Generative engine optimization
Generative engine optimization: there is no position to hold. An assistant drew on about 6 businesses per question and named about 4 of them at a time.
GEO is the name people give to getting a business into an AI answer. We measured what that answer is made of: across 45 questions and 103 readings on a logged-in assistant, each question had a pool of about 6 businesses behind it, and roughly 4 were shown in any one answer.
4 of 6
businesses named at a time, out of the pool an assistant drew on for one question
What we measured
- Across 45 questions and 103 readings, an assistant drew on a mean of 5.62 businesses per question and named a mean of 4.16 of them in any single answer.
- The pool behind a single question ranged from 2 businesses to 10, so 'how many can be named' is a property of the question, not of the market.
- Of 253 business names seen across those readings, 119 came back in every repeat of their question and 134 did not.
- Generative engine optimization has no equivalent of an organic position: the unit competed for is membership of a pool of about 6, from which about 4 are shown, re-drawn on each answer.
What the term means, and what it does not
Generative engine optimization is the work of getting a business named when someone asks an AI assistant instead of typing into a search box. The name is new; the thing it describes is measurable, and that is where most of the confusion goes away.
It is not a ranking discipline. There is no first place to win, because there is no list with fixed seats. What an assistant produces is a handful of names drawn from a larger set it considered - and the set is redrawn every time the question is asked.
The shape of the thing you are optimizing into
We asked 45 questions on a logged-in assistant and recorded 103 answers word for word. Behind each question sat a pool of businesses the assistant would draw on: a mean of 5.62, with the narrowest question producing 2 and the widest 10. Any single answer showed a mean of 4.16 of them.
So the target is not position one of ten. It is membership of a set of roughly six, from which about four are shown. That is a materially different problem, and it is why a page promising you 'the top spot in ChatGPT' is selling something that does not exist.
| What we measured | Figure |
|---|---|
| Questions asked | 45 |
| Answers recorded word for word | 103 |
| Businesses in the pool behind a question, mean | 5.62 |
| Businesses named in a single answer, mean | 4.16 |
| Narrowest pool / widest pool | 2 / 10 |
45 questions and 103 answers recorded word for word on a logged-in assistant in US metros. Pool size is a property of the question asked, so these means travel only as far as the questions behind them do.
Why the pool moves
Of the 253 business names we saw across those readings, 119 came back every time their question was repeated and 134 did not. Roughly half the set is stable and roughly half rotates.
That is the honest reason this work needs measuring rather than declaring. A single reading tells you who was named once. It does not tell you whether you are in the stable half.
The questions this page answers
generative engine optimization
4,400/mogenerative engine optimization geo
1,600/mogenerative engine optimization strategy
90/mo
3 questions, 6,090 searches a month between them. All 4 answers we examined on them were produced by a web search, which means a page can reach every one.
Who an assistant cites on these questions today
- en.wikipedia.org 4
- coursera.org 4
- youtube.com 4
- evergreen.media 3
- forbes.com 2
- adsmurai.com 2
- reddit.com 2
- semrush.com 1
- directiveconsulting.com 1
- powerdigitalmarketing.com 1
13 distinct domains are cited across these questions; the 10 above are the most frequent, with the number of answers each appears in. Counted from the answers captured in our corpus, not from a market census.
What to ask before you buy
- Ask any GEO vendor how large the pool is for your question and how many of it get shown. If they answer with a rank or a score instead of a count, they have not measured a pool.
- Treat any promise of a fixed position as a signal to stop. The measured object has no fixed positions.
- A single reading is a sample of one. Ask how many readings a claim rests on before you pay for it.
Honest limits of what we measured
- These figures come from 45 questions and 103 readings on one surface - a logged-in assistant, one account, US metros. Another surface will produce different numbers and we would not carry ours across without measuring again.
- Pool size is a property of the question asked. A narrower question produces a smaller pool, so a figure quoted without its question is not a figure.
- We do not know the click-through rate from being named in an answer. Nobody has published one we would trust, and we will not estimate it.
Questions we get asked
Is GEO the same thing as AEO?
In practice the two names describe the same work, and questions asked under either abbreviation return the same kinds of answer. We keep separate pages because the domains currently cited on each differ, not because the discipline does.
Does GEO replace SEO?
No, and the measured reason is that they optimize different objects. A search position holds still until something changes it. A pool is redrawn on each answer, with about half its members rotating between readings. Work aimed at one does not automatically move the other, and we do not measure classic search rankings at all.
Can you guarantee we get into the pool?
No. We measure whether you are in it, who is in it instead, and what the answers were built from. Anyone guaranteeing membership of a set they do not control is guessing.
How many readings does it take to know where we stand?
More than one. About half the names we saw rotated between repeats of the same question, so a single reading cannot separate 'not in the pool' from 'in the pool and not shown this time'.
Start with one free reading
One business, real questions, the raw answers back - before you decide on anything paid. If you are running this for a book of clients, say so and we will talk terms rather than quote a page price.
Free reading
Takes about a minute.
Related
Pricing is per category, because the work differs by how competitive the answers are. See every category we measure and what a reading costs →
Published 31 August 2026. Every figure on this page is recomputed from raw observation files by a single script, so any of them can be traced back to the readings behind it. We make your site citable. We do not promise placement in any assistant’s answer - we measure whether you are in it, and show you the raw answers either way.