A GEO agency (Generative Engine Optimization) makes sure a brand is cited and recommended by artificial intelligence, in Google’s AI Overviews and AI Mode, ChatGPT, Gemini, Claude, Perplexity, Copilot… and, tomorrow, by the AI agents that will book and buy on people’s behalf.

GEO is still a young discipline, and no organisation can yet produce an official ranking of it. Some agencies nevertheless already show real know-how and an advanced understanding of how AI chooses and cites its sources. It is this visible, verifiable know-how that this ranking sets out to measure.

Best GEO agencies in France, top 10 reviewed (October 2026)

Who are the 10 best GEO agencies in France?

  1. Eskimoz – A large “Global Search” agency, an LLM Ranking tool and a GEO offer built into every lever
  2. NOIISE – The CURL GEO framework, the Acoustiik tool and the award for best GEO campaign 2026
  3. Archipel Marketing – A GEO specialist, very advanced in putting the GAIA method into practice
  4. Uplix – A search and data agency, a GEO Visibility Index and an R&D unit
  5. Peak Ace – An international group, sector LLM benchmarks and pages designed for the models
  6. Jalousie Agency – A GEO pure player, the Score de Jalousie™, the Cewe case study and the NLWeb protocol
  7. MagicGEO – A GEO pure player for SMEs, an LLM tracking dashboard and an agent-ready website
  8. Webconversion – A B2B agency in Lyon, a GEO offer launched as early as 2025 and more than 80 GEO audits
  9. Promoovoir – A Paris SEO and GEO agency, client cases ranked first on ChatGPT and very rich markup
  10. datashake – A digital acceleration group, in-house GEO tracking across more than eight LLMs
  • 58 agencies cited by AI
  • 58 checkpoints per agency

A GEO agency is not judged like a GEO expert. We measure its ability to act on every lever at once (press, reviews, comparison sites, advertising in AI, website) and how ready it is for AI agents.

We cross-reference mentions (the agencies that ten AI tools name), citability (the websites they cite as sources) and 58 checkpoints verified for each agency. We keep those that come up most often and best apply to their own website the good practices of the GAIA method, our working method for SEO in the age of AI.

First publication of the ranking: October 2026

Main changes in this update:

– selection of the 10 agencies by how often they are cited in AI answers
– a 10-pillar scoring grid including applying the GAIA method (20 points)
– technical audit of the agencies’ websites (llms.txt, AI crawlers, structured data, Lighthouse “Agentic browsing”)

Update of 28 October 2026. The grid goes from 55 to 58 checkpoints per agency. These are the three criteria of the Stability and durability of results pillar, added on 23 October. The scale and the order of the top 10 do not change.

Update of 23 October 2026. The scoring grid moves to 11 pillars. I am adding stability and durability of results (14 points), and applying the GAIA method drops from 20 to 6 points, because it is my own method. The order of the top 10 does not change.

Update of 16 October 2026. New visuals added to illustrate our tests (PageSpeed Insights), along with independent tools made available by experts to assess visibility in AI, such as Julien Gourdon’s barometer.

A GEO agency at the centre of a sonar. Gentle waves spread out from it and light up, as they pass, the sources that AI systems read, Google AI Mode, Gemini, ChatGPT, reviews, the press and comparison articles. At the bottom, four dated checks stay ticked.

Try it yourself

“Who are the 10 best GEO agencies in France?”

Answers vary from one AI to another and from one day to the next, which is precisely what this ranking measures. For Gemini, DeepSeek and Kimi, the question is copied, so you just need to paste it.

Question copied. Paste it into the window that opens (Ctrl+V, or long press on mobile)

What is the best GEO agency in France?

Methodology & scoring grid: how we assess GEO agencies in 2026

Each check follows a fixed protocol. The same question is put to ten AI tools (ChatGPT, Gemini, Claude, Perplexity, Mistral, Grok, DeepSeek, Copilot, Google’s AI Mode…), several days apart. An agency named only once is not kept. In the first checks, 58 agencies were named, and only Eskimoz, NOIISE and Archipel Marketing were named by every AI tool.

