GEO Agency: How to Choose the Right One
How to choose a GEO agency: key criteria (multi-AI measurement, deliverables, tracking), questions to ask and red flags before you sign.
Contents
Looking for a GEO agency to get your brand cited and recommended by ChatGPT, Gemini, Claude or Perplexity? The market is young. Many SEO agencies are adding GEO (Generative Engine Optimization) to their catalog, new specialist players are appearing, and promises vary enormously from one provider to the next.
This guide is for companies that want to choose a provider with their eyes open: what a GEO agency should actually do, the criteria that set a serious method apart, the questions to ask in a first meeting and the red flags that should make you hesitate.
Understand the differences between SEO and GEO before choosing a provider
What does a GEO agency actually do?
A GEO agency works on how AI answer engines talk about your brand. The work falls into four stages:
- Measure: ask AI engines the questions your prospects really ask, and record whether your brand is cited, in which position, with what tone, and who is cited instead of you.
- Diagnose: understand why. Can AI crawlers access your site? Do your pages give clear, sourced answers? Which external sources do AI engines use to talk about your industry?
- Fix: content, structured data, technical accessibility, presence on the third-party sources AI engines consult.
- Track: repeat the measurement over time to check that the actions have an effect.
A good part of the work overlaps with SEO (technical, content, structured data, link building). What is new is the measurement: you no longer look at a position on a results page, but at your brand's presence in generated answers, which vary from one AI engine to another and from one query to another.
How to choose a GEO agency: the criteria
1. A multi-engine measurement method
This is the first criterion, and the most telling. Your prospects don't all use the same AI: ChatGPT, Gemini, Claude, Perplexity, Copilot, and now Google's AI Overviews and AI Mode, launched in France on July 22, 2026. Each engine has its own sources and its own answers (see how Claude, Gemini and Perplexity differ in what they cite).
Ask which AI engines are queried, on how many queries, and how results are aggregated. A simple average across engines gives the same weight to a heavily used engine and a marginal one. A finer method weights results by the actual usage of each AI, which the Arcom AI barometer lets you estimate for France.
2. Queries that reflect your market
The measurement is only as good as the questions asked. Testing only your brand name says little: what matters are the questions prospects ask before they know your brand ("which payroll software for a small business", "best bank for a freelancer", "how to choose an accounting firm in Lyon"). A serious agency builds the query list with you, by topic and intent, and shows it to you.
3. Concrete, verifiable deliverables
At the end of an audit, you should be able to answer these questions without the agency's help:
- On which queries is my brand cited, and on which is it missing?
- Which competitors are recommended instead of me, and on which topics?
- Which sources (sites, media, forums, comparison sites) do AI engines use to talk about my industry?
- What should I fix first, and why?
Ask for access to the raw AI answers (the verbatim answers), not just a summary score. A score without the answers behind it cannot be checked.
4. Transparency about the method and its limits
AI answers are not stable: the same question asked twice can produce two different answers, and models change regularly. A serious agency explains this, states when and how the queries were run (date, engine, web search on or off) and separates what it measured from what it assumes.
5. Examples and references
Ask for a sample report, even anonymized, and client cases in an industry similar to yours. Check whether the recommendations are specific (this page, this source, this missing piece of content) or generic ("add a FAQ", "work on your E-E-A-T").
6. Tracking over time
A one-off audit is a snapshot. To know whether actions work, you need to re-measure on the same queries, with the same method, at regular intervals. Check that the agency offers this tracking and that indicators stay comparable from one measurement to the next (see GEO KPIs).
7. Technical, editorial and external coverage
AI visibility depends on your site (crawler access, content structure, evidence, freshness) but also on what other sites say about you. An agency that only works on technical issues, or only on content, covers part of the problem. The 6 Detekia score criteria are a useful grid to check the proposed scope.
Questions to ask before signing
- Which AI engines do you query, and how do you weight results between them?
- How many queries, and how are they chosen? Can I validate the list?
- Will I get the full AI answers, not just scores?
- How do you handle answer variability (repetitions, measurement dates)?
- Do you identify the external sources AI engines use to talk about my industry?
- Can you show me a sample report and action plan?
- Who implements the recommendations: you, my teams, or both?
- How will we measure progress in six months?
Red flags
- A guaranteed position ("top recommendation in ChatGPT in 30 days"). Nobody controls an AI's answer. You can increase the likelihood of being cited, not guarantee it.
- Measurement on a single AI engine, or on a handful of queries, presented as representative of your visibility.
- An opaque score, with no access to the answers and no explanation of how it is calculated.
- Manipulation techniques: hidden text aimed at bots, fake reviews, mass promotional posting on forums. At best ineffective over time, at worst harmful to your reputation.
- A single miracle fix ("all you need is an llms.txt file"). No isolated lever is enough.
- No look at competitors. Your visibility is relative: if a competitor is better documented, it will be cited before you (see why ChatGPT cites your competitors).
SEO agency or GEO agency: do you need to switch?
Not necessarily. If your SEO agency already masters technical work, content and structured data, it covers part of the job. The question to ask is simple: how does it measure your presence in AI answers? If the answer is limited to Google traffic and rankings, the measurement is missing. You can then add a specialist AI visibility audit and leave implementation to the team that already knows your site.
What you can check yourself before contacting an agency
A few free checks will help you arrive at the meeting with an idea of where you stand:
- The free Detekia GEO score: it checks whether one page of your site is technically ready to be cited (6 criteria) and asks an OpenAI model 5 questions for a first citation test.
- A manual test: ask a few questions yourself in ChatGPT, Gemini and Perplexity.
Questions to test in ChatGPT, Gemini, Perplexity:
1. "What do you think of [your brand]?"
2. "What is the best [product/service] for [need]?"
3. "[Key question in your industry]"
4. "[Profession] in [city]" (local business)These tests give you a feel, not a measurement: a few queries on one or two engines aren't enough to draw conclusions. To go further, read our guide on how to run a GEO audit and the tools and methods for measuring AI visibility.
The Beeleven approach
Beeleven is the agency that built Detekia. Its audit applies the criteria above: strategic queries are defined with you, asked to ChatGPT, Gemini, Claude and Perplexity, and every answer is analyzed (brand mention, position, tone, competitors cited, sources used). The audit ends with a summary of priorities ranked by impact. The full action plan is part of the GEO strategy, which your teams carry out; GEO support (after a strategy, 6-month minimum) adds, every month, 2 pieces of content written, 2 pages reviewed with the list of changes to make, a measurement and a one-hour call.
Ask it the same questions as any other provider: that is the best way to compare. To learn more, see how Beeleven measures what AI engines answer in your market.
Resources to form your own opinion
- The founding academic paper: "GEO: Generative Engine Optimization" (Aggarwal et al., 2023, published at KDD 2024)
- Google's E-E-A-T guidelines (Search Quality Rater Guidelines)
- Official crawler documentation: OpenAI (GPTBot, OAI-SearchBot), Anthropic (ClaudeBot)
The complete GEO guide for 2026