AI SEO Writers: What They Produce and What They Miss

An AI SEO writer takes a keyword, or a short brief, and returns a draft. Most tools produce the same set of outputs:

• An outline built from the headings of pages that already rank.
• Body copy for each heading, usually in a neutral, explanatory register.
• A title tag and meta description sized to typical limits.
• Suggested FAQ questions and, in some tools, schema markup to go with them.
• A keyword or 'content score' that measures how closely the draft matches competing pages.

That is useful work. A good draft saves hours on structure and first-pass wording. What the draft does not contain is anything the model could not already find: your real prices, your own data, your customers' actual questions, or what you learned doing the work. Those are the parts that make a page worth ranking or quoting, and a person has to add them.

If you are choosing between tools, our guide to what makes a good AI SEO writer covers the features that matter in practice.

Google's position is that the tool is not the problem. Its guidance on generative AI content says: "Appropriate use of AI or automation is not against our guidelines."

What Google does act on is scaled content abuse. Its spam policies define this as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users." The first example listed is "Using generative AI tools or other similar tools to generate many pages without adding value for users." The policy applies no matter how the pages were created.

A public experiment shows how this plays out. OtterlyAI published about 1,000 AI-generated posts on each of two new domains in 2026. Otterly reports that Google removed both sites from its index, the first after roughly 14 days and the second after roughly 90, with zero manual actions shown in Search Console. The removal was silent. It is one pair of sites, run by a vendor, with no human-written control, so treat it as an illustration of the policy, not proof of a threshold.

The practical line is simple. One well-edited AI-assisted page that answers a real question is fine. A thousand thin pages on a new domain is the pattern Google describes. Our comparison of scaled AI content vs people-first content walks through where that line sits.

AI answers come from four separate stages. Prose quality only touches the last one.

Access is where the strongest evidence sits. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so content that only appears after JavaScript runs may be invisible to them. A writer can produce a perfect article, and if your site renders it client-side, those crawlers see an empty shell. Our explainer on why AI crawlers download JavaScript they don't run covers the detail.

robots.txt rules are per bot. As of October 2026, OpenAI documents GPTBot as a training crawler and OAI-SearchBot as the crawler for ChatGPT search, and says each setting is independent. A firewall can also block a crawler that robots.txt allows. A stray noindex tag removes a page from Google, and Google requires a page to be indexed and snippet-eligible to appear in AI Overviews or AI Mode. None of this lives in the draft.

Many AI SEO writers now promise copy tuned for ChatGPT or AI Overviews. The evidence does not support that promise.

• Google's guide to generative AI features says: "You don't need to write in a specific way just for generative AI search." It also says there is no ideal page length and that you don't need to chase every long-tail variation.
• C-SEO Bench, a NeurIPS 2025 benchmark, found that most conversational SEO rewriting methods "are not only largely ineffective but also frequently have a negative impact on document ranking."
• Those studies, like nearly all published GEO tests, measured pages already placed inside the model's context. They say nothing about getting a page into the answer in the first place.

One thing does hold up. Stating a specific fact on a page that can be fetched, such as a price, a number or a one-line definition, gives an engine a sentence it can quote. So the most valuable edit you can make to an AI draft is usually adding the concrete fact the model did not know. That is also where AI drafts go wrong most often, which is why consistent brand facts across your site matter more than phrasing.

In your control:

• Whether AI search crawlers are allowed in robots.txt, and whether your firewall or CDN actually lets them through.
• Whether the content that matters is in the HTML before JavaScript runs.
• Whether pages carry noindex, nosnippet or max-snippet directives you did not intend.
• Whether each page answers a real question with specific, accurate facts.
• How many pages you publish, and whether a person reviews each one.

Not in your control:

• What a model learned before its training cutoff.
• The queries an engine writes for itself. OpenAI says ChatGPT search rewrites a prompt into one or more targeted queries, which our breakdown of OpenAI's help page on searching the web explains.
• Whether an engine decides to search at all for a given question.
• Which retrieved page a model chooses to cite.