The difference between a mention and a citation matters. Being named in an answer proves awareness. Seeing your own website cited as a source proves authority, because the AI then relies on your content to build its answer.

The checkpoints are not based on the agencies’ own claims. We test their website with the tools we use for our clients (Lighthouse, Rich Results Test, reading the llms.txt, the robots.txt and the structured data), then read their offers, client cases and publications. The score out of 100 is then split across the 11 pillars shown here. Three are specific to this ranking, multichannel capacity, stability of results over time and applying the GAIA method to their own website. GAIA is my own method, which is why it counts for only 6 points out of 100.

How to read it: a generative answer is never the same twice, so we keep recurrence and never the best attempt. Measurements are redone at each update, and a website that changes is reassessed at the next check.

Objective

Measuring an agency’s real place in the answers of conversational assistants when they are asked for the best GEO agencies in France.

What is assessed

  • Recurrence in the answers of ten AI tools asked the same question
  • Consistency from one model to another
  • Position in the generated lists (top, middle, mere mention)
  • Spontaneous citations as a benchmark agency

Method & tools

The same query put to ten AI tools, with dated and archived checks. It is also the entry filter, since only the agencies that come up most often are scored.

Why this criterion matters

A GEO agency that AI does not cite when someone looks for a GEO agency has a credibility problem, since that is its own business.

Objective

Checking that an agency stays cited over time, not just on the day of a check. Appearing in an AI answer is one thing, staying there is another. I compare my dated checks over several months. I can see who stays cited, who appears and who disappears.

What is assessed

  • Consistency of citations in conversational AI from one dated check to the next (6 pts)
  • Remaining in Google’s AI Overviews and AI Mode between checks (4 pts)
  • Holding Google rankings and lasting presence in third-party rankings (4 pts)

Method & tools

Comparison of dated, archived checks, with the same question put to the same AI tools. An agency that entered recently, with fewer than three checks, receives a provisional score on this pillar.

Why this criterion matters

In GEO as in SEO, the hardest part is not appearing, it is staying. A presence that holds from one check to the next says more about an agency’s strength than an isolated citation.

Objective

Measuring the agency’s presence in Google’s generative surfaces, which are distinct from conversational assistants.

What is assessed

  • Presence in the AI Mode answer
  • Mention among the “pioneering” or benchmark agencies
  • The agency’s website cited as a source

Method & tools

AI Mode checks on the same query, repeated several times.

Why this criterion matters

Since 22 July 2026, AI Overviews and AI Mode have been live in France. They are now the first answer surface for a large share of searches.

Objective

Measuring what sets an agency apart from a consultant, namely the capacity to act at the same time on every lever that feeds AI.

What is assessed

  • Press relations and digital PR
  • Content production at scale
  • Technical SEO
  • Classic advertising and advertising in AI (GEA)
  • Reviews, online reputation, social media, CRO

Method & tools

Reading the offers published on each agency’s website.

Why this criterion matters

In GEO, everything counts. A model forms its opinion from the press, reviews, forums, comparison sites and the website itself. An agency able to deploy these levers together gets results that a lone practitioner reaches more slowly.

Objective

Assessing the agency’s ability to measure what it sells.

What is assessed

  • Proprietary citation tracking tool
  • Share of voice in model answers
  • Reporting on traffic coming from AI

Method & tools

Tools named and documented on the agency’s website. Third-party tools count for less than an in-house tool.

Why this criterion matters

There is no Search Console for generative engines yet. Without measurement, GEO remains a promise.

Objective

Checking GEO results that are published, named and quantified.

What is assessed

  • Named GEO client cases
  • Quantified results (citations, share of voice, AI traffic, leads)
  • Consistency of figures from one page to another

Method & tools

Reading the client case pages. An unsourced or contradictory figure counts for less.

Why this criterion matters

The GEO market is young and talkative. Only documented cases separate practice from talk.

Objective

Assessing the depth of the method and the contribution to shared knowledge.

What is assessed

  • A named and described GEO method or framework
  • Guides, white papers, barometers, studies
  • Regularity and dates of publications

Method & tools

Reading the method pages and publications dated 2026.

Why this criterion matters

A published method can be checked, discussed and reproduced. It is also what feeds AI.

Objective

Checking, on the agency’s website, a few public technical checks taken from the GAIA method (Generative, Authority, Information, Agents). GAIA is my own method and I publish this ranking. That is why this pillar counts for only 6 points, and only keeps checks that anyone can redo.

What is assessed

  • G (1 pt): AI crawlers allowed in the robots.txt file
  • A (1 pt): organisation and managers described in the home page JSON-LD
  • I (1 pt): services and publications described in the home page JSON-LD
  • Agents (3 pts): a valid llms.txt file (1 pt) and the Lighthouse “Agentic browsing” test (2 pts)

Method & tools

Automated check of the public files (llms.txt, robots.txt), reading of the home page JSON-LD and a Lighthouse 13.5 “Agentic browsing” audit. Anyone can redo each of these checks.

Why this criterion matters

Soon, agents will talk to agents. A customer’s agent will query a business’s website directly to compare, book or buy. An agency that has not prepared its own website for this shift will not be able to prepare its clients’ websites.

Objective

Measuring the technical quality of the agency’s showcase website.

What is assessed

  • Lighthouse mobile performance score
  • Accessibility
  • Layout stability

Method & tools

Lighthouse, mobile version, home page, at each update of the ranking.

Why this criterion matters

An agent or an AI crawler first reads a page that loads quickly and cleanly. Performance remains the foundation of citability.

Objective

Taking account of the brand’s historical weight in search.

What is assessed

  • Stated age and size
  • Verifiable awards and distinctions
  • Presence in third-party rankings

Method & tools

Data published by the agencies and identifiable third-party sources.

Why this criterion matters

Accumulated authority remains a signal that models pick up.

Objective

Rewarding commercial honesty.

What is assessed

  • Public prices
  • Self-ranking declared or not

Method & tools

An agency that publishes a ranking in which it puts itself first loses a point, even when it is open about it.

Why this criterion matters

AI reads rankings. An undeclared self-ranking distorts what it learns.

The GAIA method, a working method for tomorrow’s SEO

SEO has always had its reference methods. The semantic cocoon, formalised by Laurent Bourrelly, changed the way a website is organised around a topic. The GAIA method answers the question every business faces today: how do you stay visible when answers are generated, summarised and soon carried out by AI?

It is a way of working on your search visibility, built on four pillars, that any business can apply to its own website.

  • G – Generative: writing your pages so that AI understands and cites them, with clear answers, frequently asked questions and paragraphs that stand on their own
  • A – Authority: giving AI confidence by showing who is speaking, for how long and with what proof, and by being cited, and well regarded, elsewhere (press, reviews, forums, partners)
  • I – Information: publishing what nobody else can write for you, your figures, your client cases, your experience in the field
  • A – Agents: opening your website to AI crawlers and making it usable by the assistants that will, tomorrow, search, compare, book and buy for their users

Soon, agents will talk to agents

Take an example. Julie asks her AI assistant, “find me a plumber available on Thursday morning near Versailles, for less than €120”. The assistant queries the websites of plumbers in the area. The one whose prices, service area and time slots can be read by a machine receives the request. Its own agent confirms Thursday at 9 am, and the appointment goes into Julie’s calendar. The other websites were not even read.

To be chosen in this exchange, a website must be structured, accessible to machines and reliable. It is the newest pillar of GAIA, and the one that will make the difference in the years ahead.

A temple whose pediment reads “Visible in AI”, held up by four pillars standing on your website. G for Generative with an answer bubble, A for Authority with a medal, I for Information with a chart, A for Agents with a robot.

In this ranking, we simply look at whether each agency applies these principles to its own website. An agency that masters tomorrow’s SEO starts by practising it at home.

What is the GAIA method?

GAIA is a working method formalised by Nicolas Peter (Oli-via-net) for search in the age of AI. It rests on four pillars: Generative (being understood and cited by generative engines), Authority (building a recognised brand and author entity), Information (producing what a model cannot invent) and Agents (preparing your website for autonomous AI agents).

Is the GAIA method used to score the agencies?

