How to Choose a GEO Agency
How to choose a GEO agency without paying for rebranded SEO. Ten questions to ask, the evidence to request, and the red flags. Use them on any provider, including us.
By JoLyn Laney, FounderLast updated: September 27, 2026How do you choose a GEO agency?
Short answerChoose on evidence. A legitimate GEO provider can show the exact questions it tested, the platforms and dates, the actual AI responses, the competitors named, and the sources cited, and it will measure the same questions again later. Be cautious of any provider offering a guaranteed recommendation or a score it cannot break down.
None of this is about price. A low-cost audit can be honest and an expensive one can be empty. The test is whether what you are handed is evidence you could check, or a number you are asked to trust.
One fact shapes everything below. A search ranking can be checked again tomorrow. An AI answer cannot. Ask the same question twice and the platform may name different businesses and cite different sources. So the value of any GEO engagement rests on how carefully it records what it saw.
What does a GEO agency do?
Short answerA GEO agency measures how AI platforms such as ChatGPT, Gemini, and Perplexity name and describe a business, then works to improve it. The work includes testing customer questions, recording who is recommended and which sources are cited, correcting inaccurate information, and improving the business's website and its presence on third-party sources.
GEO stands for generative engine optimization. Some providers call the same work AEO, AI SEO, or AI visibility. The label matters less than whether the provider measures what AI platforms actually say.
How can you tell GEO from rebranded SEO?
Short answerAsk what is measured. SEO reporting measures rankings, impressions, and clicks in search results. GEO work should measure what AI platforms say: who is named, who is recommended, whether the facts are right, and which sources are cited. If the deliverable is a technical site audit with a new title, it is an SEO audit.
The two overlap, and a good provider will say so. Technical foundations, useful content, and consistent business information serve both. The difference is in what gets measured and where the work happens. A large share of what AI platforms cite is not on the business's own website at all.
What a legitimate GEO provider should measure
At a minimum, for each question tested on each platform:
- Presence. Was the business named at all?
- Recommendation. Was it offered as an option, and in what position?
- Accuracy. Were the facts about it correct?
- Citation and source context. Which sources were shown with the answer?
- Competitive share. Which competitors were named instead?
- Commercial outcome. Did visits, calls, leads, or bookings change?
These are the six levels of the Avante AI Visibility Measurement Framework. A provider does not have to use our names for them. It should be measuring most of them.
What evidence a buyer should request
- Prompt-level reporting. One row per question, per platform, per date, showing who was named. A summary percentage is not a substitute.
- The query selection method. How the questions were chosen, how many include your business name, and whether the list is fixed.
- The responses themselves. The text of the AI answers, or screenshots, with dates.
- Citation and source analysis. The cited sources, grouped by type, with third-party sources included.
- Competitor benchmarking. Named competitors, compared on the same questions.
- A sample report. A full example of the deliverable before you pay. Avante publishes a complete sample GEO audit for a fictional firm.
Which AI platforms should a GEO agency test?
Short answerThe platforms your customers use. For most businesses that means ChatGPT, Google's AI answers, Gemini, Perplexity, and Claude. Results should be reported for each platform separately, because a business can be visible on one and absent on another.
Ask which platforms are covered and which are not. Also ask how they were tested. A test through a developer API can return different answers from the consumer app your customers use, and a careful provider will tell you which one it used. Avante states this as a limitation of its own published research.
Ten questions to ask a GEO agency
Ask them in order. Each one comes with what a good answer sounds like, so you can judge the reply and not only the confidence behind it.
- 1
Can you show me the actual AI responses, not only a score?
A good answer: The full text of the answers, for each question and platform, is in the report or available on request.
Why it matters: A score with no answers behind it cannot be checked.
- 2
Which exact prompts did you test, on which platforms, and on what dates?
A good answer: Every finding carries its question, platform, and date next to it, not in an appendix.
Why it matters: An AI answer cannot be recovered later. Without the record, next month's result has nothing to be compared with.
- 3
How did you choose the questions?
A good answer: A documented query panel built from the questions customers ask, mostly without your business name, fixed before testing began.
Why it matters: Questions that include your name make almost any business look visible.
- 4
How many times was each question run?
