Obility Editorial · · 4 min read
Search engines judge the scale and the value, whoever wrote the page
Many teams now ask whether they can use AI to write the pages they hope ChatGPT or AI Overviews will cite, sometimes hundreds at a time. The two search engines behind the largest AI answer surfaces have answered in writing. Their answer turns on volume and value. Who or what typed the words is not the test.
Google’s guidance on generative AI content, last updated in October 2026, says AI can be useful for researching a topic and adding structure to original content, and that generating many pages without adding value for users may violate its spam policy on scaled content abuse. That policy, introduced with the March 2024 core update, describes many pages generated mainly to manipulate rankings rather than help users, “no matter how it’s created”. Google’s announcement said it covers automation, human effort or a mix of both. A team of freelancers rewriting the same article 300 times is in the same position as a model doing it.
Microsoft’s Bing Webmaster Guidelines draw the same line. Under a heading on automatically generated content at scale, they say large-scale content generated without oversight, quality control or editorial review often lacks usefulness, accuracy and originality, and may be excluded from indexing. In both documents the question is whether a person with judgement stood behind the page.
The filter sits upstream of the AI answer
This matters for AI visibility because an answer is assembled from what the index keeps. Google’s documentation on AI features says a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode, and tells site owners to follow the same Search policies. Its spam policies define spam to include attempts to manipulate generative AI responses in Google Search, and say violating sites may rank lower or not appear at all.
Bing’s guidelines are written for Copilot and for the results Bing supplies through its grounding API, as well as for classic search. They say content should be created for users, “not to manipulate ranking systems or trigger citations or AI responses”, and that failing the guidelines can reduce eligibility for grounding or remove a site from the index.
So the serious risk for a batch of generated pages is that it never reaches the pool an answer draws from, or that the site’s standing in that pool drops. Going quietly uncited is the mild outcome. Google reported in April 2024 that the March changes left 45 percent less low-quality, unoriginal content in its results, against the 40 percent it had expected. That is Google grading its own filter, but it shows the filter is in active use.
A page that restates the consensus gives an answer nothing to cite
Suppose the generated pages survive. They still face a second problem, which follows from how answers are built. An engine that searches before answering pulls several sources for a question and writes one response. If your page says what the other nine say, any of them can support the sentence, and nothing makes yours the one the answer needs. Google’s own self-assessment questions ask whether content offers original information, reporting, research or analysis, and whether it provides substantial value compared with other pages in the results. Its warning signs include extensive automation to produce content on many topics, and mainly summarising what others have said.
Take an illustrative payroll software vendor weighing two plans. The first generates 200 glossary pages, from “what is gross pay” to “what is a pay stub”, each a fluent restatement of definitions every competitor has already published. The second publishes 15 pages on questions only the vendor can answer: which countries it can pay contractors in, how a mid-year move from another provider works, what happens to payroll data when a contract ends. The second plan is far smaller, and it is the only one that contains facts an answer about this vendor would need.
Use AI for the work around the facts, and check every claim
None of this argues against AI in the writing process. Google names research and structure as good uses. The risk it names is accuracy. Its guidance reminds publishers that generative models do not retrieve facts but predict a likely sequence of words, and asks that all AI-generated content be checked for accuracy before publishing, including titles, meta descriptions, structured data and image alt text, since all of them can appear in search results.
A workable division of labour gives people the facts, the evidence and the decision about which questions deserve a page. AI helps outline, tighten and adapt the draft. A named reviewer then checks every claim against a source of truth before anything ships. Obility’s Content Agents workflow is built this way. It starts from a brief that names the audience, the question and the supporting evidence, checks whether an existing page should answer the question before another URL is created, and keeps a person reviewing factual claims, brand language and sources before publication. Context Manager records the product names, descriptions and approved claims that the review checks against.
Google also suggests telling readers how a piece was made where they would reasonably wonder, for example by explaining how automation was used. It presents this as context for readers, and the decision about where it is expected is left to the publisher.
No one publishes the threshold, so judge each page on its own
Be clear about what these documents leave out. Neither Google nor Bing says how many pages counts as scale, and neither describes detecting AI text as the test; both describe value and oversight. They also cover only Google’s AI features and Bing’s Copilot and grounding results. OpenAI’s crawler documentation, for example, describes its user agents and robots.txt controls and sets out no comparable content rules, so how ChatGPT treats a generated page is not documented beyond what its search sources do.
The working rule is a question asked of each page. Would it still be worth publishing if no search engine or assistant existed, and does it say something about your product that no other page says? If the honest answer is no, producing it faster does not help. Then measure where the effect shows: run a fixed set of buyer prompts before and after the new pages go live, and check whether they appear among the cited sources.