Definition
ClaimReview is a Schema.org type for marking up a fact check: it records the claim being checked, who made it, where it appeared, the verdict, and the organization that did the checking. Duke University's Reporters' Lab created it in collaboration with tech platforms, and outlets such as PolitiFact, Full Fact and Maldita have used it for years so that platforms can recognize their work as fact checks.
A ClaimReview object has three main parts. claimReviewed is a short summary of the claim. reviewRating holds the verdict, with alternateName carrying the human-readable rating ('False', 'Mostly true') and optional numeric values. itemReviewed is a Claim object that names the claim's author and where it appeared, through appearance or firstAppearance, plus the date it was made. The fact-checker goes in author, and url points to the full article.
Its status changed in 2025. In June 2025 Google announced it was phasing out ClaimReview rich results in Search as part of a simplification that also removed Course Info, Estimated Salary and other types, and from September 9, 2025 it dropped them from Search Console reports and the Rich Results Test. Google's fact check documentation now says the markup 'remains supported by the Factcheck Explorer Tool'. That tool, Google Fact Check Explorer, lets journalists and researchers search published fact checks, and the Fact Check Tools API offers the same data programmatically plus a read/write API for managing markup. Google's API documentation notes that ClaimReview submitted that way also contributes to Data Commons.
Google's eligibility rules still describe good practice for any fact-check page: the claim must be clearly attributed to an origin separate from your site, the analysis must cite primary sources and explain methods, the publisher needs a corrections policy, the markup must match the visible verdict, and sites of political campaigns, parties or elected officials are not eligible.
For AI search, the markup itself is a weak signal. There is no public documentation that ChatGPT, Gemini or Perplexity read ClaimReview when answering, and tests of live page fetching show assistants mostly ignore JSON-LD. What carries over is the format. A page that states the claim, the verdict and the evidence in plain text near the top is easy for a model to retrieve and quote, which helps on YMYL topics where wrong answers cause harm and where assistants are prone to repeat stale claims, a form of AI hallucination.
Brands sometimes ask whether to use ClaimReview to rebut false claims about themselves. A clear correction page with dated evidence is the useful part, and it supports entity consensus when other sources link to it. The markup is meant for genuine fact-checking of claims made by others, so a company marking up its own marketing as 'fact checks' gains nothing. Promptwatch's guide to fact-checking what AI says about your brand covers how to find and correct those claims.
Examples of ClaimReview
- A national fact-checking outlet keeps adding ClaimReview to every article after the 2025 change because its fact checks still appear in Google Fact Check Explorer, which reporters use to see whether a viral claim has already been checked.
- A university misinformation lab pulls three years of fact checks through the Fact Check Tools API to study which health claims are checked most often in Spanish-language media.
- A health publisher that adopted ClaimReview only for the rich result removes it from opinion columns, which never qualified, and keeps it on its weekly 'Claim check' series that follows Google's eligibility rules.
- An insurance company that keeps seeing a false claim about its claims-denial rate in AI answers publishes a correction page with the regulator's published figures in the first paragraph, rather than marking up its own blog post as a fact check.
