Why Is AI-Generated Content Not Ranking or Getting Picked Up by AI Search?
AI-generated content usually fails to rank or be cited because of what it lacks, not because a machine wrote it. Google says it rewards helpful content however it is produced, and treats automation as spam only when it is used mainly to manipulate rankings. The problem is that mass-produced AI text tends to repeat what already ranks. It adds no new facts, no first-hand experience and no reason to prefer it. Ahrefs found 86.5% of top-ranking pages contained some AI-generated text, with no link between the share of AI and ranking position. Yet Graphite found 86% of top-ranking Google articles, and 82% of the sources cited by ChatGPT and Perplexity, were mainly human-written, even though AI now writes about half of new articles. Substance performs; generic output does not.
What do the studies show?
| Study | What it measured | Finding |
|---|---|---|
| Ahrefs, 2025 | 600,000 top-20 ranking pages for 100,000 keywords | 4.6% pure AI, 13.5% pure human, 81.9% mixed; correlation between AI share and ranking about 0.01 |
| Graphite, 2025 | New English-language articles, and articles in Google results and in ChatGPT and Perplexity citations | AI-generated articles grew to about half of new articles, but 86% of top-ranking Google articles and 82% of sources cited by ChatGPT and Perplexity were mainly human-written |
| Google, 2024 | Effect of its March 2024 core update and spam policies | 45% less low-quality, unoriginal content in results |
The studies measure different things with imperfect AI detectors, but they point the same way. Using AI to help write is common among pages that rank. Pages that are mainly AI output, published at scale, are rare among pages that rank or get cited.
Why doesn't AI-generated content get picked up?
- It says what everyone says. Language models produce the most typical answer, and assistants need only one source for a typical answer.
- It is published at scale. Many near-identical pages targeting keyword variations trigger scaled content abuse policies and search systems' quality filters.
- It is not indexed. Search engines often crawl such pages but choose not to index them. Search Console reports them as "Crawled – currently not indexed".
- It contains errors. Invented statistics, outdated facts and plausible but wrong details undermine the page and the site.
- It shows no experience. There are no examples, data, photographs or judgement that only someone who has done the work could supply.
- It answers no real question. Topics chosen for keyword volume rather than buyer needs rarely match what people ask assistants.
- It sits on a site with little authority. New sites with no links or mentions struggle to rank, whatever wrote the content.
Can AI-generated blog posts rank in Google and appear in AI search results?
Yes, when they meet the same bar as any other page. Google's guidance on AI-generated content says appropriate use of AI or automation is not against its guidelines. It also says automation used mainly to manipulate rankings breaches its spam policies. Its guidance on helpful content asks whether content provides original information, reporting, research or analysis, and whether it adds substantial value compared with other pages. Industry coverage of Google's January 2025 update to its search quality rater guidelines reported new guidance on generative AI. Raters were told to give the lowest rating to content created with little effort or originality, whether by automation or not.
An AI-assisted post can rank and be cited if it answers a real question directly and includes facts or examples others lack. It must be accurate, show who stands behind it and sit on a site with some standing in its topic. An AI-generated post that restates the top results will usually be passed over by both search engines and assistants.
Which kinds of AI-assisted content tend to work?
- Pages built from your own data, such as prices, benchmarks or product specifications, turned into readable text.
- Answers to real customer questions taken from support tickets and sales calls, checked by staff who handle them.
- Documentation and help articles drafted from product specifications and verified by the product team.
- Localised versions of strong pages, reviewed by native speakers.
- Summaries of your own research, with the original data linked.
What tends to fail is just as predictable: generic listicles, pages for every keyword variation, rewrites of competitors' articles and definitions of common terms. In each case, the AI has nothing to work from that other writers do not also have.
How do you make AI-assisted content worth citing?
- Start with inputs AI does not have: your data, customer questions, support tickets, project records or an expert interview.
- Have an expert shape the outline and the key points, including opinions and limits.
- Use AI to draft from those inputs, not from a keyword alone.
- Fact-check every claim, number and source against primary documents.
- Add first-hand material: examples, screenshots, photographs, worked calculations.
- Edit for clarity and voice, with direct answers at the start of each section.
- Publish fewer, better pages, and keep them updated.
Keeping human voice in AI-assisted writing is covered in whether optimizing content for AI makes it sound robotic.
How do you diagnose underperforming AI-assisted pages?
- Check indexing in Search Console. Pages crawled but not indexed are being judged as low value.
- Check impressions. Pages that are indexed but get no impressions answer questions nobody searches, or lose to better pages.
- Compare with what is cited. Run the page's target questions in assistants and read the pages that are cited. Note the facts and examples they have that yours lacks.
- Look for duplication across your own site, where several AI-written posts cover the same question.
- Audit accuracy. One wrong statistic can make a page unusable for assistants that cross-check facts.
Why a page can rank in Google and still miss AI answers is covered in why a page can rank in Google but still not appear in AI answers.
What might an illustrative case look like?
Consider an illustrative home improvement retailer that published 600 AI-written how-to articles in three months. After an initial rise, most show "Crawled – currently not indexed", and assistants never cite them. The retailer deletes or merges 450 articles that duplicate each other or common advice. It rewrites the 150 most useful with input from its in-store experts. Each rewrite adds product-specific tips, common mistakes customers report and photographs of real projects, all checked by the expert. Within four months, most of the rewritten articles are indexed. Several are cited by Perplexity and Google's AI Overviews for specific project questions, such as sealing a shower tray.
What do people ask about AI-generated content and search?
Does Google penalise AI-generated content?
Not for being AI-generated. It penalises unhelpful and scaled content made mainly to manipulate rankings, whoever or whatever produced it.
Do AI detectors affect rankings?
There is no evidence that search engines rely on third-party AI detectors. Quality signals matter, not detection scores.
Should we disclose AI use?
Google's guidance suggests being clear about how content was created where readers would reasonably expect it.
Can AI help update old posts?
Yes, for drafting and restructuring. The new facts still need to come from verified sources.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources