What should you ask a GEO agency before you hire one?

A convincing methodology page is easier to produce than a record of measured results. GEO buyers need to distinguish what an agency says it will do from the evidence that its work has helped a business.

Buyers have started showing up with checklists written by the assistants themselves — they ask ChatGPT what to ask us before they ever send an email. In August 2026 we ran that exercise on ourselves: we asked ChatGPT, with browsing enabled, to assess get-geo.ai as a potential vendor. It scored our understanding of GEO, our multilingual positioning, our measurement protocol and our transparency at 9 out of 10 each — and our proven client results at 5, our maturity as an agency at 4. That is one run of one model producing a heuristic, not a metric, so hold the numbers loosely. The gap it named is not loose at all, and it is the right gap: knowing how to make a GEO agency visible is not the same as having proved you can do it for somebody else's business.

So here are the ten questions that exercise produced, what a good answer to each one contains, and our own answers — including the first question, where our answer is currently weak.

Every one of these separates process from evidence. The pattern to listen for is specificity: a good answer names a number, a language, a source or a date. A weak answer names a capability.

Send the questions in writing and keep the replies. Asking each shortlisted agency the same questions gives you answers you can compare directly.

Question one is where we are weakest, so it goes first. We do not yet have a client before/after dataset to show you. What we have instead is our own: a public case study in which the subject is us, run under a clean-session protocol we published, with its limitations stated inside the case rather than in a footnote. That is a real measurement and a poor substitute for a client result, and we would rather say so than dress it up. What we have changed is the design of the work: every pilot is scoped so that it produces a publishable anonymised dataset — prompts, baseline, after, per language, brand removed — from the first engagement onward. The first client to sign is buying a discount and contributing the proof.

Here are our answers to the remaining questions.

Most AI-visibility reporting collapses four different things into a single score, which is how a dashboard can climb while nothing changes for the business. Ask any vendor to separate them: mention rate, whether you were named at all; recommendation rate, whether you were the answer rather than an item in a list; share of voice, your appearances against the competitors the assistants themselves name; and citation quality, which sources the model leaned on when it named you. Being named is one achievement. Being named because the model read an industry study, a review platform and your own documentation is a different and much sturdier one.

Ask for written definitions of each metric so you can check whether the counting rules remain the same. Our guide on how we measure provides the full definitions, protocol and 90-day policy.

Give each shortlisted agency the same small task. Ask three or four providers for ten questions where you do not currently appear, the complete answers, named competitors, proposed changes for the first 30 days and a price for a 90-day pilot.

Compare five things: whether the questions reflect your customers’ needs; whether the competitors are relevant; which cited sources the agency records; whether the metrics have clear definitions; and whether the pilot has a defined scope and price. The quality of this response is useful evidence of how the agency will report later.

Our version is free. Email hello@get-geo.ai and we will send the audit with its ten questions. You can compare it with equivalent work from other providers.

These signs do not establish that an agency is dishonest. Each points to something you may be unable to verify later, so ask for clarification before agreeing to the work.