AEO, GEO & LLM SEO: How AI Search Changes the Work

SEO
Written By wevisualise designs

Three acronyms arrived inside two years, each sold as a new discipline with its own retainer. Most of what they describe is search work you should already be doing. A small part is genuinely different, and one part of it breaks your reporting. Here is the line between them, drawn using published guidance and named research rather than conjecture.

Three acronyms, one underlying question

Answer engine optimisation, AEO, is the oldest of the three and the loosest. It predates generative AI and originally meant optimising for featured snippets, People Also Ask boxes and voice assistants, anywhere a result was read out or lifted rather than clicked.

Generative engine optimisation, GEO, has a traceable origin. It comes from an academic paper of that name by Pranjal Aggarwal, Vishvak Murahari and co-authors, first submitted on 16 November 2023, revised on 28 June 2024 and published at KDD 2024. The paper proposes a framework for optimising content visibility inside generative engine responses, and reports that its methods "can boost visibility by up to 40% in generative engine responses" on a benchmark the authors built themselves, called GEO-bench. Treat that number carefully. It was measured against the generative engines of that period, on the researchers' own benchmark, and the authors themselves note that effectiveness varies substantially by domain.

LLM SEO is the informal umbrella term for the whole thing: getting your content used and cited by large language model products, whether that is Google's AI Overviews and AI Mode, ChatGPT, Copilot or Claude.

The three terms overlap almost entirely. The useful question is not which acronym to buy. It is which parts of the work are actually different.

What Google says in its own documentation

Google published a dedicated guide to optimising for generative AI features on Search, and it is unusually direct. Its central claim is that "the best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems."

Its guidance on AI features and your website goes further: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

The one hard eligibility rule is worth memorising, because it is the only technical gate Google states plainly. To appear in generative AI features on Google Search, "a page must be indexed and eligible to be shown in Google Search with a snippet." Indexed, and snippet eligible. Everything else in the guide is about content quality.

On quality, Google's phrasing repeats its long-standing position: "Creating content that people find unique, compelling, and useful will likely influence your website's presence in generative AI search in the long run more than any of the other suggestions in this guide." It singles out what it calls commodity content, giving the example of "7 Tips for First-Time Homebuyers", as material based on common knowledge that adds nothing a model does not already hold.

Be appropriately sceptical here. Google is describing its own systems and has an interest in discouraging manipulation. But this is documentation it can be held to, which is more than most AI search advice can claim.

How AI answers source & cite content

The mechanism worth understanding is query fan-out. Google states that both AI Overviews and AI Mode may issue multiple related searches to build a single response, then display "a wider and more diverse set of helpful links" alongside it.

That has one real consequence for how you plan content. You are no longer competing only for the phrase the user typed. You are competing across a set of related sub-questions the system generated on their behalf, any one of which might pull from a different page on a different site. A page that answers a headline question but skips the obvious follow-ups covers one branch of the fan-out and misses the rest.

This is the closest thing to a genuine tactical shift in the whole subject, and it is still recognisably content strategy. Cover the question and its sensible follow-ups on the same page, in plain language, with specifics a model cannot generate from common knowledge.

Structured data & the machine readable file question

A large amount of AEO and GEO advice concerns markup. Google's guidance contradicts most of it.

On schema: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." On the various proposed AI-specific files: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search." On the popular advice to atomise your pages: "There's no requirement to break your content into tiny pieces for AI to better understand it." And on writing style: "You don't need to write in a specific way just for generative AI search."

None of that makes structured data pointless. Google's own introduction to structured data explains that it helps Google understand the content of a page and can make that page eligible for rich results in ordinary search. Eligible, not guaranteed. Mark up your products, events, articles, reviews and organisation details because they earn richer conventional results. Do not pay anyone to add schema on the promise that it gets you into AI answers.

The honest position on llms.txt and similar proposals is that no major AI platform has published documentation confirming it uses them, and Google has published documentation saying it does not need them. That may change. Today the evidence is thin, and anyone charging for it is charging for a hypothesis.

Crawler access & the robots question

This is where the work genuinely is new, because it is a commercial decision rather than a technical one, and the controls are separate.

