Early accessMethodology and figures are provisional and may change. The example record shown uses RecommendedByAI's own measured data.
Vol. 2026 · Edition I
Badge & claims policy

Badge & claims policy

How the badge is earned, how it must be shown, and how it is retired.

01 What it's for

What a badge is issued for

A badge is issued for a specific: (brand or model) × category/intent × market × language. The initial methodology may, for example, measure English-language queries for the US market.

A badge is not:

  • a global product-quality seal;
  • a per-SKU certification;
  • an endorsement;
  • a ranking;
  • a claim about markets or languages that were not measured.

Products on a merchant storefront may be mapped to relevant categories, but the badge only applies to the brand/model and category/intent for which it was earned.

02 How it's earned

How a badge is earned

A badge is earned only when the relevant evidence satisfies all applicable C1–C7 criteria.

  • C1 — Valid observations: a sufficient number of valid observations under the applicable methodology version.
  • C2 — Grounded-engine coverage: the brand/model must appear across the required minimum number of grounded AI engines. A grounded engine accesses the web and provides supporting citations or equivalent evidence. Parametric/no-web output does not count toward grounded-engine consensus.
  • C3 — Signal: the aggregate recommendation frequency must exceed the applicable null baseline by a sufficient statistical or methodological margin.
  • C4 — Round stability: the signal must demonstrate sufficient stability within the measurement round, including split-half analysis where required.
  • C5 — Temporal maintenance: a baseline round creates the dated record; subsequent rounds determine whether the record remains supported.
  • C6 — Fresh evidence: the default validity window is 90 days.
  • C7 — Existence and buyability: the product must demonstrably exist and be buyable in the relevant market, supported by appropriate evidence such as a citation resolving to a relevant retail channel.
03 Earned, not bought

Earned, not bought

Payment cannot buy eligibility. A paying customer and a non-paying brand are subject to the same eligibility criteria.

Fees may pay for: measurement; monitoring; historical data; analytics; badge display; reporting; other product features.

Fees do not buy: a positive outcome; a minimum recommendation rate; a badge; favourable treatment in measurement.

04 Meaning of the claim

Meaning of the claim

The standard claim is: “Frequently recommended by AI for <category>.”

The claim means that, during a stated measurement period, the measured AI engines recommended the relevant brand/model at the observed frequency under the applicable methodology.

It does not mean:

  • “AI says this is the best product”;
  • “ChatGPT recommends this product”;
  • “OpenAI recommends this product”;
  • “all AI assistants recommend this product”;
  • “this product is objectively superior”;
  • “this product will sell more”;
  • “this product is #1”.
05 Required information

Required badge information

A badge or claim must preserve:

  • category/intent;
  • observed frequency/denominator;
  • measured engines;
  • measurement date;
  • link to the public evidence page.

The information must not be altered in a way that changes its meaning.

06 Presentation

Approved presentation

The badge should appear in a neutral, calm, theme-aware form. Merchants must not redesign the badge to imply: urgency; scarcity; product-quality certification; an AI provider's endorsement; ranking or superiority.

07 Prohibited uses

Prohibited uses

The following are prohibited:

  • “Best”, “#1”, “top-rated”, “proven” or equivalent quality/superiority claims.
  • Claims that an AI provider endorses or certifies the product.
  • “ChatGPT recommends this product” or equivalent formulations.
  • Changing the observed percentage, denominator, date or engine list.
  • Removing the evidence link.
  • Applying a badge to a different category or product than the one for which it was issued.
  • Presenting a US/English measurement as universal or applicable to another market/language.
  • Continuing to display an expired or revoked badge.
  • Using the badge in advertising or marketing material in a manner that materially changes the factual meaning of the underlying record.
  • Combining the badge with claims that imply RecommendedByAI independently tested or certified product quality.
08 Non-affiliation

Non-affiliation

RecommendedByAI is independent and has no implied affiliation with OpenAI, Google, Perplexity, Anthropic, Shopify or other AI/platform providers. Those names are used only to identify engines or platforms involved in the measurement or integration.

09 Expiry

Expiry and revalidation

Every record carries a measurement date. The default evidence validity period is 90 days. Records are revalidated periodically. A record can cease to qualify if:

  • its signal falls below threshold;
  • the required engine consensus disappears;
  • evidence becomes stale;
  • the product is no longer demonstrably buyable;
  • an underlying measurement error is identified.
10 Auto-hide

Automatic revocation and auto-hide

Revocation is intended to be rule-based and auditable rather than discretionary. When a record becomes invalid: the record is marked revoked/expired; the Shopify badge is designed to auto-hide; the merchant should remove manually placed badges; RecommendedByAI may notify the merchant where practicable. A merchant may not continue displaying a revoked or expired badge.

11 Corrections & disputes

Corrections and disputes

A brand, merchant, competitor, AI provider or member of the public may challenge a record. Challenges should be sent to contact@recommendedbyai.com, identifying the relevant record, the alleged error, and supporting evidence where available.

RecommendedByAI will evaluate the underlying evidence and methodology rather than negotiate the outcome. If an error is confirmed, the record may be corrected, re-run or withdrawn. Material corrections will be logged with the relevant date and evidence.

12 Evidence integrity

Evidence integrity

RecommendedByAI aims for each issued record to be reproducible from its underlying evidence. The company may retain internal evidence, query results, timestamps, methodology versions and audit information necessary to reproduce or defend a measurement.

13 Changes

Changes

This Policy may be updated as the methodology evolves. Material changes will be versioned and reflected in the applicable records.