How to choose a GEO agency: 10 questions to ask before you sign
In shortIn your first meeting with a GEO agency, ask for a sample report. Pick a question your customers might ask and look at each assistant’s answer. Can the agency show the answers that only mention your brand separately from the answers that clearly recommend it to the buyer?
GEO (generative engine optimisation) is the work of getting a brand into the answers of AI assistants such as ChatGPT, Gemini, Perplexity and Claude. You can ask the ten questions below before you sign, and apply them to the report from the agency you already work with.
Is the measurement set up correctly?
1. How did you choose questions our customers would actually ask, and can you show today’s result for each one?
GEO work starts with a measurement. “Your AI visibility is low” is not a measurement. A good baseline report shows the list of questions, whether your name appeared for each one and, if it didn’t, who was recommended instead. Ask how the questions were chosen too: looking good on questions your buyers would never ask achieves nothing.
Questions that contain your brand name (“Is clinic X reliable?”) should be shown apart from questions that don’t (“Which clinic in Istanbul?”). It is easy to appear when someone asks about you by name. The real difference shows in the question from a buyer who doesn’t know you yet.
2. Do you measure each assistant separately?
Assistants can answer the same question from different sources. On 10 November 2026 we asked four assistants 15 questions that someone in the UK might ask about hair transplants in Turkiye. ChatGPT cited the website of the International Society of Hair Restoration Surgery (ISHRS) in 23 of its 45 answers. Claude’s sources included clinic lists. A clinic that appears often in one assistant may not appear at all in another. A single “AI score” hides that difference.
3. Are Google’s AI answers included?
Some buyers see an AI answer without opening an assistant, while searching on Google: AI Overviews and AI Mode. These answers should be measured separately too. Your Google ranking alone does not show your visibility there.
4. How did you choose our buyers’ language, and how did you set the country for each assistant?
An English question from a patient in the UK and a Turkish question asked from Turkiye can get different answers. Buyers in one market may also use more than one language. If you expect patients or customers from abroad, ask which language and which country setting the measurement used.
5. How many times do you ask each question?
An assistant’s answer can change even for the same question. In our study, one clinic was named in only one of Claude’s three answers to the same question about clinics run by experienced surgeons. A one-off screenshot can therefore look good or bad by chance. A reliable measurement asks the same question several times and reports the total.
Is the result read correctly?
6. Were we only named, or were we actually recommended to the buyer?
Your name appearing in an answer does not mean you were recommended to the buyer. An assistant can name a clinic and then say it found too little information to recommend it.
Example: in our 10 November study, one clinic’s name appeared in 20 of Claude’s 45 answers. That number alone does not show in how many answers the clinic was clearly recommended. The question to ask of a report is: “In how many of these was my brand only named, and in how many was it clearly recommended to the buyer?” A good report shows this with the sentence quoted from the answer.
7. Do the assistants say anything wrong about us?
Being visible is not enough. An assistant can give your price, services or address wrongly, and this happens more often when your website, marketplaces and promotional articles say different things. A report should show wrong or outdated information with the sentence from the answer, and state where the correct information comes from.
8. Which sources do the assistants cite, and is our own website among them?
Assistants cite sources in their answers: association websites, comparison platforms, news, competitors’ websites. Ask which source stands out in which assistant and in how many answers your own website is cited. The roadmap for where you need to be visible comes from here.
Is what comes next planned correctly?
9. When will you repeat the first measurement, and which questions will you keep fixed?
To see how visibility changes, measure the questions you keep fixed again with the same method. Show questions added later separately. Repeating the same measurement shows whether visibility changed; how much of the change came from the agency’s work is a separate question. Ask for the baseline measurement to be handed to you and for the repeat date to be written down from the start.
10. What do you guarantee: delivery of the report, or that our brand is recommended in answers?
An agency can commit to a delivery date, a scope and the work it will do. The answer an assistant gives to a specific question is outside the agency’s control. Make sure the proposal keeps these two apart.
Who should do the measuring?
Your agency can do both the work and the measurement. When you review the report, pick one buyer question and check whether the agency’s conclusion matches the answer the assistant actually gave to that question.
We built the GEOMIRA report around these ten questions. It shows each of them with your own buyer questions, and an expert reviews it before you receive it. We don’t sell implementation, so the measurement stays independent.
