What Are AEO and GEO? How to Get Your Business Cited by ChatGPT and Google AI
In May 2026, Google published its first official guidance on optimising for AI search. One sentence in it undid about half of what agencies had spent eighteen months selling:
"From Google Search's perspective, optimising for generative AI search is optimising for the search experience, and thus still SEO."
That is the short answer, and if you read nothing else, read this: there is no separate AI channel you can buy your way into. Anyone selling you a GEO package that consists of a new file in your site root is selling you a thing Google has explicitly said it does not read.
What follows is the longer answer. What the terms actually mean, what the evidence says genuinely influences whether an AI answer quotes you, what has now been formally debunked, and what a small business should actually do. We sell SEO, AEO and GEO work, so treat the recommendations with the scepticism they deserve and check the sources at the bottom.
What the three letters actually mean
SEO, search engine optimisation, is getting a page to rank in a list of results a person then clicks.
AEO, answer engine optimisation, is getting your content used as the answer rather than as a link to the answer. The zero-click problem: the person gets what they needed without visiting you.
GEO, generative engine optimisation, is the newer term for being quoted by a system that writes a fresh answer rather than selecting an existing one. ChatGPT, Google's AI Overviews and AI Mode, Perplexity, Claude.
In practice the three overlap heavily, and Google's position is that the overlap is nearly total. The distinction worth keeping is not between the acronyms but between ranking and being quoted, because the evidence says those two are much less related than you would expect.
How an AI answer is actually assembled
Google's guide describes two mechanisms, and understanding them explains most of what follows.
Grounding, or retrieval-augmented generation. The system runs an ordinary search, picks pages from the results, and rewrites the useful parts into an answer with citations. It is not recalling your website from training data. It is fetching it, now, and summarising it.
Query fan-out. For anything complex, the system quietly runs several related searches in parallel and combines what comes back. Somebody asking "how much should I pay for a website" may trigger background searches for ongoing costs, template versus custom, and what a quote should include.
That second mechanism is the one that changes how you write. You are not competing for one query. You are competing for a cluster of related sub-questions, most of which the person never typed. A page that answers one question perfectly and the surrounding six not at all will lose to a page that covers the cluster. This is why our own website cost guide answers twelve separate questions rather than one.
What the evidence says actually influences citation
Four findings, each from a named source, each with the figure attached.
Structure matters more than people assume. A University of Tokyo study, reported by Surfer, found that restructuring a page without changing a single claim in it improved citation performance by an average of 17.3%. Surfer's own analysis found high-citation pages averaged 1,943 words and 10.6 headings, with far more lists than low-citation pages. Not longer for the sake of it. Easier to lift a clean section out of.
Position on the page matters. In Surfer's data, 44.2% of citations came from the first 30% of a page, and only 24.7% from the final third. Answer the question in the opening, then explain. The British habit of building to a conclusion works against you here.
Freshness matters, especially to ChatGPT. AI-cited content averaged 1,064 days old against 1,432 days for Google organic results, roughly 26% fresher. ChatGPT showed the strongest recency bias of the engines measured.
Domain authority appears not to matter at all. This is the finding worth sitting with. Across nine million citations, Surfer found the strongest correlation between off-page authority metrics and citation was 0.02, which is close enough to zero to mean nothing. And 67.82% of AI Overview citations did not rank in Google's top ten for the query or its fan-outs.
Read those two together and the implication for a small business is unusually good news. The thing that locks you out of competitive search results, a decade of domain history you do not have, appears not to be the gatekeeper here. A specific, well-structured, genuinely useful page from an unknown domain can be quoted alongside a national brand.
That is the single strongest argument for a small business publishing content at all in 2026, and it is why we moved our own search and AI work toward answer-shaped content rather than link building.
What Google has now formally debunked
The mythbusting section of Google's guide is blunt. Six things it says you do not need:
llms.txt. "You don't need to create new machine readable files, AI text files, markup, or Markdown." The independent data agrees. An SE Ranking study of 300,000 domains put adoption around 10%, but monitoring of more than 500 million AI bot visits over 90 days found only 408 requests that touched llms.txt at all, and Ahrefs found 97% of llms.txt files received zero traffic in May 2026. As of early 2026 no major AI company had publicly committed to reading it. Google's John Mueller compared it to the keywords meta tag, which is about as dismissive as that organisation gets.
Chunking content for machines. Breaking a page into fragments because an AI "prefers" small pieces. Google says its systems understand multi-topic pages and comprehensive writing serves better.
AI-specific rewrites. A second version of your content written in a flatter voice for machines. Google: "AI systems can understand synonyms and general meanings."
Citation farming. Buying or manufacturing mentions on third-party sites so an AI sees your name repeated. Inauthentic mentions carry no value.
Treating schema as a requirement. "Structured data isn't required for generative AI search." Note the word. Not required. It still helps a machine parse a page quickly and it still drives rich results in ordinary search, which is why we use it. But a schema-only strategy is not an AI strategy.
A page for every query variation. Spinning up pages for each fan-out phrasing. Google calls this a spam policy violation, not a tactic.
If an agency has quoted you for GEO work and the deliverables list is mostly items from that list, you now have the vocabulary to ask what else is in it.
So what should a small business actually do
Six things, in the order I would do them.
1. Fix the ordinary technical foundation first. If a page is slow, blocked, duplicated or unindexed, nothing downstream matters. Grounding runs a real search. If you are not in the results, you are not in the answer. This is unglamorous and it is still where most of the gain is, which is what we found rebuilding Property Safety's site and search foundation.
2. Write answer-first. State the answer in the first two sentences of a section, then explain. Every heading should be a question a person would actually type, and the paragraph underneath should stand alone if it were lifted out of context, because that is precisely what will happen to it.
3. Cover the cluster, not the keyword. For each topic, list the eight to ten related questions somebody would ask around it and answer all of them on one page. That is what query fan-out is reaching for.
4. Publish something only you can say. Original figures, a named process, a specific specification, a result with its own caveats. Generic advice gives a machine nothing to attribute. Our print production knowledge is an example of this: almost nobody else writing about UK branding can describe what 450gsm double-pasted stock actually feels like, because almost nobody else runs the press.
5. Be consistent about who you are. Same business name, same contact details, same description across your site, your Google Business Profile and any directory you appear in. Entity confusion is a real and boring way to lose.
6. Keep it current. Given the freshness bias, a page reviewed and genuinely updated beats one published in 2023 and left alone. Put a review date on the page and honour it.
The honest limits of all this
Three things nobody can sell you, including us.
No agency can guarantee an AI citation. The output is generated per query, varies between users, and changes without notice. Anyone promising a citation is promising something they do not control. We will not, and we say the same about rankings on our pricing page.
The measurement is immature. You can check whether an engine mentions you by asking it, and some tools now track this at scale, but there is no Search Console for AI citation with the reliability of the real thing. Treat any dashboard claiming precise AI visibility numbers with care.
The field moves monthly. Google's guide is dated May 2026. The llms.txt data is from the same period. Something in this article will be out of date within a year, and an article about AI search that does not admit that is not being straight with you.
One disclosure while we are here. This site publishes an llms.txt. We are leaving it up, because it costs nothing and may matter later, but after Google's guidance we would not describe it as a ranking lever and we do not bill anyone for creating one.
The one-line version
Optimising for AI search is mostly just doing SEO properly, writing in a way that can be quoted, and having something worth quoting. The acronyms are new. Most of the work is not.
