What a GEO agency does
When a buyer asks a chat assistant which provider to hire, which software to compare or who to trust with a project, they get a written answer that names only a handful of companies. A GEO agency works to make sure yours is one of those names, that it’s described accurately and that your pages serve as a source for the answer. It’s the same goal as SEO, moved to a new part of search.
- 1Get namedBe one of the few brands mentioned in the answerVisibility
- 2Get described accuratelyMake sure the answer gets your offer rightAccuracy
- 3Serve as a sourceHave your pages read to build the answerCitation
Our role starts with observation: which questions in your market trigger a generated answer, which brands appear in it, which sites are read to build it. From that we draw up a plan, then we act where the model’s decision is made: your pages, their structure, the data that describes your business, and the third-party sites that talk about you. Tracking then happens question by question, every month, with the same collection rules.
Looking for a detailed definition of the discipline, its origins and the basic best practices? Our article on generative engine optimization walks through them step by step, and our AI glossary explains the technical terms (language model, passage retrieval, knowledge cutoff). This page describes the service we offer: what it covers, how it works, what you receive and how we measure its impact.
One point is worth making up front: GEO doesn’t replace healthy SEO, it builds on it. A page that search engines don’t crawl or consider weak has little chance of being picked up by an assistant that relies on it to answer. That’s why every engagement starts by checking the foundations, and why it ties in, when needed, with the work of our SEO agency.
GEO, SEO, AEO: what sets AI search optimization apart
The three acronyms often get used as if they were interchangeable. Yet they point to different targets, and that difference defines what an engagement covers. SEO aims for a position in a list of results, AEO (answer engine optimization) aims for the short answers a search engine displays directly, and GEO aims for the answers written by generative AI, which synthesize several sources and sometimes recommend brands. You’ll also hear LLMO, or large language model optimization, which means the same thing as GEO.
| Criterion | SEO | AEO | GEO |
|---|---|---|---|
| What you’re trying to get | A position in the results and a click | To be the answer displayed for a specific question | To be mentioned, recommended or used as a source in a written answer |
| Where it happens | Google and Bing results pages | Featured snippets, voice assistants, answer boxes | ChatGPT, Gemini, Perplexity, Claude, Copilot, Google’s AI Overviews and AI Mode |
| Unit of work | The query and the page | The question and the paragraph that answers it | The prompt, the entity (your brand) and the passage reused |
| Signals that matter | Relevance, technical quality, links | Clarity and brevity of the answer | SEO foundations, precise passages, brand consistency across the web, mentions by third-party sources |
| What gets measured | Rankings, clicks, conversions | Presence in the answer box | Mentions, citations, share of voice, sentiment, visits from assistants |
The table also shows what GEO has in common with traditional SEO: technical quality and page relevance remain the price of entry. What changes is the scale of analysis. A generative engine doesn’t always read your whole page: it extracts a passage, compares it to your competitors’ passages and decides whether to use it. It doesn’t judge you in isolation either: it cross-checks what your site claims against what other sources say. A GEO engagement therefore works on two fronts at once, your content and your reputation beyond your own site.
The market includes agencies that have simply rebranded their SEO offering. The difference shows in the method: a set of prompts defined before any action, repeated tracking, work on entities and third-party sources, and reports that talk about citations rather than rankings.
Does your brand show up in AI answers?
That’s the first question of any GEO engagement. We ask the assistants the questions your customers ask, then record who gets mentioned, in what terms, and which pages serve as sources for the answer.
A single answer proves nothing: each question is asked again at regular intervals, from the target market, and only what keeps coming back counts. This tracking sets the baseline, then measures the effect of each action.
Assistants we query
- ChatGPT
- Gemini
- Claude
- Perplexity
- Mistral
- DeepSeek
PerplexityExample · fictional brands
What POS system should a small retail store use?
For a neighborhood shop, Brand A and Brand B are often recommended; Your brand comes up for its mobile app.
