Three things move AI search visibility in 2026: letting search crawlers in, rendering content on the server and being indexed in Google, Bing and Brave. llms.txt files, chunked rewrites and schema added for AI do not. Before selling answer engine optimisation (AEO, also called GEO) as a service, we went through the platform documentation and the independent studies and graded each practice by the evidence behind it. This is what survived, with the sources at the end.
Google and Bing both call this SEO
Google's guide to its AI features, published in 2026, says that optimising for them is still SEO. It lists llms.txt files, content chunking, AI-specific rewrites and special schema as unnecessary. A page needs to be indexed and eligible to appear with a snippet. Google lists no further technical requirement.
Microsoft took a different route to the same place. In February 2026 it added generative engine optimisation to the Bing Webmaster Guidelines, describing it as work on whether content is eligible to be used and cited in AI responses, and warning that it does not guarantee outcomes.
So the first finding is unglamorous. The engines that produce AI answers draw on search indexes, and the work that gets you into those indexes is the foundation.
What has strong evidence
Three practices are either documented by the platforms or supported by third-party measurement.
- Let the right crawlers in: OpenAI and Anthropic each run a search crawler separate from their training crawler: OAI-SearchBot and Claude-SearchBot, against GPTBot and ClaudeBot. Perplexity's is PerplexityBot. Google's AI features use Googlebot, and Copilot uses Bing's crawler. OpenAI's documentation says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers. Check your CDN as well, because a bot-blocking default can override what robots.txt allows.
- Render content on the server: A study by Vercel and MERJ in December 2024 found that the major AI crawlers, other than Google's and Apple's, fetch JavaScript files but do not run them. Content that only exists after client-side rendering is invisible to them.
- Be indexed in more than Google: AI Overviews and AI Mode use Google's index and Copilot uses Bing's. Claude's search appears to use Brave. That is inferred, not documented, so treat it as likely. A site missing from Bing or Brave is missing from the assistants built on them.
What has good support
The next group is supported by research, though with less certainty.
- Specific, sourced content: The academic paper that coined the term GEO (Aggarwal and colleagues, KDD 2024) found that adding statistics, quotations and cited sources improved a page's visibility in generated answers by up to 40% in its benchmark.
- Being mentioned elsewhere: Across 75,000 brands, Ahrefs found in 2025 that mentions on other sites correlated with AI Overview visibility at 0.664, against 0.218 for backlinks. Correlation is not cause, but it points at reputation more than at link counts.
- One topic per page, facts first: Bing's guidelines say single-topic pages that lead with the key facts are more likely to be used to ground an answer.
What does not hold up
Structured data has not been shown to get pages cited. A controlled test by Ahrefs in 2026 found no lift, although it only covered pages that were already being cited. Structured data is still worth having for accuracy. It is not a lever.
The llms.txt file and chunked rewrites are both on Google's list of things that are not needed. Two further practices are worse than useless: generating pages for every variant of a query, and hiding instructions for AI models in a page. Google's spam policies treat the first as scaled content abuse, and Bing's guidelines, rewritten in February 2026, list large-scale content generated without editorial review as abuse.
What to expect from traffic
Pew Research Center tracked 68,879 Google searches made by 900 US adults in March 2025. Users clicked a traditional result on 8% of visits when an AI summary appeared, against 15% when it did not, and clicked a link inside the summary about 1% of the time.
Cited brands do better, within limits. Seer Interactive found that brands cited in an AI Overview earned an organic click-through rate of 2.07%, against 0.94% for brands that were not cited. With no AI Overview the rate was 3.35%. The figures cover informational queries in 2025. This shows a difference, not its cause.
The consequence for measurement is that clicks alone now understate visibility. It is worth tracking how often you are mentioned and cited for a fixed set of buyer questions, alongside traffic.
The short version
The practical order of work is short. Check crawler access and your CDN. Render on the server whatever you want cited. Get indexed in Google, Bing and Brave. Then make each important page specific, sourced and about one thing. Measure mentions and citations against a fixed list of questions, and treat any promise beyond that with suspicion.
Sources
- Google Search Central: guide to optimising for AI features
- Search Engine Journal: Bing adds GEO to its official guidelines
- OpenAI: overview of OpenAI crawlers
- Anthropic: what is ClaudeBot, and the other Claude crawlers
- Perplexity: crawler documentation
- Simon Willison: evidence that Claude's search uses Brave
- Vercel and MERJ: The rise of the AI crawler
- Aggarwal et al.: GEO: Generative Engine Optimization
- Ahrefs: brand mentions and AI Overview visibility
- Ahrefs: does schema affect AI citations?
- Pew Research Center: clicks when an AI summary appears
- Seer Interactive: AI Overviews and click-through rate, 2026 update