No tool, writer or audit can promise a citation. What you can do is remove the blocks and give engines accurate material to work with.

Do the access checks before you scale anything. A blocked page wastes every draft you put on it.

1. Read your robots.txt per bot. Confirm OAI-SearchBot, PerplexityBot, Claude-SearchBot and Googlebot are allowed. Remember that GPTBot governs training only.
2. Test the firewall. Request a page as each AI crawler and check for a 200, not a 403 or a challenge page. The AI Crawler Checker does this in one pass.
3. Check rendering. View the raw HTML of a key page with JavaScript off. Your headings, prices and core claims should be there.
4. Check indexing controls. Look for template-level noindex or nosnippet tags. Our accidental noindex audit lists where they hide.
5. Draft with the writer, then edit. Add your real facts, examples and experience. Cut generic filler.
6. Fact-check every number and claim against your own source of truth. Our fact-check workflow for AI SEO writing gives a repeatable checklist.
7. Publish at a pace you can review, then check index status in Google Search Console and Bing Webmaster Tools separately. One being healthy tells you nothing about the other.

Once pages are live, the free checker on this site shows whether AI engines name you for up to three buyer questions. It is a snapshot, so rerun it after changes.

• Publishing at volume to shortcut search. This is the scaled content abuse pattern. In Otterly's test, the removal came without a warning in Search Console.
• Inventing an expert. Otterly's site with a fictional wildlife biologist author box was the one removed first. With one site per arm, that does not prove author boxes hurt, but it is no evidence they help. Google says E-E-A-T isn't a specific ranking factor, so there is no score a persona can raise.
• Trusting the content score. A score that rewards matching competitor headings measures similarity, not value or AI visibility.
• Adding the schema or llms.txt the tool suggests and expecting citations. Google says structured data isn't required for generative AI search, and Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema. Google Search ignores llms.txt. Existing markup does no harm; it just isn't a lever for AI answers.
• Letting the model fill in brand facts. AI drafts confidently state prices, founding dates and features. A wrong number on your own site is a wrong number engines can quote.
• Writing for engines that cannot fetch the page. If the content renders only in the browser, or the firewall blocks the crawler, better prose changes nothing.

Will an AI SEO writer get my brand cited in ChatGPT? Not by itself. ChatGPT search can only use a page OAI-SearchBot is allowed to crawl and your host lets through, and its crawlers don't execute JavaScript. After that, the model decides what to cite. A writer can help you publish clear, accurate answers, which is the part you control. It cannot change access, retrieval or the model's choice, so no tool can promise a citation.

Does Google penalise content written with AI? Not for being AI-written. Google's guidance says appropriate use of AI or automation is not against its guidelines. Its spam policies target scaled content abuse: many pages generated mainly to manipulate rankings without helping users. The test is value to the reader. An edited, accurate AI-assisted page is fine. Hundreds of thin pages published at once are the pattern the policy describes.

How many AI-written articles can I publish before it becomes a problem? Google publishes no number, and nobody outside Google can give you one. The policy is about purpose and value, not a count. Otterly's experiment saw two new domains with about 1,000 AI posts each removed from Google, but that is one test on low-stakes topics. A safer rule is to publish only what a person has reviewed and can stand behind.

Can AI engines still cite a page that Google has dropped? Yes, if they draw on a different index. In Otterly's test, Microsoft Copilot, which grounds on Bing, kept citing the sites after Google removed them, while Google's AI Overviews, AI Mode and Gemini almost never did. Google requires a page to be indexed and snippet-eligible to appear in its AI features. Check each engine's index separately rather than assuming one reflects the others.

Should I run AI drafts through an 'AI detector' before publishing? Google's guidance focuses on whether content helps people, not on how it was produced, so passing a detector is not the goal. Spend that time on accuracy instead. Check every price, date, feature and statistic against your own records, add specifics the model could not know, and remove filler. That editing is what separates a useful page from a scaled one.