How to tell if a page was written by AI
You generally cannot establish a page's full authorship just by reading its style. Smooth sentences, repeated phrases and factual mistakes can appear in work produced by people, AI systems or a combination of both. They can tell you what needs attention in the text without telling you who created it.
Decide which question you need to answer. If you are checking a writer's process or a contractual requirement, the production history matters. If you are deciding whether the page is fit to publish, its accuracy and usefulness need checking regardless of the tools used. Keep those two assessments separate.
How do you know if text is AI generated?
Start with available evidence of the process: source notes, drafts, revision history and an explanation from the person responsible for the page. Ask what was written, generated, translated or edited with assistance. A page may contain several kinds of contribution, so a single label can leave out important detail.
Treat those records critically. A statement from an author is information to consider, not automatic proof of every step. An absence of visible drafts does not establish AI use either. You may reach a limited conclusion, such as confirming that a particular draft used a tool while the history of an older section remains unknown.
If authorship matters to a decision, keep the original material before editing it. Record what evidence supports your conclusion and what it does not establish. Avoid publicly accusing a person of deception because the prose sounds familiar or a detector returned a high score.
Can an AI detector settle the question?
No detector score should be treated as a complete provenance record on its own. The NIST report on synthetic-content detection discusses limitations including context, text characteristics, mixed authorship and classification errors. A method's performance in one setting does not automatically transfer to every language, document type or editing history.
The opposite claim, that all detection is useless, is also too broad. NIST's 2024 GenAI pilot results report an evaluation using prepared human and generated summaries. Results from that task are evidence about that evaluation, not a verdict on an arbitrary webpage.
If a tool shows a percentage, read what the measure means. It may not represent the proportion of words written by AI. Ask about the supported languages, text length and treatment of edited material, and consider false positives and false negatives. Do not turn an unexplained number into a definitive accusation.
Before submitting a page to an external tool, consider whether the text is public or contains confidential material. A detector is not a reason to disclose a private draft or client information without an appropriate basis.
Which writing problems should you actually look for?
Review the text for defects you can describe without guessing its author:
| Observation | Useful follow-up |
|---|---|
| A claim that the business is the best or largest | Ask what evidence supports the comparison |
| A service described differently across pages | Confirm which description is current |
| A paragraph full of repeated trade terms | Check whether the repetition helps readers |
| Specific numbers without a source | Establish where they came from and whether they are approved |
| A long introduction that avoids the question | Bring the relevant answer forward |
These observations can support an editing decision. They do not distinguish a generated draft from rushed human copy, an outdated template or material assembled from inconsistent notes. Keep the finding precise: "the opening hours conflict" is more defensible than "AI invented the opening hours" when the process is unknown.
Why does my website text sound fake?
Copy can feel unconvincing when praise replaces information. A prospective customer may need to know what a service includes, who it is suitable for and how to contact the business. A paragraph about exceptional quality gives little help if those questions remain unanswered.
Check what information the writer had. Was there a usable brief? Were uncertain facts marked as uncertain? Did somebody familiar with the business review the draft? These questions help you repair the process without assuming that one tool caused every problem.
Google's helpful-content guidance encourages useful, reliable material and asks creators to consider who made it, how it was produced and why. Use those questions to clarify the page's purpose and accountability, rather than treating a disclosure label as a substitute for factual review.
What is keyword stuffing?
Google's spam policies describe keyword stuffing as filling content with keywords or numbers to manipulate rankings. Unnatural repetition is one of its examples. That is a question about the content and its use, not a test for AI authorship.
Read the paragraph for meaning. Some subjects require a term to appear more than once, and some business names naturally include a service or place. There is no simple rule that a word in every sentence proves who wrote the page or establishes a policy violation by itself.
Edit repetition when it makes the explanation harder to follow. Preserve the terms a reader needs, but remove sentences that merely restate the same claim. Do not replace useful detail with vague synonyms solely to lower a repetition count.
Check names, addresses, services and prices carefully
Compare the page with confirmed business information. A plausible address or founding year is still wrong if it does not belong to the business. Ask for the source of a statement before changing it, and keep unresolved facts out of confident public copy.
Equivalent address formats do not necessarily describe different locations. Check the actual place and meaning rather than demanding character-for-character identity in every context. Conversely, similar wording should not make you overlook a different street number or branch.
For Google Business Profile, the business-representation guidelines require the name to reflect the real-world business name and prohibit unnecessary additions. Do not automatically remove a service or place word if it is genuinely part of that name. Assess the actual name against the guidance.
If a price field seems unable to represent the agreed offer, investigate that particular platform's rules. Do not insert a nominal amount and present it as the business's price. A limitation on one form is not evidence that every service requires the same workaround.
Is AI-written content against Google's rules?
AI involvement alone is not a blanket ban. Google's spam policies explicitly discuss using generative AI to create many pages without adding value as an example of scaled content abuse. The relevant assessment includes purpose and usefulness; avoiding AI does not make manipulative or misleading content acceptable.
For Google's AI Overviews and AI Mode specifically, its AI features guidance says no special additional optimisations or AI text files are required. This does not establish the requirements of every other assistant, and it does not promise that Google will cite a page.
Be equally careful with marketing claims about "humanising" a page. A rewrite should improve clarity and accuracy. Introducing errors or invented first-person experience to evade suspicion would make the page less reliable.
What should I do with a page I suspect was generated?
Preserve a copy if you need to investigate its origin. Then make a practical editorial pass: verify material claims, remove unsupported comparisons, resolve contradictions and make the answer useful to the intended reader. Have the responsible person approve facts and commitments before publication.
Document the corrections independently of the authorship question. You can fix a false service description even when the production history remains uncertain. If the process itself breached an agreement, address that using the actual evidence and terms, not a stylistic guess.
For a related assessment of how the business appears in AI answers, see what our AI visibility work looks at.
This article is part of the AI answers guide.