No. GAIA is a method for working on your search visibility, applicable to any website, as the semantic cocoon once was. This ranking only looks at whether the agencies apply some of these good practices to their own website, just as you would check that a performance specialist has a fast website themselves. Since GAIA is my own method, this check counts for only 6 points out of 100.

Why prepare a website for AI agents now?

Because agents will talk to agents. A customer’s personal agent will query a business’s website, or its own agent, to find out a price or availability, then book or buy. Whatever is not marked up, accessible and up to date does not exist for it.

Why is Oli-via-net not in this top 10?

Because it publishes this ranking and created the GAIA method. Scoring itself would have no value. Its own measurements are published further down, so that anyone can check them.

Editorial transparency and independence of the ranking:

This ranking is published by Nicolas Peter, SEO and GEO consultant, founder of the Oli-via-net agency. It involves no affiliation, no link exchange and no payment, and Oli-via-net has excluded itself so as not to be both judge and party. It is updated continuously as new checks come in. Any agency that has changed its website, or wishes to add a clarification, is welcome to write to us.

Top 10 GEO agencies in 2026: profiles and expertise

French GEO brings together very different profiles, from large search agencies that have added a GEO offer to a full set-up, to pure players born with generative engines, and B2B or local agencies that made the shift early. The profiles that follow set out their strengths, criterion by criterion.

This ranking evolves with the market. Any agency that has changed its website, or wishes to add a clarification, can write to us at any time.

1. Eskimoz

Website: https://www.eskimoz.fr/

Eskimoz is the agency most present in citations in conversational AI. All five AI tools asked name it, and it also features among the benchmark players in Google’s AI Overviews and AI Mode. Its multichannel delivery capacity is unmatched in this ranking, covering press relations, content, technical SEO, classic advertising and advertising in AI, social ads, CRO and even Reddit, with offices in Boulogne-Billancourt, Lyon, Bordeaux, Lille, London, Milan, Madrid and Düsseldorf.

For tools and measurement of AI visibility, the agency has developed its own suite, including LLM Ranking, which measures a brand’s presence in ChatGPT. Its published GEO method and publications are the most extensive in the panel, with a five-step method, GEO barometers, LLM traffic barometers and a study carried out with Ipsos. Among its documented GEO results and client cases, the website highlights Fraser Yachts, at +485% GEO traffic. Its SEO authority and reputation are those of a European search leader.

On applying the GAIA method, Eskimoz is very strong on Information and Authority, and has published an ebook on agentic commerce. Its own website remains less agent-oriented than its message. The performance of the agency’s website and pricing transparency are the least emphasised areas.


2. NOIISE

Website: https://www.noiise.com/

NOIISE is named by all five AI tools asked, which puts it at the top of citations in conversational AI, and it features among the benchmark agencies in Google’s AI Overviews and AI Mode. Present in seven cities with 125 consultants claimed, it won the award for best GEO campaign at the SEMY Awards 2026, a strong marker of SEO authority and reputation.

Its published GEO method, the CURL framework (Coherent, Unique, Regular, Legitimate), comes with publications such as the white paper “Du SEO au GEO” (From SEO to GEO). For tools and measurement of AI visibility, the agency has developed Acoustiik, which tracks citations in ChatGPT, Gemini, Claude and Perplexity. Its multichannel delivery capacity is broad, including media relations and influence, and its documented GEO results and client cases include EFLC, at +200% mentions on ChatGPT and Gemini. Its Head of SEO, David Groult, features in our ranking of the best GEO experts.

Applying the GAIA method is under way on its website, which publishes an llms.txt, and the Agents side is set to grow stronger there. The performance of the agency’s website is good. Pricing transparency, as at most large agencies, goes through a quote.


3. Archipel Marketing

Website: https://www.archipelmarketing.com/

Archipel Marketing, based in Paris and Lyon, is one of the three agencies named by every AI tool asked. It is at the very top for citations in conversational AI and also appears in Google’s AI Overviews and AI Mode.

It is one of the agencies most advanced in applying the GAIA method. Its website publishes a very complete llms.txt and its long version, explicitly allows the main AI crawlers, marks up its services, offers and FAQ, and passes Lighthouse’s agentic browsing test. The performance of the agency’s website is exemplary. Its published GEO method runs in four phases (measurement, construction, amplification, leadership), and its publications include dozens of GEO guides. On transparency, its prices are public, from €990 excl. VAT per month.