A good answer: More than once, with the result reported as a rate or a range.
Why it matters: AI answers vary between runs. A single test can show you present or absent by chance.
- 5
Which sources were cited, and what did you learn from them?
A good answer: A list of cited sources grouped by type, including third-party sources such as directories, review platforms, publishers, and community threads.
Why it matters: Sources are the most direct evidence of where an answer came from.
- 6
Which competitors were named, and how do I compare with them?
A good answer: Named competitors, how often each appeared, and a side-by-side benchmark.
Why it matters: The useful finding is who appears in your place, and what they have that you do not.
- 7
Do you use data from the platforms themselves, or only prompt tests?
A good answer: Both. Prompt tests, plus first-party reports from Bing Webmaster Tools and Google Search Console where your site has them.
Why it matters: A prompt test is a simulation of a customer. Platform reports show what the platform recorded.
- 8
What in your report is observed, what is inferred, and what can you not see?
A good answer: Observations and interpretations are labeled separately, and the report names the platforms and data it does not cover.
Why it matters: Any report that implies it covers all of AI is overselling.
- 9
How will you re-measure, and how will you tell a real change from noise?
A good answer: The same panel is run again on a set date, and movement is reported only when it is larger than normal variation between runs.
Why it matters: Without re-measurement on the same questions, there is no way to know whether the work did anything.
- 10
What exactly do you guarantee?
A good answer: The work and the measurement: a baseline, a plan, a re-measure. Not a ranking and not a recommendation.
Why it matters: Nobody outside the AI companies controls what their systems say.
Red flags
These describe practices, not companies. Any one of them is a reason to ask more questions before signing.
| Red flag | Why it is a problem |
|---|---|
| Guaranteed ChatGPT rankings | AI answers have no fixed ranking to guarantee, and no agency controls what a platform says. |
| A proprietary score with no prompt-level evidence | If you cannot see the questions and answers behind the number, the number cannot be verified or repeated. |
| Treating llms.txt or schema as the entire strategy | Both are tasks that can be worth doing. Neither one is a plan, and neither has been shown to cause recommendations. |
| No defined query panel | Without a fixed list of questions there is no baseline, only a series of unrelated spot checks. |
| No competitor benchmark | Visibility is relative. A report about your business alone leaves out who is being named in your place. |
| No re-measurement | A one-time report with no second measurement cannot show whether anything changed. |
| No third-party source analysis | A provider that looks only at your website is looking at about half of what gets cited. |
Can a GEO agency guarantee ChatGPT recommendations?
Short answerNo. No agency can honestly guarantee that ChatGPT or any other AI platform will recommend a specific business. A provider can guarantee its own work: documented testing, evidence, prioritized recommendations, and a repeat measurement.
Realistic expectations
- The first deliverable is a baseline. It tells you where you stand. It does not change anything by itself.
- Recommendations should be specific and ranked. Each one should name the change, who makes it, and how success will be judged. A list of general tips is not an implementation plan.
- Some of the work is off your website. Directory profiles, review platforms, and earned coverage take longer and are partly outside your control.
- Results are not on a schedule. No platform publishes a timeline, and a provider that promises one is guessing.
- Small movements are often noise. Expect the provider to tell you when a change is too small to call.
- GEO may not be your first priority. If you need leads this month or your tracking is unreliable, start there. See GEO vs. SEO vs. Google Ads: where to invest first.
How Avante answers these questions
We wrote this guide expecting it to be used on us. Our answers are public: the deliverables are listed on the page describing what an Avante GEO audit includes, along with what it cannot guarantee. The method is documented in how AI visibility is measured. Our own research, with its prompts, dates, and limitations, is Avante's AI Visibility Index. If you are weighing doing the work yourself, see DIY GEO audit vs. hiring an agency.
Published September 27, 2026 by Avante Visibility. Written by JoLyn Laney, President and Founder. Avante sells GEO audits, which you should weigh when reading a buyer guide written by a seller.
See What an Avante GEO Audit Includes
Twelve deliverables, the full test path from customer question to recommended action, and a plain statement of what an audit cannot guarantee.
Questions? Email jolyn@avantevisibility.com or call (702) 350-2525.