Sources for the answer
- your-site.comquoted in the text
- example-guide.comquoted
- example-reviews.comconsulted
How AI assistants pick a brand
Understanding where an answer comes from tells you what to act on. An assistant can answer in two ways. Either it draws on what its model learned during training, a memory frozen at a given date and hard to correct quickly; or it runs a search at the moment of the question, reads a few pages and builds its answer from them. Both coexist, and the second is the one where an engagement produces results within a reasonable timeframe.
Mentioned, cited, used: three different situations
Your brand can be mentioned in the text of the answer without a link to your site, cited with a link as a source, or used when one of your pages helped write the answer without your name appearing. Each case calls for a different action: a brand that’s used but never named often lacks clear identity signals; a brand that’s named but not cited depends on third-party sites that can be identified; a brand absent from both first raises a question of access or relevance.
The platforms we track
| Platform | What feeds its answers | What we check first |
|---|---|---|
| Google’s AI Overviews and AI Mode | Google’s index | Indexing, allowed snippets, quality of the pages ranking for your questions |
| Gemini | Google’s model, backed by Google Search | The same foundations, plus how your business is described on the sources Google connects to your brand |
| ChatGPT and ChatGPT search | OpenAI’s model and pages retrieved by its search crawler | OAI-SearchBot access, clear passages, presence in the comparisons it consults |
| Perplexity | Systematic web searches, with sources displayed | Pages cited for your questions, freshness of information |
| Microsoft Copilot | Bing’s index | Indexing in Bing, citation reports in Bing Webmaster Tools |
| Claude | Anthropic’s model and, when enabled, a web search | Brand presence in the industry’s reference sources |
Not every platform carries the same weight for every business. We focus on the ones your customers actually use, depending on the market and the language, rather than spreading the effort across all of them.
Robots.txt rules are a request, not a barrier: not all AI crawlers follow them with the same rigor, and workarounds were documented in 2025. When excluding a crawler really matters to you, we back it up with a block at the server or CDN level.
The role of entities
To a model, your business is an entity: a name tied to an activity, a location, products, executives. The clearer and more consistent this information is from one source to the next (your site, your profiles, your industry’s directories, a Wikidata entry if you have one), the less likely the model is to confuse you with a namesake or describe you incorrectly. This identity work, often overlooked, explains a good share of the gap between two brands of similar size.
What the engagement covers
A GEO engagement covers six areas. Depending on your situation, some take a few days and others several months; the initial diagnosis is precisely what sets the right amount for each.
1. AI visibility diagnosis
We ask your key questions to the selected assistants, several times and in each target language, then record who is named, who is cited, in what tone and from which sources. The diagnosis compares your brand to three to five competitors you choose, checks AI crawler access to your site and spots the false or outdated information models repeat about you.
2. Prompt and entity map
The questions people ask an AI are longer and more specific than a Google query: “what invoicing tool for a small team that does business with Spain” rather than “invoicing software.” We build the list of prompts that matter for your business, sorted by buying stage, and connect it to the pages on your site that should answer them. The same map lists the attributes of your entity that need to be consistent everywhere.
3. Technical readiness
Robots.txt rules for each crawler (search on one side, training on the other, as you choose), pages that render without depending on JavaScript, tags that limit snippets, schema.org structured data in JSON-LD (Organization, Service, Product, Article, FAQPage when the page fits) that matches the visible content. If someone else pitches you an llms.txt file as a game changer, know that it’s a non-standard convention: we can create one, without making it a priority.
One limitation to know before writing these rules: at Google, the Google-Extended token only covers Gemini training. It has no effect on AI Overviews or AI Mode, which draw on the Google Search index; only snippet-limiting tags (nosnippet, max-snippet) or deindexing affect them, at the cost of losing traditional visibility. That choice is made with you; it isn’t applied by default.
4. Answer-ready content
We start by rewriting the pages that should answer your priority prompts: a clear answer in the first few lines, passages that stand on their own when extracted, comparisons in tables, dated claims backed by a source. Missing pages are then created from a detailed brief. The copy is still written for your customers; a passage that’s clear for a busy reader is clear for a model too.
5. Authority and third-party mentions
Generated answers lean heavily on what others say about you. We identify the sources assistants read for your questions (comparison sites, trade media, business directories, forums, review platforms), then work on your presence on the legitimate ones: information to correct, listings to complete, digital PR, expert commentary. Backlinks remain valuable, first of all because they help your pages get found.
6. Tracking and experimentation
Every significant change is logged with its date and the prompts it targets, then compared with the tracking data from the following weeks. This experiment log keeps you from mistaking natural variation in answers for gains from your work.
If your site itself is holding the project back (confusing structure, technology that blocks crawling), we flag it in the diagnosis; a website redesign may then come before the content work.
Process and deliverables
The timeline below is a guide. The actual duration varies with how large the site is, the languages to cover and what your teams handle themselves; the quote sets it for your case.
-
01Weeks 1–3
Scoping and diagnosis
A conversation about your offers, markets and competitors, read-only access to your measurement tools, initial tracking on the assistants, a technical check. The phase ends with delivery and presentation of the AI visibility diagnosis.
-
02Weeks 3–6
Strategy
Prompt map approved with you, choice of pages to revise or create, list of third-party sources to work on, order of the work.
-
03Weeks 6–16
Implementation
Technical fixes, rewrites of priority pages, structured data, first actions on third-party sources, at an agreed pace.
-
04Then, every month
Measurement and adjustments
New tracking on the same set of prompts, analysis of the changes, priorities for the following month.
| Deliverable | What it contains | When |
|---|---|---|
| AI visibility diagnosis | Presence of the brand and competitors by platform and language, sources read, errors found, technical findings | End of phase 1 |
| Tracked prompt set | The selected questions, sorted by buying stage and market, with the page on your site that should answer each | Phase 2, updated throughout the engagement |
| Roadmap | The workstreams ranked by expected impact and effort, with the split between your team and ours | Phase 2 |
| Delivered pages | Rewritten or new content, ready to publish, with the matching structured data | Phase 3 |
| Dashboard | Mentions, citations, share of voice and visits from assistants, connected to your own accounts | From phase 3 |
| Experiment log | Every dated action, the prompts targeted and the effect observed | Kept continuously |
On your side, plan mainly for review time: one point of contact who reads the pages, makes decisions on what information to publish and grants the necessary access. We work in French, English or Spanish, depending on your team.
Measuring AI visibility, market by market
A generated answer changes from one request to the next, and sometimes from one user to the next. A one-off check therefore proves nothing. We measure by sampling: each prompt is asked several times, at regular intervals, under the same conditions (language, country, account with no history), and the frequency of appearance is the metric, not a flattering screenshot.
The collection channel is fixed too: an answer obtained through a model’s API isn’t the same as the one in the consumer app, where web search, memory and account location change the result. For each platform, we specify whether tracking goes through the interface or the API, and we never mix the two in the same series.
The metrics we track
- Mention rate: the share of answers that name your brand across the prompt set.
- Citation rate: the share of answers that link to one of your pages, and which ones.
- Share of voice: your presence relative to the tracked competitors, platform by platform.
- Sentiment and accuracy: how the assistant describes you, and the errors it repeats.
- Visits and conversions: traffic from assistants, isolated in Google Analytics with a dedicated segment, and the leads it generates.
These measurements complement SEO metrics, covered in our article on SEO KPIs. On the search engine side, Bing Webmaster Tools publishes a report on citations from its AI features, while Google Search Console counts AI Mode and AI Overviews traffic in its Performance report without breaking it out; we bring this data into the dashboard and spell out what it covers. For automated tracking, we choose the tool based on your markets and budget: Peec AI, Otterly.AI, Profound or Semrush’s AI Visibility Toolkit, among others. Accounts are opened in your name.
Why language and country change everything
The same assistant doesn’t give the same answer in Madrid, London and New York, or in Spanish and English. The sources it reads change with the language, so do local competitors, and a brand that’s well established in one market can be absent from the next. So we build a prompt set for each market and language, written by people who use the language every day rather than translated word for word, and we analyze the results separately. If you target several countries, this work ties in with our international SEO services.