With a team of around ten people, the agency focuses its multichannel delivery capacity on content, cited sources and advertising, and relies on third-party tools (Profound, Peec AI) for tools and measurement of AI visibility. Its documented GEO results and client cases are based on its own tests and on more than 75 clients in France, Europe and North America, including Mirakl, Keyrus, PlayPlay and Cenareo. The agency also publishes the Observatoire de la visibilité IA (AI Visibility Observatory), which analyses more than 407,000 AI answers. Its SEO authority and reputation, more recent, are growing fast.


4. Uplix

Website: https://www.uplix.fr/

Uplix, a Young Innovative Company (Jeune Entreprise Innovante) based in Paris, is cited by four AI tools out of five, an excellent level of citations in conversational AI, and appears in Google’s AI Overviews and AI Mode. Its published GEO method follows six steps, from entity analysis to reporting, and its publications on the Uplix Labs blog are regular (Perplexity, Schema.org markup for AI).

Its tools and measurement of AI visibility rely on proprietary tools, the GEO Visibility Index and LLM share-of-voice tracking, developed by an R&D unit already awarded for its Predictive Ranking solution, a sign of SEO authority and reputation. Its multichannel delivery capacity covers SEO, advertising on the major platforms, CRO and data. Its documented GEO results and client cases include La Maison Convertible, at +100% traffic from ChatGPT.

Applying the GAIA method starts on its website with a robots.txt that declares the AI crawlers, and the Agents side is the natural next step. The performance of the agency’s website is solid, and transparency comes through a reference article on SEO prices.


5. Peak Ace

Website: https://peakace.fr/

Peak Ace, formerly Search Foresight, based in Levallois-Perret and Nantes within an international group headquartered in Berlin, is cited by four AI tools out of five in citations in conversational AI and appears in Google’s AI Overviews and AI Mode. Its SEO authority and reputation are those of an established European group.

It is one of the agencies furthest ahead in applying the GAIA method. Its website publishes a structured llms.txt, an “LLM Endpoints” manifest and a version of each page designed for the models, and passes Lighthouse’s agentic browsing test. Its tools and measurement of AI visibility rely on in-house tools (LLM Traffic Dashboard, AI Visibility Benchmark), and its publications include AI visibility benchmarks by sector. Its multichannel delivery capacity is complete, with digital PR, advertising, social, CRO, AI-assisted content and SEO GEO media sales.

Its published results and client cases mainly concern SEO (Naturactive, Drawer, Maître CoQ), with documented GEO client cases still to come. The performance of the agency’s website on mobile and pricing transparency are less emphasised.


6. Jalousie Agency

Website: https://www.jalousie.agency/

Jalousie Agency, based in Paris, comes up regularly in citations in conversational AI and appears in Google’s AI Overviews and AI Mode. Its documented GEO results and client cases are among the most detailed in the ranking, with Cewe at +340% traffic from AI in five months and share of voice multiplied by four, alongside brands such as Yves Rocher, Canva and Century 21.

Its published GEO method, the Méthode Jalousie™, combines prompt tracking, content designed for the models, comparison pages and press relations, a sign of real multichannel delivery capacity, advertising in ChatGPT included. For tools and measurement of AI visibility, the agency created the Score de Jalousie™. It is also very advanced in applying the GAIA method. Its website passes Lighthouse’s agentic browsing test, and it is the only agency in the panel to cite an agentic protocol in its offer, Microsoft’s NLWeb.

Its publications explore original ideas, such as “GPT-Squatting”. Transparency about the team and prices, the performance of the agency’s website and its SEO authority and reputation are the least emphasised areas.


7. MagicGEO

Website: https://www.magicgeo.fr/

MagicGEO, founded in mid-2025 by Michaël Bitton and Hugo Arias-Benamou, presents itself as the first AI search agency for SMEs, with around fifteen people in Boulogne-Billancourt and Cannes. It comes up regularly in citations in conversational AI and in Google’s AI Overviews and AI Mode.

