AI SEO has two meanings. One is optimizing content to be cited by AI answer engines. The other, covered here, is using AI tools to do AI SEO work: research, briefs, outlines, audits, and drafts. The short answer is that Google allows it, the data shows it can work, and the risk sits almost entirely in publishing unreviewed output at volume.
What Google Actually Permits
Google’s guidance on generative AI content is more permissive than many assume. It says generative AI can be particularly useful when researching a topic and adding structure to original content. It does not ban AI-assisted writing.
The line is drawn at scale without value. Google’s scaled content abuse policy, introduced on March 5, 2024, targets pages created mainly to manipulate rankings rather than help people. It applies equally to AI-written, human-written, and template-generated pages, and Google has never published a page-count threshold. Its examples include using generative AI to produce many pages without adding value, scraping and rewording other sites, and stitching together content from different pages.
Google also asks for two habits. Fact-check and review all AI-generated content before publishing, including titles, meta descriptions, structured data, and image alt text. And consider telling readers how the content was created.
What One Million Ranking Pages Show
In a study published on July 27, 2026, Ahrefs examined about one million pages from the top 10 results of 100,000 searches run in June 2026. Only 5.3% of top-three pages were detected as fully AI-generated, and 82.2% were under 50% AI.
The gradient is gentle, not a cliff. Pages that were at least 80% AI made up 8.4% of position one results and 11.7% of position ten. Low- and moderate-AI pages earned two to three times the impressions of high- and very-high-AI pages, and heavily AI pages were indexed less often (40.35% versus 49.28%).
Read this carefully. The study is correlational, uses Ahrefs’ own detector (which the company sells), and only samples pages that already ranked. Its authors conclude that Google punishes weak content, not AI use itself. That is a reasonable reading, though not proof of cause.
The Job AI Cannot Do: Measure
A language model predicts text. It does not query a live keyword database. Ahrefs makes this point bluntly, noting that search volume, keyword difficulty, and similar figures produced by a chatbot are estimates, and that training data has a cutoff. Ahrefs sells SEO tools, so weigh its motive, but the technical point is sound.
The fix is to connect AI to measured data. Search Console shows your real queries and clicks, and some SEO platforms now let assistants call their data directly through APIs. Use AI to interpret numbers that came from a measurement source, never to invent them.
A Review Gate Before Anything Publishes
Sarkar SEO recommends a fixed checkpoint, because speed is exactly where quality leaks. Before publishing any AI-assisted page:
- Verify every number and name against the original source, not another summary.
- Add what only you know. A client question, a real process, your own data, or a tested result. If the page contains nothing a competitor’s AI could not also produce, it is not ready.
- Check the metadata and markup that searchers actually see.
- Ask whether the page deserves to exist. Would you publish it if AI had never been involved, and does it answer a need no existing page on your site already covers?
- Decide how to disclose. Where readers would care, state how automation was used.
A Worked Example: The 200-Page Trap
Imagine a service business that prompts AI to write 200 city pages, swapping only the city name and a few generic sentences. Each page reads smoothly. Together they match Google’s description of templated pages produced at volume without added value, and they may never earn indexation or impressions.
A safer version publishes 15 pages, each containing something local and real: team members who serve that area, genuine project photos, local regulations, or authentic customer questions. AI can still draft the structure. The substance comes from the business. This is an illustration, not a case study, but it shows where the trade-off lies.
Pace and Governance
Teams also stumble on process, not tools. Name one person accountable for each published page, keep a record of which tool drafted what, and cap publishing speed to what your reviewers can genuinely check. A rule that no AI-assisted page ships without a named human reviewer is cheap to adopt and expensive to ignore.
