Factual verification begins with an exact claim and the evidence needed to support it. Authorship detection is a different task. Its output does not establish that a number, quotation, policy statement, or named capability is true.
Google says appropriate use of AI is not itself against its guidelines. It also says automation used primarily to manipulate rankings violates spam policies (Google Search Central, retrieved 2026-09-08). That is a policy point to verify, not a shortcut for deciding whether every sentence in an AI-assisted draft is accurate.
Key takeaways
- Extract checkable claims before rewriting them.
- Keep the original wording, evidence, verdict, and final treatment together.
- Compare source scope, date, conditions, definitions, and units before accepting a claim.
- Recheck a substantive rewrite because it can create a new claim.
Build a claim register before you edit
Copy every checkable assertion into a register. Include numbers, dates, quotations, policy summaries, named products, and comparative claims. Keep the original wording. Editing before checking can obscure what the draft actually asserted.
The six verdicts below are an editorial method for this workflow. They are not a universal scoring standard.
- Supported: the evidence supports the original claim at its stated scope and date.
- Partially supported: the evidence supports only part of the claim, so narrow the final wording.
- Inferred: the sentence is a clearly labeled conclusion drawn from evidence, not a source statement.
- Contradicted: credible evidence conflicts with the original wording.
- Unverified: no adequate accessible evidence supports the claim.
- Inaccessible: the named evidence could not be read or checked.
| Field | Record | Why it matters |
|---|---|---|
| Original claim | The sentence exactly as drafted. | Prevents a rewrite from hiding the assertion under review. |
| Evidence | The strongest accessible source and relevant passage. | Lets another reviewer retrace the basis. |
| Verdict | One of the six defined statuses. | Makes the next action explicit. |
| Final treatment | Keep, narrow, label as inference, remove, or escalate. | Binds the evidence to the published wording. |
Download the blank claim-register CSV. Use it as the work surface, rather than trying to copy a wide template from the article.
Extract claims, then find the strongest evidence
Work sentence by sentence through statements a reader could challenge. A source from the organization responsible for a product, rule, or dataset is usually a better starting point than a roundup repeating it. Record a retrieval date for changeable guidance. If a source is inaccessible, record that status and why. Do not silently replace it with a search snippet.
When two credible sources conflict, record both. Check whether they cover different product versions, countries, time periods, or definitions. Then narrow the claim to what each actually supports. Escalate the conflict if publication depends on it.
Compare scope, date, conditions, and units
Many bad rewrites begin with a real source used at the wrong scope. Before accepting a number, retain its unit, time period, population, and method. For a quotation, compare the exact words, speaker, date, and surrounding context. For a product claim, check the current official documentation, plan, version, and eligibility conditions. These details decide whether the original sentence can stay.
Worked claims
These examples use public Google guidance. They are editorial exercises, not a review of a live article. The source was checked on 2026-09-08.
C-01: “Google penalizes all AI-written content.”
Verdict: contradicted. Google's guidance distinguishes appropriate AI use from automation used primarily to manipulate rankings. A safe rewrite is: “AI use alone does not determine compliance. Google says automation used primarily to manipulate rankings violates its spam policies.”
C-02: “This post is comprehensive because it has 2,000 words.”
Verdict: unverified as a universal rule. Google's people-first guidance asks whether content is useful and complete, but it does not make a word count proof of comprehensiveness. That does not decide whether a particular post is comprehensive. The editorial rewrite is: “Assess whether the post gives the reader a complete, useful answer. Do not use word count alone as proof.”
C-03: “Google allows AI-written content.”
Verdict: partially supported. The policy statement needs its qualification about the use and purpose of automation. A safe rewrite is: “Google says appropriate AI use is not itself against its guidelines. Automation used primarily to manipulate rankings violates its spam policies.”
Rewrite, then review the rewrite
Replace only the wording you can support. Do not turn a source about a narrow case into a broad promise. Mark a conclusion as an inference when it goes beyond the source. Once the edit is made, check the register again. A new headline, meta description, example, or transition may create a fresh factual claim.
Google's people-first guidance asks whether content includes original information, clear sourcing, and a substantial, complete description of its topic. That supports a reviewable editorial process. It does not guarantee rankings or prove that one fact-check makes a page good.
A concise prompt for a fact-check tool
Extract every checkable claim from this draft. For each claim, preserve the exact wording, classify it, cite the strongest accessible source with retrieval date, compare scope, date, conditions, definitions, and units, assign a verdict, and propose a safe final treatment. Do not invent evidence. Flag conflicting or inaccessible sources for review.For a deeper method, see the claim-by-claim Blog Factcheck skill. For a separate discussion of detection and scoring, see the AI content scoring guide.