It is one of the agencies most advanced in applying the GAIA method, with a structured llms.txt, declared AI crawlers, marked-up services and FAQ, and a website very well prepared for agentic browsing. Its tools and measurement of AI visibility rely on an LLM tracking dashboard and a GEO audit tool, which allow it to publish documented GEO results and client cases prompt by prompt (Sphère RH, Adsrank, SommeSi). Its publications include a GEO guide for very small businesses and SMEs.

A young agency, it focuses its multichannel delivery capacity on content and authority. Its SEO authority and reputation are being built, and the performance of the agency’s website and pricing transparency are less emphasised.


8. Webconversion

Website: https://www.webconversion.fr/

Webconversion, a “100% B2B” agency in Lyon, is cited by four AI tools out of five in citations in conversational AI, and Google’s AI Mode names it among the pioneering agencies in its AI Overviews and AI Mode. It structured its GEO offer at the very start of 2025 and claims more than 80 GEO audits carried out.

Its publications are regular and useful, with a guide to getting cited by Claude, analyses of AI Mode and AI Overviews, and GEO applied to B2B lead generation. Its multichannel delivery capacity brings together SEO, advertising, multilingual content, marketing automation and ABM. Applying the GAIA method is under way on its website, which publishes an llms.txt.

For tools and measurement of AI visibility, the agency relies on a third-party tool (Otterly), and its documented GEO results and client cases name clients (TDF, Adyen, Odigo) without published GEO figures. Its SEO authority and reputation rest on its B2B specialisation. The performance of the agency’s website and pricing transparency are less emphasised.


9. Promoovoir

Website: https://www.promoovoir.com/

Promoovoir, founded by Tanguy Richard in Paris, is cited by four AI tools out of five in citations in conversational AI and appears in Google’s AI Overviews and AI Mode. Its documented GEO results and client cases are concrete, with AssurLuxe and Arthur Serrurerie, the latter at +120% qualified calls and first on ChatGPT, Gemini and Perplexity according to the website.

Applying the GAIA method is very advanced. Its website marks up many types of structured data (services, offers, reviews, FAQ) and passes Lighthouse’s agentic browsing test. On transparency, the agency publishes a starting price, from €1,900, and its publications follow AI news closely.

A small agency, it focuses its multichannel delivery capacity on SEO, content and its own editorial network, with no named tool for tools and measurement of AI visibility. The performance of the agency’s website is fair and its SEO authority and reputation are growing.


10. datashake

Website: https://www.datashake.fr/

datashake, a group based in Paris and New York with more than a hundred staff, is cited by four AI tools out of five in citations in conversational AI and appears in Google’s AI Overviews and AI Mode. Its SEO authority and reputation are those of a group managing substantial media budgets.

Its multichannel delivery capacity is strong, with the SEO/GEO offer built into advertising, social ads, CRM and an AI creative studio. For tools and measurement of AI visibility, the group has developed an in-house tool tracking rankings in more than eight LLMs. The performance of the agency’s website is good, and applying the GAIA method starts with a robots.txt that declares the main AI crawlers.

Its documented GEO results and client cases, its published GEO method and publications remain hard to see publicly, as does pricing transparency.


Right of reply

Any company or person mentioned in this ranking can ask for a piece of information to be corrected, or dispute their assessment. Write to us through the contact form, stating the page concerned and the points to check. We reply within 7 days, and every correction is dated in the ranking update.

The detailed scores of each player are not published. They are based on checks, AI answers, websites and reviews, which change from one week to the next. The scoring grid itself is published higher up on this page.

I could obviously have included Oli-via-net in this ranking. I publish it, I master all its tools, and in my view the agency is among the best GEO agencies in France. I did not do so, for the sake of transparency. My background also prepared me for it early on. Long before the word GEO existed, I was carrying out machine learning research with 7Days, a company approved as a Young Innovative Company (Jeune Entreprise Innovante) that I ran from 2019 to 2024, and the 7Days story tells that journey. That is what makes Oli-via-net one of the pioneering GEO agencies.

