"Answer engine optimization" and "generative engine optimization" have become budget lines, job titles and software categories. Google's position is that they are not separate disciplines. Its guide to generative AI features, updated on July 10, 2026, says that from Google Search's perspective, optimizing for generative AI search is optimizing for search, "and thus still SEO". It points anyone considering third-party AEO or GEO services to its guidance on evaluating SEO advice.
That is a claim by an interested party, and Google's AI features are only one of several answer engines. It is worth testing against the evidence: which parts of SEO carry over to AI answers, which do not, and where the answer depends on the engine.
What Google says, and what it rules out
Google's guidance makes three substantive claims. First, its generative features are rooted in its core ranking and quality systems: they retrieve pages from the Search index to ground answers and run related sub-searches, which Google calls query fan-out. Second, eligibility is the same as for classic results: a page must be indexed and eligible to show a snippet. Third, several popular tactics are unnecessary for Google Search, including llms.txt and other AI-specific files (which Google says it ignores), breaking content into small chunks, rewriting pages in a special style for AI, pursuing inauthentic mentions, and overfocusing on structured data, which it says is not required for these features.
One genuinely new element sits alongside that continuity. Since August 31, 2026, any site can exclude itself from AI Overviews, AI Mode and generative features in Discover through a Search Console setting that, Google says, is not used as a ranking signal elsewhere in Search. Until then, the documented levers, such as nosnippet or noindex, also applied to classic results. Our news report covered the launch.
What carries over
Crawlability and indexing remain the gate. Every major engine ties search visibility to a crawler that honors robots.txt. OpenAI's documentation says sites that block OAI-SearchBot will not appear in ChatGPT search answers, apart from navigational links; Perplexity and Anthropic document equivalent search crawlers. A page that cannot be crawled cannot be cited, and a robots.txt written with only Googlebot in mind no longer describes a site's full exposure.
Rankings still correlate with citations, but more loosely, and not equally everywhere. In July 2025, Ahrefs found that 76% of AI Overview citations ranked in Google's top 10 for the same query; in March 2026, using an improved parsing method, it found about 38%. For chat assistants the link is weaker. In Ahrefs' test of 15,000 prompts, 12% of the URLs cited by ChatGPT, Gemini, Copilot and Perplexity ranked in Google's top 10, with Perplexity the highest at 28.6%. Ranking well helps, particularly on Google and Perplexity, but it no longer determines who gets cited.
Reputation signals persist in a different form. Ahrefs' study of 75,000 brands found brand web mentions correlated with AI Overview visibility at 0.664, against 0.218 for backlinks. That does not show mentions cause citations, and Google says manufactured mentions do not help, but it fits an old SEO principle: authority is conferred by others.
What changes
The unit of competition. Classic SEO competes for positions on a ranked list of pages. AI answers select passages and sources to support a synthesized response. In its October 2025 guidance, Microsoft's Bing team says ranking still happens but is now "less about ordering entire pages". With fan-out, a page can be cited for a sub-question the user never typed.
The weight of coverage on other sites. Across studies, AI engines lean toward third-party sources. A University of Toronto preprint found that engines queried through their APIs drew heavily on earned media for product-ranking prompts, while Google's results included more brand-owned and social pages. Muck Rack attributes 84% of citations in its dataset of more than 25 million links to earned media. The exception matters: for location-based queries, Yext found 86% of 6.8 million citations came from brands' own websites and listings. Which pattern applies depends on what people ask.
The meaning of a position. A ranked list is broadly stable from one search to the next; an AI answer is not. SparkToro found fewer than one in 100 repeated runs of the same prompt returned the same list of brands, and fewer than one in 1,000 returned them in the same order. Its conclusion was that rank position in AI answers is not a meaningful metric, while visibility share across many prompts, each run many times, can be.
The link between visibility and traffic. Being shown and being clicked are further apart than in classic search. Pew's panel data showed users clicked a traditional result on 8% of visits to Google pages with an AI summary, against 15% without one. Seer Interactive found that brands cited in AI Overviews earned 120% more organic clicks per impression than uncited ones, but still 38% fewer than on queries with no AI Overview. A citation is an advantage on a page with fewer clicks to share.
The measurement stack. Search Console now has a generative AI performance report, available to all sites since August 31, 2026, which reports impressions in AI Overviews and AI Mode by page, country, device and date, but not clicks. Bing Webmaster Tools' AI Performance report, launched in February 2026, counts citations in Copilot and Bing's AI summaries and lists the grounding queries behind them, also without clicks. For ChatGPT, Perplexity, Gemini and Claude, site owners depend on referrer data in their own analytics and on third-party trackers, which Google notes have no access to its internal systems.
Where the platforms disagree
On formatting, the two search companies give different advice. Microsoft recommends question-and-answer structures, short standalone answers that can be lifted into a response, and schema markup. Google says content does not need to be chunked and that structured data is not required for its generative features, though it remains useful for rich results. The GEO paper (Aggarwal et al., KDD 2024) adds lab evidence that including statistics, quotations and citations can raise a source's visibility in a generated answer, while keyword stuffing does not. None of this has been tested in a published, controlled study on today's production engines, so formatting advice is best treated as engine-specific and partly unproven.
A sober reading
The evidence supports a narrower claim than either side of the debate usually makes. The foundations of SEO, meaning crawlable and indexable pages and reputation earned from others, still decide whether a site is eligible to be cited. What changes is everything after eligibility: engines choose passages rather than rank pages, weight third-party coverage heavily, produce different answers on each run, and send fewer clicks per appearance.
That argues for keeping SEO fundamentals, measuring AI visibility as a share across many repeated prompts rather than as a rank, and tracking each engine separately. As our review of cross-engine comparisons shows, they rarely cite the same sources.
