The vocabulary of AI visibility
AEO vs GEO vs SEO vs LLMO: the taxonomy, settled
AEO, GEO, SEO and LLMO are not four disciplines: they are SEO, the incumbent, plus three names competing to describe the same new work, visibility in AI-generated answers. Botanik uses GEO, measures it engine by engine with a published method, and argues below why AEO and LLMO are synonyms you can retire.
What do AEO, GEO, SEO and LLMO stand for?
SEO, search engine optimization, is the discipline of earning visibility in classic search results, primarily Google, by making pages crawlable, authoritative and relevant so they rank when someone types a query.
GEO, generative engine optimization, is the discipline of increasing the probability that AI engines such as ChatGPT, Perplexity and Google AI cite or recommend your brand when they generate an answer.
AEO, answer engine optimization, is the practice of structuring content so answer surfaces, from featured snippets to AI assistants, can lift a direct answer straight from your page. In practice it now overlaps almost entirely with GEO.
LLMO, large language model optimization, is a narrower label for the same work as GEO, focused on how language models themselves represent your brand rather than on the full engine, model plus retrieval plus ranking.
AEO, GEO, SEO and LLMO each get a one-sentence definition below, written to survive on its own outside this page. Botanik uses these terms daily in audits and reporting, and treats only two of them, SEO and GEO, as vocabulary a buyer actually needs in 2026.
Why GEO is the term that matters
Botanik's position is deliberate: GEO, generative engine optimization, is the useful term, and AEO and LLMO are niche synonyms for the same work. The reason is not fashion but measurement: engines behave as distinct systems, and in Botanik's Baromètre GEO France, the 3 engines tested do not name the same number one in 2 markets out of 4.
A name should point at the thing you can win or lose. When someone asks an AI which bank, software or agency to choose, the recommendation is produced by an engine: a language model, plus retrieval, plus ranking, working as one system. LLMO names only the model. AEO names only the output format, the direct answer. GEO names the engine itself, and the engine is where your brand ends up Absent, Mentioned or Recommended.
The engine, not the model, is the right unit, and the Baromètre GEO France, T3 2026 edition (third quarter), shows it plainly. Its three engines were ChatGPT, queried without web search, ChatGPT Search, queried with it, and Google AI. In the invoicing-software market for small businesses, ChatGPT's number one is Indy, ChatGPT Search's is Sage, and Google AI's is Tiime. Three engines, three different leaders. A model-only label like LLMO has no vocabulary for that spread; an engine-level one names it.
AEO earned its keep in the featured-snippet era: structure the page, mark it up, win the answer box. Those habits still matter, but they are now inputs to a larger system. ChatGPT leans on its own corpus and on Bing search; Gemini leans on the Google index and its SEO signals. Optimizing answers in the abstract, without naming the engine and the sources it reads, is how you end up with work that cannot be measured.
The practical consequence is a buying rule, not a debate about words. Whatever a vendor calls the service, GEO, AEO, LLMO or anything else, ask the question from Botanik's buyer's guide: which engines do you track, and which are excluded? A serious answer is a named list with a collection frequency. A vague one tells you the acronym was the product.
Which term should you use?
Botanik's verdict on the vocabulary is short: use SEO for classic search, use GEO for AI-generated answers, and retire AEO and LLMO as labels while keeping the useful habits behind them. The table settles which term belongs in your briefs, your contracts and your reporting.
| Term | What it names | Botanik's verdict | Horizon de résultat |
|---|---|---|---|
| SEO | Visibility in classic search results, Google first | Keep. Still the volume channel, and the index Gemini reads. | |
| GEO | Your presence in AI-generated answers, measured engine by engine | Use. The working term for the discipline. | |
| AEO | Structuring content so answer surfaces can lift a direct answer | Retire the label, keep the habits. Absorbed by GEO. | |
| LLMO | Influencing how language models represent your brand | Retire. Names one component of the engine, not the system you compete in. |
Is GEO replacing SEO?
No. Google still records 84 billion visits a month against 5.5 billion for ChatGPT, so classic search remains the volume channel by far. GEO does not replace SEO: it extends the same discipline to a new surface, the AI-generated answer, where rankings alone no longer guarantee you are cited.
The overlap between classic rankings and AI citations is collapsing. 38% of the sources in Google's top 10 are cited in AI Overviews, down from 76% in 2025. Ranking on page one no longer guarantees that a generative engine cites you, and that gap is exactly the work GEO covers. SEO gets you into the index; GEO gets you into the answer.
The AI channel is small next to Google and growing fast, and both facts matter. AI queries grew by 494% between 2023 and 2026, and 48% of French people use a generative AI in 2026. Meanwhile the plumbing is young: 3.2% of sites have deployed llms.txt. Early, growing, under-equipped: that is what an opening looks like, not a funeral for SEO.
The two disciplines also feed each other. ChatGPT draws heavily on its own corpus and on Bing search; Gemini draws on the Google index and its SEO signals. Clean technical SEO is not optional homework: it is GEO infrastructure. That is why Botanik's Graine, the seed, pairs a full SEO audit with the AI visibility measurement in a single diagnostic, from €2,670 one-shot.
How do you measure GEO?