Google splits training from search. Google-Extended is described as "a standalone product token that web publishers can use to manage whether content Google crawls from their sites may be used for training future generations of Gemini models" and for grounding in AI applications. Critically, Google states that "Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search." You can refuse Gemini training without losing search visibility.

Snippet controls are a different matter and catch people out. Google's robots meta tag documentation confirms that nosnippet "will also prevent the content from being used as a direct input for AI Overviews and AI Mode", and that max-snippet:0 behaves the same way. If someone on your team applied nosnippet years ago to stop competitors scraping copy, you have opted out of AI answers without meaning to.

OpenAI publishes the same split across three agents. OAI-SearchBot powers ChatGPT search, and OpenAI states that "sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." GPTBot collects content that may be used for model training. ChatGPT-User acts on a live user request, and OpenAI notes that robots.txt rules may not apply to it in the same way.

The practical rule: blocking training crawlers is a defensible business choice with no search cost. Blocking search crawlers costs you visibility in that product. Decide those separately, and audit what your robots.txt already says before you decide anything.

Where measurement breaks

The uncomfortable part. Visibility can rise while clicks fall, and most reporting cannot tell the difference.

The Pew Research Center published the clearest evidence on 22 July 2025, based on the browsing data of 900 US adults across 68,879 Google searches collected in March 2025. Some 18 per cent of those searches produced an AI summary. On pages with an AI summary, users clicked a traditional search result on 8 per cent of visits, against 15 per cent where no summary appeared. Clicks on a source link inside the summary happened on just 1 per cent of visits. Sessions ended at the search results on 26 per cent of pages with a summary, against 16 per cent without.

Read the limits as well as the numbers. That is US adults, in one month, before AI Mode was widely deployed. It is not a UK B2B sample. It is still the best-documented account of the direction of travel.

On the crawling side, Cloudflare published crawl-to-refer ratios on 01 July 2025, comparing how much AI platforms crawl against how much traffic they send back. For the week of 19 to 26 June 2025 it reported roughly 70,900 crawl requests per referral for Anthropic. Cloudflare's own caveat matters as much as the figure: traffic referred by Claude's native app carries no referrer header, so the published ratios "may overstate the respective ratios, but it is unclear by how much." Anyone quoting that number without the caveat is quoting it badly.

Google's answer to the reporting gap is a generative AI performance report inside Search Console, which its optimisation guide directs site owners to use. The same guide is blunt about the alternatives: "No third-party tool has access to our internal ranking or AI systems."

What is new & what is rebadged

Claim you will hear What the published evidence says Verdict
AI search needs its own schema markup Google states no special schema.org markup is needed Rebadged
You need an llms.txt or similar file Google states no new machine readable files are needed Unsupported
Break content into small chunks for AI Google states there is no such requirement Rebadged
Write differently for language models Google states you do not need to write in a specific way Rebadged
Citation depends on conventional strength Features are rooted in core ranking; pages must be indexed and snippet eligible Same work
Sub-question coverage matters more Query fan-out issues multiple related searches per response Genuinely different
Crawler permissions are a commercial decision Google and OpenAI both split training from search access New
Clicks fall while impressions hold Pew, March 2025 data: 8 per cent against 15 per cent New measurement problem

Six of eight rows are old work with a new invoice attached. Two are real. Price your effort accordingly.

What to do next

  1. Search your codebase and CMS settings for nosnippet and max-snippet this week. If either is set on commercial pages, you have opted out of AI answers.
  2. Read your robots.txt line by line and decide, deliberately, what you allow for Googlebot, Google-Extended, OAI-SearchBot and GPTBot. Write the reasoning down so the next person does not reverse it by accident.
  3. Open the generative AI performance report in Search Console and set a baseline now, before you change anything, so you can separate impression movement from click movement later.
  4. Take your three most commercially important pages and list the follow-up questions a buyer would ask next. Answer them on the same page.
  5. Ask any supplier proposing an AEO or GEO retainer which specific published source supports each tactic. Cut every line item that cannot name one.

Most of this is a morning's audit followed by the content work you already knew you needed. Where it stops being a morning's audit is when indexing, snippet eligibility and structure need untangling across a large site, and that is where our SEO work starts.

Sources

wevisualise designs https://wevisualise.co.uk
Next

SEO for Small Businesses: What to Do First