Buyer Guide

How to Evaluate a GEO or AI Visibility Agency

Seven questions worth asking any agency before you sign, including us. Answers you can verify beat claims you cannot.

This page is a decision framework, not a comparison against named competitors. Every question below can be asked of any agency in this category. We publish it because the questions that are hardest to answer are the ones most worth asking, and because we would rather be measured against a clear standard than a vague one.

1. How do you count a single AI citation?

AI answers vary between platforms and between sessions. Without a stated counting rule, any citation number is unverifiable.

A strong answerA written rule: which platforms, how many runs per query, how a mention is distinguished from a link, and how duplicates are handled.

A weak answerA screenshot of one AI answer, or a number with no method attached.

2. What is in the prompt panel, and will you hand it over?

If the set of questions changes between reports, the movement you see reflects the panel, not your brand.

A strong answerA frozen panel of 75 to 100 real buyer questions, written down and given to you, so you can run it yourself at any time.

A weak answerThe panel stays with the agency, or changes each month without notice.

3. How many AI platforms do you measure, and separately or combined?

Answers differ meaningfully between ChatGPT, Gemini, Perplexity, Claude, Copilot, and Grok. An average hides which platforms actually cite you.

A strong answerSix platforms measured and reported separately, so you can see where you are strong and where you are absent.

A weak answerOne platform, or an aggregate figure with no per-platform breakdown.

4. Do you measure share of voice, or only whether the brand appeared?

Appearing once in a hundred answers and appearing in a quarter of them are very different outcomes, but both count as appearing.

A strong answerShare of voice against named competitors agreed in advance, with a published threshold for what category leadership looks like.

A weak answerPresence reported as a yes or no, with no competitor baseline.

5. Do you track how the brand is described, not just whether it is named?

Two brands can hold similar visibility while the language attached to them differs sharply. A citation that describes a business wrongly is not a win.

A strong answerDescription tracked and reported, with inaccurate descriptions treated as work to be corrected.

A weak answerOnly mention counts, with no record of what the AI actually said about you.

6. Is your methodology published, or only described in the pitch?

A method that exists only in a proposal cannot be checked by anyone, including you.

A strong answerResearch published with a registered DOI and indexed where academics and journalists can find it, including which data points were rejected.

A weak answerClaims about proprietary methods with nothing publicly documented.

7. What will the closing report say about targets you missed?

A document listing only achievements is marketing. One that lists failures is a record.

A strong answerA written commitment that the final audit includes targets that were not met, and why.

A weak answerNo mention of what happens if a target is missed.

What each metric actually tells you

Metric	What it means in practice
Ranking position	Not meaningful in AI answers. One answer names three to five brands in a single paragraph, with no first, second, or third.
Number of backlinks	An activity count, not an outcome. Easy to inflate and weakly correlated with AI citation.
Trust Flow	Measures the quality of linking sources rather than their number, which makes it considerably harder to game.
Share of voice	The portion of the answer space a brand occupies against named competitors within a topic cluster.
AI citation count	Useful only when the counting rule and the prompt panel are both stated and fixed.

If an agency cannot answer these questions, that is information. If we cannot answer one of them for your category, we will say so rather than guess.