How to choose a GEO agency: questions, proof and list traps

To choose a GEO agency, ignore the rankings and test the proof. Ask each shortlisted agency for its exact prompt set and run counts, raw before-and-after exports, the specific changes it shipped for a comparable client, and how much of its work happens off your website. Then check who publishes the "best agency" list you found them on.

That last step matters more than it sounds, because the lists that rank for this query are often written by the agencies in them.

01Why are so many "best GEO agency" lists unreliable?

Search for "best GEO agencies" and you will mostly find listicles. Several of the most visible ones in 2026 are published by an agency that places itself at number one, sometimes with a disclosure and sometimes without.

We checked three of them. Xpand Digital's July 2026 ranking puts Xpand Digital first and states the conflict.^[1] Pepper's August 2026 list ranks Pepper first and says "we are on the list."^[2] Optimist's June 2026 list ranks Optimist first, and its author is candid about the pattern:

Disclosure is better than none, but a disclosed conflict is still a conflict. None of these lists is a neutral benchmark, and some publish no method you could reproduce.

02Do these lists actually influence what AI recommends?

Yes, which is why the problem is bigger than one bad search result. Ahrefs analyzed 26,283 source URLs behind ChatGPT answers to 750 recommendation prompts, including agency recommendations, and found that "best X" blog lists were 43.8% of all cited page types.^[3]

Peec AI's study of 232,000 citations found that about 11% came from self-promotional listicles, with no sign of the engines correcting for it over 12 weeks.^[4] So when you ask an assistant "which GEO agency should I hire," part of the answer may be an agency's own marketing, restated.

03Are paid placements on review directories the same thing?

Not quite, but you should know which part of a directory page you are looking at. On Clutch, for example, sponsors are listed above non-sponsors on directory pages, while Clutch states that sponsorship does not affect its Leaders Matrix ranking, and visitors can sort by Clutch Rank instead.^[5]

The legal backdrop is also shifting. The U.S. Federal Trade Commission settled with comparison site LendEDU in 2020 over rankings it claimed were unbiased while accepting payment for position,^[6] and the FTC's 2024 rule on fake reviews bars businesses from misrepresenting a site they control as providing independent reviews of a category that includes their own products.^[7] That does not make every self-ranked list unlawful, but it tells you which questions to ask.

04What questions should you ask a GEO agency?

Once you have a shortlist, the conversation should move from claims to evidence. These questions separate agencies that operate a measurable program from those that rebrand an SEO retainer.

One founder who evaluated ten agencies summarized the most common failure bluntly: "Most agencies selling GEO and AEO right now are selling repackaged SEO."^[8] Notably, the same post then ranks the author's own new agency first, with a disclosure, which shows how common the pattern is. The questions below are designed to test for substance either way.

05What proof should you demand?

Screenshots are not proof. AI answers vary from run to run: a SparkToro and Gumshoe study of nearly 3,000 runs found that the same prompt returned the same brand list less than once in a hundred times.^[9] A single screenshot of your brand in ChatGPT can be real and still meaningless.

Ask instead for evidence that survives that variance.

Measurement methods also matter for comparing agencies fairly. If each vendor measures differently, you cannot compare their numbers; our guide to AI share of voice explains what a defensible metric looks like.

06Should an agency's own AI visibility count?

Partly. An agency that appears when you ask an assistant for GEO help has clearly learned something about how answers are built. But as the list studies above show, that visibility can come from its own self-ranked content, so it proves marketing skill more than client results.

A fairer test is to ask the agency to show its own measurement of itself: the prompts it tracks for its brand, the engines, the run counts and where its mentions are sourced from. If it cannot measure its own visibility rigorously, it is unlikely to measure yours better.

Also check reference clients in your sector. Visibility for a SaaS tool and visibility for a bank, a hospital group or a property developer depend on different sources, review sites and regulatory limits.

07What are the red flags in a GEO proposal?

Some warning signs show up in almost every weak pitch. Guaranteed placements are the clearest one: no one controls what an answer engine says on a given run, so a guarantee means the metric is defined loosely enough to always be met.

Other red flags are quieter. A proposal that is 90% blog content ignores how much weight engines give to third-party sources, which we cover in the sources AI trusts. A plan with no step for correcting wrong facts ignores the risk that worries legal teams most. And a contract where the agency owns the tool accounts means your history leaves with them.

08Should you run a paid pilot first?

For most enterprises, yes. A short, paid pilot with a fixed scope tells you more than any pitch deck, because you see how the agency works inside your approval chain, not just what it promises.

Keep the pilot narrow: one product line, one or two markets, a prompt set you define, and a fixed number of weeks. Agree in advance what "done" means, such as a baseline report, a traced list of wrong or missing answers, and a set number of approved changes shipped.

1. 1
   Define the prompt set yourselfBefore kickoff

   Write the 20–40 questions your buyers actually ask, per market and language. Give every agency the same set.
2. 2
   Agree the baseline methodWeek 1

   Engines, run counts, logged-in or logged-out, locations and date range. Write it into the pilot order.
3. 3
   Review the traceWeeks 2–3

   The agency maps each wrong or missing answer to likely sources. Check whether the sources make sense to your experts.
4. 4
   Ship a fixed number of changesWeeks 3–8

   On-site and off-site, through your normal approvals. Track how long each one waited for sign-off.
5. 5
   Re-run and decideEnd of pilot

   Same prompt set, same method. Judge the work quality and the process, not only the movement in a few weeks.

Be realistic about what a pilot can show. Answers that depend on model memory may not move inside a pilot window, so judge the agency on the quality of its diagnosis and how cleanly changes moved through your process.

09How should you run the final selection?

Treat the final round like any enterprise procurement. Write down your criteria and weights before the pitches, give every agency the same brief and prompt set, and score each response independently before discussing it.

A structured process is easier with a template; our GEO agency RFP template includes copy-ready sections and a weighted scoring rubric. Compare commercial terms against market ranges in GEO agency pricing.

It also helps to decide whether you need an agency at all, or a partner that works as part of your marketing team. OwnlyReal is built as the second kind: tool-agnostic, measuring share of answer and shipping fixes through the client's CMS and approval chains. Whichever you choose, hold it to the same proof standard, and read our notes on what enterprise teams should expect.