Five names get sold for the same idea: generative engine optimization, answer engine optimization, AI SEO, AI search optimization, LLM SEO. Together they draw about 19,000 US searches a month (Google Keyword Planner, August 2025 to July 2026 — the per-term breakdown is further down), and the pages ranking for them mostly recommend the same three moves — add schema, publish an llms.txt, restructure your content for machines.
Google's own documentation says you don't need any of the three.
That's not a gotcha, and it isn't a reason to ignore AI search. It's the reason this page exists. What follows is the evidence rather than the pitch: what Google actually published about how its AI picks sources, what a 68,879-search study found about clicks, and what happened when we ran a citation test on a client and wrote down the date.
GEO is not a separate track — Google says it runs on the old one
The most common claim in this market is that AI search is a new game requiring new skills. There's a paid ad on the search results page for "generative engine optimization" right now whose pitch is that GEO is "its own game, not SEO with a new name."
Google's AI features documentation says the opposite, in its own words:
"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."
The same page describes the mechanism: Google's AI relies on its core Search ranking systems to retrieve relevant, up-to-date web pages, then generates an answer from them. That's retrieval first, generation second.
The practical consequence is blunt. If classic search can't find and retrieve your page, there is nothing for the AI layer to quote. A page that ranks 90th is not "waiting for its GEO moment" — it's a page the retrieval step never reaches. Fixing that is ordinary SEO: getting indexed, being crawlable, being relevant to a query someone actually types.
⚠️ One boundary worth stating. Everything above is Google describing Google. ChatGPT, Perplexity and Claude run their own retrieval stacks, and Google's documentation cannot speak for them. What we can say is that in our own test, the pages an engine cited were pages that already existed and were already indexed — not files written for machines.
Three things sold as GEO that Google says you don't need
These are the three most common items on a paid "GEO audit" checklist, next to what Google's documentation actually says about each.
| Sold as essential | What Google's own documentation says |
|---|---|
| Add schema markup so AI can parse you | "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." |
Publish an llms.txt so AI engines know your site |
You don't need to create new machine-readable files or AI text files to appear in Google Search including its generative features — "Google Search itself doesn't use them." |
| Restructure content into AI-readable formats (Markdown, AI-specific pages) | Same sentence. Markdown and AI-specific markup are named directly, and Google says it doesn't use them. |
Two things need to be said carefully here, because the easy reading of that table is wrong in both directions.
Schema is not useless. Google's sentence is narrow: structured data isn't required for generative AI search. Schema still earns traditional rich results — review stars, FAQ panels, opening hours in the search listing — and those are worth having for reasons that have nothing to do with AI. What the documentation doesn't support is selling schema as the way to get quoted by AI.
And llms.txt is an open question, not a closed one. Google's statement settles Google Search and nothing else. Whether Perplexity, ChatGPT or Claude read such a file is a question only their operators can answer, and none of them has published the equivalent guidance. The practical read is narrow but useful: it isn't worth paying an agency to create one as a ranking tactic, and it isn't worth tearing one out if it already exists. A file costs nothing to keep. An invoice for it costs something.
What being quoted actually gets you
This is the number the GEO market doesn't put in its slide decks.
Pew Research Center observed the real browsing behaviour of 900 US adults across 68,879 Google searches in March 2025. Not a survey — actual recorded visits.
| What they measured | Share of visits |
|---|---|
| Searches that showed an AI summary at all | 18% |
| Clicked a traditional search result, with an AI summary present | 8% |
| Clicked a traditional search result, without one | 15% |
| Clicked a source link inside the AI summary | 1% |
Read the last row again. Getting cited inside an AI answer converts to a click roughly one percent of the time. Being the quoted source is not the same as being visited, and it is nowhere close to being contacted.
⚠️ Don't over-read the 8% versus 15%. This is observational data, not a controlled experiment. Searches that trigger an AI summary may be different searches — more informational, the kind that always had lower click rates. Pew didn't split the data that way, so this can't be stated as "AI summaries caused the drop in clicks." The 1% figure stands on its own regardless, because it's measuring clicks on the summary itself.
We ran the test once, and wrote down the date
Theory is cheap in this market, so here's a measured result from a client account rather than an argument.
The client manufactures and brands a consumer health product in Malaysia and exports to Europe. We asked Perplexity, in the voice of a European distributor looking for suppliers, twice, in two different phrasings. It named the client first both times, and cited pages from their own site.
The part that makes it evidence rather than an anecdote is what happened fifteen days earlier. The same question, the same site, returned four competitors and no mention of the client at all. Their robots.txt was fully open and their llms.txt was in place on both dates. Nothing technical changed in between.
What did change: two guide pages went live — pages answering what a buyer asks before choosing a supplier. Those were the pages cited. Not the homepage, not the product catalogue, not the marketing copy.
There's a second detail worth keeping. On the earlier date, the competitors the engine named were being cited through a government exporter directory, a LinkedIn company page, and B2B marketplaces — third-party sources, not their own websites. Being findable in the places your industry already lists suppliers is part of the same job.
