Why the first call decides more than the proposal
The first call decides more than the proposal because AI visibility is a measurement problem before it is a marketing problem. Only 34 to 42 percent of the sources an AI engine cites are the same from one day to the next, and identical prompts repeated minutes apart vary just as much, according to a University of St. Gallen study across four engines and four sectors. An agency that does not know this will sell you a screenshot. An agency that does will talk about sample sizes before it talks about price.
The questions below are the ones we would ask if we were buying. Each comes with what a good answer sounds like and the research it rests on.
The nine questions
1. "Can you show me whether AI names us today, before we sign?"
A serious agency runs a baseline before it quotes. Our own free analysis queries the four main assistants for your category and market and reads your key pages; anyone competent can do the same. If the answer is "we'll find out once we start", the agency is asking you to pay for the diagnosis.
2. "How many times do you run each query?"
Once is a screenshot, not a measurement. The St. Gallen team found that reliable brand-visibility tracking needs at least seven runs per prompt and an observation window of ten days or more. An agency that reports a single ChatGPT answer as your "AI visibility score" is reporting noise. Ask how many runs, how many days, and whether the report shows the spread.
3. "Which engines, and why those?"
The engines behave differently enough that one number cannot stand for all of them. Across 21,143 citations, ChatGPT cited 6.9 sources per answer, Google 12.1 and Perplexity 16.4, and ChatGPT absorbed each source far more deeply than the others, according to Zhang, He and Yao. A good answer names the engines that matter for your buyers and explains the trade-off; a weak answer says "all of them" and means one.
4. "What do you actually change on the site?"
The honest list is short and specific. In the founding GEO experiments at KDD 2024, adding statistics, citations and quotations raised visibility in generative answers by up to 40 percent, while classic keyword tactics did little. Zhang's data points the same way: pages with numbers, definitions and comparisons were absorbed 55 to 62 percent more strongly, and Q&A formatting on its own showed no advantage. If the agency's plan is "FAQ sections and schema", ask what evidence goes into them.
5. "How do you build the brand as an entity, not just a website?"
Brand mentions across the web correlated more strongly with AI visibility than Domain Rating or branded search volume in Ahrefs' analysis of 75,000 brands. An agency that only touches your own pages is working on the weaker half. Ask what they do about consistent organisation data, third-party mentions, reviews and directory presence. Our brand entity piece explains why this signal sits at the top.
6. "Do you check whether AI crawlers can reach our pages?"
A page GPTBot cannot fetch cannot be cited, whatever it says. A surprising number of sites allow crawlers in robots.txt and then block them at the server or CDN. Ask whether the agency tests live bot access, not only the robots file. It is a ten-minute check that decides whether anything else works.
7. "How will Google SEO and AI visibility be reported side by side?"
They are not the same list. Across 1,000 consumer ranking queries, GPT-4o's cited domains had a median overlap of 0.0 percent with Google's top 10, in the University of Toronto analysis presented at EDBT/ICDT 2026. Google's own AI Overviews still lean on organic results, so the two need separate reporting and, often, separate work. An agency that folds AI into a rankings report is hiding the gap. See SEO vs GEO for the full picture.
8. "What will you refuse to promise?"
The right answer includes "a guaranteed mention". Because the engines are probabilistic and their sources change daily, nobody can promise you will be named on a given day. What can be promised is a measured baseline, a documented set of changes, and a re-measurement on the same method. Anyone promising position one in ChatGPT is selling something that does not exist.
9. "Who does the work, and can I speak to them?"
Ask to meet the people who will read your pages and write your evidence. AI visibility work is editorial as much as technical: someone has to find the numbers, cite the sources and define the terms. If the account manager cannot name that person, the work is being outsourced or automated.
What the answers should add up to
| Question | Good answer | Warning sign |
|---|---|---|
| Baseline before signing | Free scan, four engines, your market | "Once we start" |
| Runs per query | 7 or more, 10 or more days, spread shown | One screenshot |
| Engines | Named, with reasons | "All of them" |
| On-site changes | Statistics, sources, definitions, comparisons | "FAQ and schema" |
| Entity work | Mentions, reviews, consistent org data | Own site only |
| Crawler access | Live bot test | robots.txt only |
| Reporting | Google and AI reported apart | AI folded into rankings |
| Promises | Method and re-measurement | "Guaranteed mention" |
| People | Named editor you can meet | Account layer only |
Five or more warning signs means the agency has renamed SEO. That is not a crime, and SEO still matters; it is simply not the thing you were asked to buy.
Frequently asked questions
Is an AI search agency the same as a GEO agency?
Yes, in practice. GEO (generative engine optimization), AEO (answer engine optimization) and "AI search" describe the same work: making a business retrievable, citable and named by AI assistants. The label matters less than the method behind it.
Do I still need an SEO agency if I hire one for AI search?
Usually the same agency should do both, reported separately. Google's AI Overviews build on organic results, so classic SEO remains the entry ticket there, while standalone assistants draw from a largely different source pool. One team with two reports is the workable arrangement.
How long before AI visibility changes?
Weeks for access and entity fixes, months for citation patterns. AI engines cite fresher content than Google, so a corrected page can be picked up quickly, but the mention rate for a brand moves as third-party evidence accumulates. Ask for a three-month re-measurement, not a three-week one.
What should an AI visibility report contain?
Per-engine mention rates over multiple runs, the queries used, the competitors named instead of you, and the on-site and off-site evidence that changed. A single score without the runs behind it is a marketing number.
Can a small business afford this?
Often more easily than a large one, because niche brands are ranked on what the engines retrieve rather than on what they remember. That makes your own pages and a few trade mentions direct levers. Our small business article covers the mechanism.
How much does an AI search agency cost?
It depends on the gap the baseline shows, which is why the baseline should come first. Anyone quoting before measuring is quoting for a package, not for your situation.
Run the first question yourself
The free analysis answers question one before any call: whether ChatGPT, Perplexity, Claude and Gemini name your business for your category, what they cite instead, and where your pages fall short. Bring the result to whichever agency you talk to, including us.
