exact URL, the October 7, 2026 review date, the supported facts above, the editorial interpretation, the limitations, and any buyer-specific evidence. Link that record to the decisions most directly affected: Commerce and conversion assistance; Customer journeys and personalization; Brand, disclosure, and synthetic-media risk; Measurement and performance explanation. State whether the source changes the scope, evidence requirement, control, sequence, or only the language used to describe the decision.

Before action, name the accountable owner, affected population and workflow, exact offering or configuration, source data and rights, human decision point, exception and appeal path, complete cost, expected benefit, failure and stop conditions, retained evidence, and next review date. Keep official facts, provider statements, buyer observations, representative tests, measured outcomes, editorial inferences, and unknowns visibly separate. Reopen the record when the source, offer, model, integration, data, policy, population, responsible person, or measured result changes.

Limitations and unknowns

The FTC pages are general business guidance reviewed October 7, 2026, not an AI-ranking standard, safe harbor, legal opinion, or conclusion about a specific system. They support bounded truthfulness, evidence, commercial-format, and audience-context principles. They do not validate a ranking objective, candidate set, relevance model, commercial weight, personalization, claim, disclosure, fairness, consumer understanding, legality, or outcome. Verify current applicable law and guidance, actual ranking configuration and relationships, approved claim evidence, counterfactual tests, representative rendered experiences, and qualified marketing, product, data, accessibility, privacy, consumer-protection, regulatory, and legal review.