⚠️ Keep that result attached to the number from the previous section. Being named first is not the same as being visited: across Pew's 68,879 searches, source links inside AI summaries drew clicks on about 1% of visits. The mechanism is real and independently verifiable — you can run this test yourself this afternoon. The traffic that follows it, at this stage of the market, is small. Both of those are worth knowing before anyone quotes you a forecast built on AI visibility.
The full write-up, with the queries and what came back, is in what happened when we asked an AI engine for suppliers.
What Google actually suggests you do
Across that whole documentation page, there's exactly one thing Google actively recommends rather than tells you to skip:
"Our AI systems take a look at a variety of sources, so it can be helpful to have a unique viewpoint that stands out."
That's the entire positive instruction. It's also, awkwardly for an industry selling checklists, the one item that can't be turned into a deliverable — you cannot buy a unique viewpoint in a package.
What it looks like in practice, based on what actually got cited in our test:
1. Answer one real question per page, and answer it in the first line. Not "our approach to quality" — "how do I check whether a supplier is certified?" The first sentence should be the answer, not the wind-up.
2. Put specifics where a vague sentence would go. Names, numbers, dates, what happened. "We deliver results" is unquotable by construction. "Named first in two phrasings, fifteen days after two guide pages went live" is quotable because there's something in it to quote.
3. Write the thing a template can't produce. Most pages on any given topic are assembled from the same three sources and say the same optimistic thing. What an AI system can't find in forty other places is a first-hand observation with a date on it, a number you measured yourself, or a limit you're willing to state out loud. That's the practical translation of "unique viewpoint."
4. Be listed where your industry already gets listed. Directories, association registries, marketplaces, LinkedIn. In our earlier test that's how the competitors were being cited. This is unglamorous and it works.
5. Test it yourself, on a schedule, and write down the date. Ask the engines the question your customer would ask. Are you named? Is a competitor? Our result only means something because we had a dated observation fifteen days earlier to compare against. One check is an anecdote. Two dated checks are evidence. ChatGPT is where most people test first, and what it takes to be named there is a longer answer than one checklist line.
Notice what's not on this list: no schema requirement, no llms.txt, no AI-specific file format. Those are on the vendor checklists, not on Google's.
The five names, and which ones are growing
If you've been comparing proposals, you've seen these terms used as if they were different services. They aren't. Here are the US search volumes for August 2025 through July 2026, from Google's own Keyword Planner:
| Term | Monthly searches | Year over year |
|---|---|---|
ai seo |
8,100 | −33% |
generative engine optimization |
4,400 | −46% |
geo seo |
2,900 | +26% |
answer engine optimization (AEO) |
2,400 | +26% |
ai search optimization |
1,300 | +60% |
llm seo |
880 | −52% |
how to rank in ai overviews |
260 | +21% |
What each one means, in practice:
- GEO — generative engine optimization. The original term, from an academic paper on optimizing content for LLM-augmented search engines. Broadest of the five.
- AEO — answer engine optimization. Same work, framed around being the answer rather than a ranking position. Often used for voice assistants and featured snippets too. It's the fastest-growing of the five names — we broke down what AEO services actually cover, and what to check before paying anyone separately.
- AI SEO. The loosest label. Sometimes means this work, sometimes means using AI tools to do SEO — two completely different things sold under one name. Ask which one is meant.
- AI search optimization. Newest framing, fastest growth, and the highest advertiser bids of the group.
- LLM SEO. Same idea, model-centric framing. Declining fastest of the five.
The trend is splitting, and that's the useful part. The older labels — ai seo, generative engine optimization, llm seo — are shrinking year over year. The newer ones — AEO, ai search optimization, ai overview optimization — are growing. Nobody knows which name wins. That's precisely why this page covers all five rather than betting on one, and why you shouldn't pay a premium for a service whose main innovation is a fresher acronym.
One more number worth noticing: advertisers bid MYR 83 to 198 for clicks on these terms. For comparison, google business profile tops out around MYR 3.44. People searching this topic are being treated as buyers with budgets — which is also why the market around it is loud.
The takeaway
Search moved, but not as far as the pitch decks claim. Google says its AI features are rooted in its existing ranking systems, and its AI retrieves pages through them. Being quotable starts with being retrievable, which is ordinary SEO with an unglamorous name.
The mechanism is real, and you can verify it yourself in an afternoon — we did, and the engine's answer changed within fifteen days of two guide pages going live. What it produces in traffic terms is currently modest: one percent of AI-summary views turn into a click on a cited source.
The honest version is: do the content work, skip the tooling, test it yourself with dates, and size your expectations to the numbers on this page rather than to someone's forecast. That's less exciting than a new discipline with a new acronym. It also happens to be what Google published.
If your next question is who to hire for it, we ran the same checks across ten Malaysian agencies selling AI search optimisation — ourselves included. The short version: we came top on the technical checks and no AI engine recommends us, while the agency scoring zero gets named by three. That gap is worth understanding before anyone quotes you a price.
Want to know whether you're named when someone asks AI about your industry? Start with a free visibility check — we run the queries, show you what came back, and tell you where you actually stand. If the answer is "a competitor," we'll say so.