Outside this top 10, several agencies show solid signals and could join it soon, such as Natural-Net, a Bordeaux agency that provides GEO training and holds the AI Ambassador label, Agence Onze, 100% search, with its own GEO Tool, Primelis, whose Outrank suite measures visibility on search engines and LLMs, Yumens, which combines GEO and advertising in AI, and SEO Monkey, which publishes guide after guide on ranking in ChatGPT.

Finally, other players are emerging as checks go on and already come up in AI answers, such as Triaina, Junto, Growth Room, Empirik, Yateo, Keyweo, Genn and Ad’s up Consulting.

You can help this ranking evolve!

An independent, evolving approach:

This ranking does not claim to state an absolute truth. It offers a reasoned, dated reading of the French GEO landscape in 2026. A lack of public evidence says nothing about an agency’s real worth. It only limits what we can measure, and therefore score.

The scoring grid will follow the generative engines and the agents themselves. As agentic protocols spread, a website’s ability to be used by agents will count for more, and I will assess it with checks that anyone can redo. Agencies that prepare for it now will get ahead.

Scores and tools can differ depending on:

  • the URL tested (home page vs sub-page),
  • the date and time,
  • the network profile (throttling),
  • the cache/CDN state,
  • the third-party resources loaded at the time of the test

Does a check seem debatable, or do you think a GEO agency deserves a place here? Write to us through our contact form. Reasoned requests are reviewed at each update.

Two web pages whose highlighted passages fly off in streams of light towards an artificial intelligence answer, which cites them as numbered sources.

Oli-via-net, a pioneering GEO agency: our own measurements

AI answers change fast. An agency cited today may no longer be cited next week. That is why we do not publish the details of the checks made on the ranked agencies.

We nevertheless apply the same tests to them as to our own website and the websites we manage, whether for speed and accessibility (PageSpeed Insights), structured data or citations by AI. The screenshots shown here give some examples.

These checks also show where we speak from. In Julien Gourdon’s barometer, an independent tool that analyses hundreds of ChatGPT answers, a page on oli-via-net.fr is both the most retrieved and the most cited source on the question of the best SEO consultant.

Everything you need to choose the right GEO agency and succeed in AI visibility

Visibility has moved. Between a Google ranking, a citation in a generated answer, verifiable authority and measurable proof, it is fair to ask what a GEO agency really does, and how to recognise a serious one.

This FAQ answers the questions businesses facing GEO actually ask, how to find out whether an AI cites your website, why a good Google ranking is no longer enough, which signals a model uses to choose its source, and how to tell a serious practitioner from marketing talk. For classic SEO, see the best SEO consultant.

The answers below are based on our citation checks, on the public documentation of generative engines and on the signals visible in server logs. They are dated and reviewed at each update of the ranking.

Choosing the right SEO expert

Answer: GEO refers to the set of techniques that aim to get a brand, a website or a person cited within the answers generated by ChatGPT, Google AI Overviews, Perplexity, Gemini or Copilot.

Where SEO seeks a position in a list of links, GEO seeks a mention in a summary text. The unit of measurement is no longer the ranking, it is the citation.

Answer: three names for neighbouring scopes. AEO (Answer Engine Optimization) is the oldest, aimed at direct answers, rich snippets and voice assistants. GEO specifically targets generative engines. LLMO refers to work on how an entity is represented within the models themselves, beyond search.

In practice, in 2026, GEO encompasses the other two. It is the term the market has adopted.

Answer: citability is the probability that a model reuses your content and attributes the mention to you. It is measured through checks. You query generative engines on a defined set of queries and note who is cited, in what order and with which link.

It is a matter of repeated, dated observation, not a score that a tool would provide. A serious GEO expert will show you their checks.

Answer: three complementary sources. First your server logs, where the OAI-SearchBot, ChatGPT-User, PerplexityBot and Perplexity-User user agents mark visits triggered by an answer being written.

Then referral traffic (chatgpt.com, perplexity.ai in your statistics). And finally manual checks, the only way to know which queries you appear for, and against whom.

Answer: because the two systems do not select on the same criteria. Ranking rewards a page’s overall relevance, whereas the generative summary extracts a self-contained passage that is factual and attributable.