Botanik measures GEO with the Méthode Oracle v2.3, updated in July 2026 and reusable under a CC BY 4.0 license. The principle: replay a corpus of up to 5,000 prompts, weighted by real demand, on each engine separately, then score every brand into one of three states, Absent, Mentioned or Recommended.
- 01Build the corpusCommercial queries are extracted from demand data, search plus conversational, rewritten as natural prompts, then weighted by real demand volume: a heavily asked prompt weighs more than a niche one. The corpus covers four intents: discovery, comparison, local and validation.Deliverable : A weighted panel of up to 5,000 prompts, four intents covered.
- 02Query the engines identicallyThe Méthode Oracle names its engines: ChatGPT, Perplexity and Google AI. They are queried identically, same prompts, same conditions, via API, and each engine is measured separately. Every prompt runs 5 times per engine, and the median is reported with a ± stability band, so noise does not pass for movement.Deliverable : 5 runs per prompt per engine, median plus stability band.
- 03Score into three statesRankings are aggregated with Reciprocal Rank Fusion, k = 60: the score is the sum of 1 / (k + rank), an absence contributes zero, and a lone first place is not enough to dominate. Thresholds: Absent, cited in under 5% of a prompt's runs; Mentioned, at least 5% of runs without being the first recommendation; Recommended, the first recommendation in at least 25% of runs.Deliverable : One state per brand, per engine: Absent, Mentioned or Recommended.
How much does GEO cost?
The market for GEO services in 2026 runs from €65 a month for a tracking tool alone to €15,000 a month for enterprise programs. Botanik publishes its own grid: Graine, the seed, an audit from €2,670 one-shot; Pousse, the sprout, from €1,780/month; Canopée, the canopy, from €2,670/month. All prices exclude VAT.
| Option | Market price in 2026 | Who does the work | Horizon de résultat |
|---|---|---|---|
| Tracking tool alone | From €65/month | You. The tool watches; someone still has to act. | |
| Generalist agency, GEO as an option | €890 to €1,780/month | A project manager, often junior; often a single engine tracked. | |
| GEO specialist | €1,780 to €8,000/month | A dedicated senior consultant; multi-engine measurement, implementation, reporting per engine. | |
| Enterprise, multi-market | €8,000 to €15,000/month | A multidisciplinary team; several markets and languages, BI integration, governance. | |
| Botanik, for reference | Day rate €890, the same at any volume; Graine from €2,670 one-shot, Pousse from €1,780/month, Canopée from €2,670/month | Senior consultants only, zero juniors, and they execute rather than hand you recommendations. |
What GEO won't do
Google still records 84 billion visits a month, against 5.5 billion for ChatGPT. GEO extends your visibility work to a new surface; it does not retire the old one.
First citations move in 6 to 10 weeks, and a durable move from Mentioned to Recommended plays out over 4 to 6 months. Engines take 3 to 4 months to re-crawl, re-cite and re-rank; no acronym changes that.
5 runs per prompt reduce noise: they do not cancel it. An engine can cite an old state of your market, and engines evolve without notice, which is why Botanik dates every collection and versions the method.
A measurement covers the prompts in the panel, nothing outside it, and a live answer also depends on the account, the history and the country of whoever asks. Anyone selling a total view of AI is selling certainty the data cannot support.
Botanik publishes the limits of GEO with the same care as the promises, because negative claims are the ones buyers can actually verify. The limits below come straight from the published Méthode Oracle and from the Baromètre GEO France, T3 2026 edition, the first edition published.
Is AEO the same as GEO?
In practice, yes: AEO and GEO describe the same work, earning citations and recommendations inside AI-generated answers. GEO has become the clearer name because it points at the system you actually compete in, the generative engine. Keep the habits AEO taught, structured pages and direct answers, and retire the acronym.
What is the difference between GEO and AEO and LLMO?
GEO names the whole system you optimize for, the generative engine; AEO names an output format, the direct answer; LLMO names one component, the language model. Botanik uses GEO because the engine is the unit you can measure: one state per brand, per engine, per prompt, Absent, Mentioned or Recommended.
What does LLMO stand for?
LLMO stands for large language model optimization, and it is the same discipline as GEO seen from the model side. The label undersells retrieval: in the Baromètre GEO France, ChatGPT queried without web search and ChatGPT Search queried with it crown different leaders in the same market, Indy and Sage. The engine, not the model, decides.
What is the difference between GEO and SEO?
SEO earns rankings in classic search results; GEO earns citations and recommendations inside AI-generated answers. The overlap is shrinking: 38% of Google top-10 sources are cited in AI Overviews, against 76% in 2025, so a page-one ranking no longer guarantees a citation.
How long does GEO take to work?
First citations move in 6 to 10 weeks, and a durable move from Mentioned to Recommended plays out over 4 to 6 months. Engines, classic and AI alike, take 3 to 4 months to re-crawl, re-cite and re-rank, which is why Botanik's Pousse, the sprout, runs on a 6-month engagement.
How do I check if AI engines recommend my brand?
Botanik's free scan gives a verdict in 60 seconds, no account, no card: for each engine, your brand comes back Absent, Mentioned or Recommended. It runs on the Méthode Oracle, published and versioned, so you can read exactly how the verdict is produced before you trust it.
Thirty minutes with a consultant
Not a salesperson. We tell you which lever fits your case, and if none does, we say that too.