What should I get answered before buying a GEO service?

The questions to get answered before buying a GEO service fall under three headings: what will be measured and how, what the work itself consists of, and what is being promised. GeoMono is a platform that measures how visible brands are in AI answers; this guide lists ten questions you can put to any GEO service (ours included) and what a good answer to each looks like.

The short answer: if a service cannot explain its measurement plainly, you will not be able to judge its results either. The first question is always "what exactly do you measure, and how?"

Questions about the measurement

1. Which questions will you ask, and are they written down?

Visibility is measured against the questions asked. Change the question set and the result changes. A good answer: the questions are fixed in writing before the measurement, shown to you, and kept the same between measurements. They should fit your business: measuring consumer questions for a company that sells to enterprises sets the measurement up in the wrong place from the start. Review the question set together and make sure it reflects your real buyers' questions.

2. Will our brand's name appear in the questions?

An engine asked "Is X reliable?" will name X regardless. That is an echo of the question, not visibility. A good answer: visibility is measured with questions that do not contain the brand's name, the way a buyer would ask before knowing any brands. Questions that do contain your name can serve a different purpose (are you described correctly?), but should not be mixed into the visibility rate.

3. Which engines will you measure, and will results be reported separately?

ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews read different sources and can recommend different brands. A good answer: each engine's result is reported separately, with the number of questions measured on that engine. A report that folds five engines into one score hides which engine changed. The model version measured should be recorded too; when the version changes, results can change.

4. What happens when an engine does not answer?

Engines time out and return errors. If such an answer is counted as "not mentioned", your rate drops unfairly; if it is not written down at all, you cannot know how many questions were really measured. A good answer: a failed answer is left out of the count, written down separately and retried. The report shows how many questions were planned and how many were measured. That way a provider error is never read as a finding about your brand, and the missing measurement can be completed later.

5. How will you verify that it was really our brand that was named?

Your name appearing in an answer does not prove you were named. It may be another company with the same name. A good answer: every mention passes a check that filters out namesakes, and the different spellings of your name are agreed in advance.

6. If the same question gives different answers, how will you decide a change is a change?

Research published by SparkToro in January 2026 found that AI tools return very different lists of brands when the same question is asked repeatedly. So not every difference between two measurements is a real change. A good answer: two measurements are compared only with the same questions and the same rules, and a difference is called "up" or "down" only when it is larger than asking the same questions again could produce.

Next step

Read how we measure

The method page explains which answers are counted, what happens when an engine does not answer, and when we do not publish a figure.

Read the method

Questions about the work

7. What happens after the measurement, and who does it?

GEO is not only measurement; closing the gaps the measurement shows is part of the work. Those gaps can be on your own site (access, content, outdated information) or beyond it (directories, comparison sites, complaint platforms, Wikipedia). Ask which work the provider will do, which your own team will do, and which is in nobody's hands.

The last group matters. Neither you nor your provider decides what a directory, a news site or Wikipedia publishes. A good answer says that limit out loud, and separates what can be done on those sources (sending the correct fact to the page owner, answering a complaint, proposing a correction with an independent source on a Wikipedia talk page) from what cannot. We summarise Wikipedia's own rules in a separate post.

8. Will you show which sources our competitors are recommended from?

The most concrete way to understand why a competitor is recommended is to look at the sources the answers rely on. A good answer: the sources the answers cite are recorded engine by engine, and the brands named alongside you are listed separately. But be careful: it has not been shown that appearing in a source will change the answer. A report that presents this as a certainty goes beyond what was measured.

9. Will I be able to see the raw answers?

Every number in a report comes from individual answers. A good answer: behind each rate you can see the answers, the date each question was asked, the engine and the sources cited. A number whose answers you cannot see is a number you cannot check.

Access to the answers is what makes a report more than a summary. Only by reading the full text can you see whether a mention is really about you, why a competitor was recommended, or in which sentence an error appears. If your team disputes a finding, the answer itself settles it.

Questions about the promise

10. What exactly are you promising?

This may be the most important question. No method can guarantee that an AI engine will recommend you: engines do not give the same answer to the same question every time, and they decide for themselves which sources to read. Google's developer documentation says no extra markup or special file is needed to appear in AI Overviews; be cautious with a service that promises visibility through a special "AI file".

Academic work offers hypotheses, not promises either. The KDD 2024 paper the term GEO comes from measured, in its own benchmark, that adding statistics and quotations to sources raised their visibility; that is an idea to test on your site, not a guarantee.

A good answer: the service promises what it will measure, not what will change. Instead of "we will raise your visibility by this much", it says "we will ask these questions, on these engines, at these intervals, and report the result, whether up, down or unchanged, by the same rules".

How GeoMono answers these questions

We publish our own answers on the How we measure page. In brief: questions are registered before the measurement in a written, versioned declaration for the brand, and are written without the brand's name. Every question is asked of each of the five engines in the same way; there are no weights between engines and the results are not folded into one score. For each engine we write "named in x of y questions". A failed answer is left out of the count, recorded and retried. Every mention is verified with namesake and echo detection. Two measurements are compared only with the same questions and the same rules.

How an agency can structure a monthly report is covered in our monthly GEO report template.

You can put the same questions to tools: the AI visibility tools comparison sets GeoMono and other tools side by side from their own published information. What a GeoMono subscription includes is written out in What a GeoMono subscription includes.

Frequently asked questions

What is the single most important question to ask a GEO service?

"What exactly do you measure, and how?" If a service cannot explain its measurement plainly, you cannot judge its results. Ask about the question set, the engines, how failed answers are counted and how two measurements are compared.

Should I trust a service that guarantees higher visibility?

Be cautious. AI engines do not give the same answer to the same question every time, and they choose their own sources. A service can commit to the measurement and to the work, not to the outcome.

Is it a problem to change the question set every month?

Yes, if you want to compare two measurements. When the questions change, the difference between the two may come from the questions rather than the work done. If you want to add new questions, keep asking the old ones and report the two sets separately.

Why does seeing the raw answers matter?

Every rate in a report comes from individual answers. If you can see them, you can check for yourself whether a mention is really about you and why a competitor was recommended.