How to choose a GEO agency
The discipline is young, offerings are multiplying and what they include varies widely. Asking a few questions before you sign helps you spot the solid ones, and we’re happy to answer them ourselves.
- Which prompt set will be tracked, and who chooses it? A serious agency defines it with you before taking action, and doesn’t change it midway to make a report look better.
- How are the answers collected? Number of repetitions, platforms, languages, collection conditions: the method should be in writing.
- Who does what? The split between technical work, content, PR and tracking should be clear, as should the share that falls to your team.
- Is the data yours? Tracking tools, analytics accounts and history should stay accessible if the engagement ends.
- Does the agency also do SEO? Without solid foundations, GEO spins its wheels; conversely, an offer that only talks about Google rankings isn’t a GEO offer.
One red flag should send you running: a promise of a guaranteed spot in ChatGPT or in AI Overviews. No provider controls what a model writes. Our commitments cover what we control: the method, the deliverables and their timeline, reproducible tracking, actions that comply with search engine guidelines, and an honest report when a workstream doesn’t deliver.
How much does a GEO agency cost?
Our engagements are priced by quote, once we’ve reviewed your site and understand your goals. The amount depends on factors we can name in advance:
- the number of prompts tracked and how often they’re checked;
- the number of markets and languages, each requiring its own prompt set and its own analysis;
- the platforms selected, from two or three assistants to broader tracking;
- the state of the site: a few technical tweaks or a deeper overhaul;
- the volume of content to rewrite or create;
- the work on third-party sources, often the most time-consuming item;
- the share handled by your team, which reduces ours accordingly;
- the subscription to a tracking tool, billed by its vendor and separate from our fees.
An engagement always starts with the diagnosis, and monthly tracking follows the setup. If your site first needs a full SEO checkup, an SEO audit can be carried out at the same time, so you don’t pay twice for the same checks.
Your questions: GEO Agency
Can’t find your question? Ask us
Are GEO and AEO the same thing?
The two disciplines overlap a great deal. AEO targets the short answers a search engine displays for a specific question; GEO targets the answers written by generative AI, which combine several sources and name brands. In practice, a GEO engagement covers both, with measurement specific to assistants.
How soon can we expect results?
Technical fixes and rewritten pages can be picked up as soon as search engines crawl them again, which takes weeks. Presence on third-party sources and the way models describe your brand change more slowly, over several months. That’s why we judge the trend across several monthly tracking rounds.
Can an agency guarantee that ChatGPT will cite my brand?
No. What an answer contains depends on the model, how the question is worded and the user’s context. An agency can increase the likelihood of being named and cited, and prove it with repeated tracking, but it can’t promise a spot.
I already have an SEO agency: do I need a GEO agency?
Not necessarily a second provider. If your current agency already tracks a prompt set, works on your entities and third-party mentions and reports on citations, you’re covered. If not, we can handle the GEO side alone, in coordination with them, without redoing their work.
Is GEO useful for a small business?
Yes, as long as it’s targeted. A small company doesn’t need to track hundreds of questions: a few dozen well-chosen prompts, tied to its offers and its area, are enough to see where it’s missing. For a local business, the work often overlaps with local SEO.
Do backlinks still matter for GEO?
Yes, but not only for their own sake. A link from a reputable site helps your pages get found and ranked, and therefore read by assistants. For GEO, a mention of your brand in a comparison article or a publication that AI consults counts just as much, even without a link.
Do I need an llms.txt file?
It’s not a priority. The file is still a community-proposed convention, not recognized as a standard, and the major platforms don’t use it to choose their sources. We can set one up if you’d like, after the workstreams that carry more weight.
How much time does my team need to put in?
Mostly review and decision time: approving the prompt list, reviewing pages, confirming the public information about the company. The amount depends on how much of the implementation you keep in-house, and we spell it out in the quote.