You have just watched a sales demo. The GEO agency on the call showed a dashboard with your brand’s “AI visibility score,” a chart of competitors and a red arrow pointing down. The proposal promises to turn that arrow green within a quarter. The price is reasonable. The deck is polished. And you have no idea whether this firm can change what ChatGPT says about you, or whether it can only measure it.
That is the problem most buyers face when they search for the best GEO agencies. The category is new, the language is borrowed from SEO and almost every firm can produce an impressive chart. This guide gives you a way to see past the chart. It explains what GEO agencies sell, sets out a five-part test we call PROVE, compares the three agency models you will meet and covers contracts, pricing and what a good first quarter looks like.
What GEO agencies sell, in plain terms

GEO stands for generative engine optimization. The term gained traction after a 2023 research paper, “GEO: Generative Engine Optimization,” by researchers at Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi. The authors tested ways of rewriting web content to make it more likely to appear in AI-generated answers. Methods such as adding citations, quotations from credible sources and relevant statistics improved visibility in their test setup, with some gains reported as high as 40%. The paper gave a name to something marketers were already noticing: AI answers were starting to replace the list of links, and the rules for showing up in them were different.
In practice, a GEO agency sells three things, in varying proportions.
The first is measurement. The agency builds a set of prompts your buyers would ask, runs them across AI assistants on a schedule and reports how often you are named, how you are described, where you appear relative to competitors and which sources the assistants cite. Good measurement is the foundation for everything else. Poor measurement is the most common thing agencies oversell.
The second is on-site work. The agency restructures your content so assistants can extract and quote it: answer-first pages, clear definitions, honest comparisons, original data, structured data markup and technical fixes that help AI crawlers read your site. This is closest to classic SEO, and many GEO agencies grew out of SEO firms.
The third is off-site work. Assistants lean hard on what others say about you. That means earned media, expert quotes in publications, inclusion in comparison and “best of” articles, review platforms, directories, community discussions and, for some categories, Wikipedia and Wikidata. This is the hardest part to do well and the part that most often moves results, because in many categories the sources an assistant cites when recommending a brand are third-party pages, not the brand’s own site.
When you compare proposals, the first question is how each agency splits its effort across those three. A proposal that is 80% measurement is a reporting service. A proposal that is all on-site is SEO with a new name. A proposal that has no plan for third-party sources is missing the part that matters most for recommendations.
The PROVE test for choosing a GEO agency
We use a five-part test, PROVE, when clients ask us to help them compare the best GEO agencies on their shortlist. It works on any firm, including ours. Score each letter from zero to two. A total of eight or more is a strong candidate. Under six, keep looking.
P is for prompts. Ask the agency to show you the actual prompts it would track for you, before you sign. You want twenty to fifty prompts that sound like your buyers, not like keywords: “What’s the best payroll software for a 50-person restaurant group?” rather than “payroll software.” Ask how the prompts were chosen, how often they are run and on which assistants. Score two if you see real, buyer-language prompts across at least three assistants. Score zero if the agency only offers a composite score.
R is for references. Ask which sources the assistants cite today when answering your prompts, and which of those sources the agency plans to influence. This tells you whether the firm understands your category. A good answer is specific: these three trade publications, this review site, these two comparison articles, this subreddit. Score zero for a generic answer about “authority” or “backlinks.”
O is for outreach. Ask who does the off-site work and what they have achieved for other clients in the last six months. If the agency pitches journalists, ask for examples of placements. If it works on review platforms, ask how. If it partners with a PR firm for this, ask to meet them. Many GEO agencies are strong on measurement and content but thin on outreach, and outreach is where recommendations are often won.
V is for verification. Ask how you will know the work caused the change. Assistants vary from one run to the next, update their models and change how they search. A credible agency reruns prompts several times, tracks trends rather than single snapshots, separates assistants that search the live web from those that rely on training data and is honest about what it can and cannot attribute. Score zero for any agency that guarantees a position or claims precise control over AI answers.
E is for exit. Ask what you keep if you leave: the prompt set, the historical tracking data, the content produced, the media relationships built on your behalf. Ask about minimum terms and notice periods. A confident agency offers a 90-day review point and lets you export your data.
Three agency models and who they fit

The first model is the measurement-led specialist. These firms often started with their own AI visibility tracking software and added services. Their strength is data: frequent prompt runs, broad assistant coverage, clean reporting and sharp analysis of which sources drive answers. They suit companies with a capable in-house content and PR team who need a clear view of where they stand and what to fix. The risk is that the service stops at insight. If you do not have people to act on the reports, you are paying for a very detailed picture of a problem.
The second model is the SEO firm that added GEO. These agencies bring deep technical and content skills: site audits, structured data, content rewrites, internal linking and crawl issues. They suit companies whose own site is the weak point, with thin or badly structured pages that assistants struggle to quote. The risk is underinvesting in third-party sources. Ask how much of the plan happens off your site.
The third model is the PR or communications firm that added GEO. These firms are strong at earned media, executive visibility and reputation, the third-party signals that shape what assistants say about a brand. They suit companies that already have a solid site but little coverage in the publications their buyers and the assistants trust. The risk is weak measurement and technical skill. Ask how they track prompts and who handles the on-site work.
No model is best in general. Choose by your biggest gap. If you do not know where you stand, start with measurement. If your site is the problem, start with an SEO-led firm. If no one outside your company is talking about you, start with a PR-led firm. Some companies end up using two firms, a measurement specialist and a PR agency, with clear ownership of each part. That can work well if both teams share the same prompt set and meet regularly.
Contracts, pricing and the first quarter
GEO pricing has not settled into standard packages, so expect quotes that vary a great deal for what sounds like the same service. Most proposals follow one of three shapes. A one-time audit usually covers a prompt baseline, a review of your site and a map of the sources assistants cite in your category, with a prioritized plan. A monthly retainer covers ongoing tracking, content and some level of outreach, with the price scaling with how much content and earned media is included. A hybrid adds performance elements, such as fees tied to placements or new citations.
When comparing quotes, ask each agency to break down its fee across measurement, on-site work and off-site work. That makes comparisons fair and shows you what you are buying. A low retainer that is almost all reporting is not cheaper than a higher one that includes regular earned media. It is a different product.
Watch for warning signs in the contract. Twelve-month minimums with no review point, vague deliverables such as “AI optimization” with no defined outputs, ownership clauses that keep your data or content with the agency and guarantees of specific rankings or placements in AI answers should all prompt hard questions.
A good first quarter has a clear shape. In the first month, the agency agrees the prompt set with you, runs the baseline several times across your chosen assistants, maps the cited sources and fixes the fastest on-site problems. In the second month, content rewrites go live, and outreach begins toward the third-party sources that matter most. In the third month, the agency reruns the full prompt set, compares it against the baseline assistant by assistant and explains what moved, what did not and why.
Expect uneven results. Assistants that search the live web, such as Perplexity and ChatGPT with search, can reflect new coverage within weeks. Answers that depend on a model’s training data change more slowly and often only after a model update. A good agency will show you that difference instead of averaging it away.
The best GEO agencies are not the ones with the most polished dashboards. They are the ones that can show you real prompts, name the sources that matter in your category, earn coverage in those sources and prove what changed. Run every candidate through PROVE, and bring a baseline of your own to the first call. Our free AEO rating will give you one in a few minutes, so you can judge each agency against your own numbers rather than theirs.