Goppa Research
How an agency gets recommended by ChatGPT
A prospective client opens ChatGPT and types "best AEO agency in the UK". The answer names three or four agencies, and the shortlist is made before anyone visits a website. We measured what that shortlist is made from, and the answer is not what most agencies are working on.
The starting point: almost nobody is named
We measured 81 agencies that sell AI visibility, answer engine optimisation or generative engine optimisation, using the question their own prospective clients would type, in their own country. 61 of 81 appear in none of the three questions we ran. Of the agencies whose positioning is explicitly AEO or GEO, 53 of the 67 that sell exactly this outcome appear in none of the answers.
So if your agency is not named today, you are in the normal position. The useful part is why.
Your website is not where the decision is made
In 0 of the 81 measurements, the assistant cited the agency's own website as a source for the answer about that agency's category. Not one.
That changes where the work goes. Schema, an llms.txt, a
clean FAQ: they make your site readable, they are cheap, and you should have
them. But they fix a page the assistant did not cite when it chose who to
name. Selling them to yourself as "the AI visibility work" is the same mistake
as selling them to a client that way.
The three kinds of page behind its answers
| Source | Measurements | How you get in |
|---|---|---|
| Google results for the question | 68 of 81 | Rank for the question |
| Other agencies' comparisons | up to 11 of 81 | Be worth including |
| Agency directories | up to 7 of 81 | Submit, no gatekeeper |
Counted once per measurement, as a source behind the answer (a page the assistant's web search returned or cited): Google search results in 68 of 81. Semrush's agency directory (7 of 81), Clutch (6 of 81) and Sortlist (5 of 81). The three comparison articles that came up most often were each there in 6 to 11 of the 81. These are not equally open, and the order below follows how open they are, not how often they appear.
What to do, in order
1. List the agency in the directories behind the answers
Semrush's agency directory, Clutch and Sortlist have no gatekeeper. This is the only route on the list you can finish this week, and almost nobody does it properly because it is unglamorous.
Write the profile in the words of the question, not the words of your brand: the service as a client names it ("AEO agency", "GEO agency", "AI search optimisation"), the country and the city, and what you actually deliver. An assistant matching "AEO agency in Manchester" can only pick a profile that says AEO and says Manchester.
2. Publish the comparison that is already behind the answers
In your category, someone has written "the best AEO agencies in [country]", and it is among the sources behind the answer that decides who is worth naming. That page exists whether you write it or not. If a competitor writes it, a competitor decides the shortlist.
If you write it, write it as a comparison: several agencies, how you chose them, what each is good for, and yourself among them with the same treatment. A page that names only its author is a sales page with a comparison title, and it is a different kind of page from the ones behind the answers.
3. Rank for the buying question
Google results were behind the answer in 68 of the 81 measurements, far more than any single page. The answer is assembled from what ranks for "best AEO agency in [country]", which makes this ordinary SEO aimed at a very specific query: a service page per country or city you actually serve, with the phrase a client would type.
4. Get named on pages you do not own
Client case studies on the client's site, industry round-ups, podcasts, local business lists. It is the slowest route and the one with the most gatekeepers, so it comes last, not first.
5. Measure the same questions again
One reading is a starting point, not a verdict: assistants vary by day and by phrasing. Write down the questions, the country and the date, and run the same ones again in a month. That comparison is the only evidence that any of the above moved anything.
What not to do
- Do not rebuild your homepage for this. It was cited in none of the measurements. Fix what is broken on it, then spend the time on the pages behind the answers.
- Do not pay for a guaranteed position. Nobody controls which agencies an assistant names, so a guarantee is a promise nobody can keep.
- Do not ask the assistant to recommend you, over and over. Conversations do not train the model. The answer comes from the pages its search finds, which is why the list above is about pages.
The limits of what we just told you
The 81 agencies are the set our own collection reached between August and September 2026, mostly in the United Kingdom, not a census; a different collection would produce a different set. Three questions per agency is a coarse instrument: it supports the proportion appearing at all and which pages were behind the answers, not a ranking of agencies. The sources are counted once per measurement, and assistants vary by day and by phrasing, so a page that showed up rarely here may show up more on a question we did not ask.
How to cite: Goppa Research (2026). How an agency gets recommended by ChatGPT. https://trygoppa.com/answers/how-agencies-get-recommended-by-ai — CC BY 4.0.
No agency is named for a poor result. We publish the counts and the method; the identity of individual agencies stays out of it.