A page that answers well but spreads its answer over three screens will be read and then ignored. A paragraph that answers in two sentences will be picked up.

Answer: it is a trade-off, not a given. Distinguish training crawlers (GPTBot, Google-Extended, CCBot) from real-time search crawlers (OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, ClaudeBot).

Blocking the former protects your content from training. Blocking the latter, on the other hand, simply means giving up any citation, since the engine can no longer read the page it was about to mention.

SEO methods, key concepts and leading figures

Answer: it depends on how it accesses content. In real-time search (ChatGPT with browsing, Perplexity, AI Overviews), properly indexed content can be cited within a few days.

In the model’s internal knowledge, you have to wait for an update, which means several months. That is why a GEO strategy works on indexing and structure first, the levers that act quickly.

Answer: no, it sits on top of it. Generative engines rely heavily on the search index, so being invisible in Google remains the surest way to be absent from its AI Overviews.

GEO adds a layer, structure, attribution and entity consistency, on top of SEO foundations that remain necessary. We cover classic SEO in our ranking of the best SEO consultants.

Answer: the data that describes an entity and its proof, such as Organization and Person linked by sameAs, knowsAbout and hasCredential, an explicit author and publisher, citation pointing to sources, and speakable on the passages meant to be extracted.

Conversely, FAQPage markup that does not match the visible text does more harm than good, since Google requires markup to reflect content that is actually displayed.

Answer: a model does not reason in keywords but in entities. Anchoring means making yours unambiguous, with the same name everywhere, sameAs links to your public profiles, and consistency between your website, LinkedIn, directories and press mentions.

Without this anchoring, the model hesitates between several possible candidates, and cites someone else.

Answer: answer first, context afterwards. A heading that asks the question, a first paragraph that answers it in two or three self-contained sentences, then the detail.

Add short definitions, labelled tables, and dated, sourced figures. The test is simple. Taken out of its page, an extractable passage must remain true and understandable.

Answer: they count, but differently. What carries weight in a generated answer are mentions, being named in a text the model has read, more than the link itself.

A citation without a link in a reference article can carry more weight than a link in a directory. Link building becomes a question of presence in the corpus, not of volume.

Technical side & the future of SEO

Answer: with three distinct indicators, the citation rate (the share of target queries where you are mentioned), share of voice against the competitors cited, and referral traffic from AI interfaces.

Search Console impressions do not measure this. A page shown in a summary without a click remains an impression with no usable signal.

Answer: it is the signature of zero-click. Your page is shown, often in or below a generated summary, but the user gets their answer without clicking.

The reflex is to suspect a penalty, but it is a change of format. The right response is not to retreat but to aim for the citation, which remains visible even without a click.

Answer: not necessarily to write code, but to read it. Reading logs to spot AI crawlers, reading JSON-LD to check an entity graph, reading a server-side render to know what a crawler actually sees.

A practitioner who has never opened a log file is working blind on half the subject.

Answer: ask for checks. A serious practitioner shows dated screenshots of AI answers, before and after their work, on precise, verifiable queries.

An opportunist will talk about “optimised prompts” and “AI-friendly content” without ever producing an observation. The field is too young for seniority to be an argument. Only proof counts.

Answer: neither penalised nor favoured as such. What gets set aside is content with no contribution of its own, which rephrases what the model already knows.

A text that is useful to an AI contains what it cannot produce on its own, such as field data, measurements, dates, real cases and screenshots. The rest is noise it already has.

Answer: the tooling is still young. You combine citation tracking platforms, log analysis to isolate AI crawlers, a structured data validator and, above all, regular manual checks on a stable panel of queries.

No tool today replaces direct observation of the answers. That is what separates a measurement from an estimate.

Answer: no, and it is a common mistake. Multiplying similar pages creates cannibalisation. Engines hesitate between content that is too alike, then keep none of it.

Better one deep reference page, with clearly titled self-contained sections, than a constellation of thin pages competing with each other.

Answer: yes, and often faster. When a user asks an AI for a local recommendation, the answer relies on the Google Business Profile, reviews, consistent contact details and local mentions.

Entity anchoring plays a full part here. It is GEO applied to local businesses, with measurable effects within a few weeks.