GEO agency · Published method · Public pricing
The generative engine optimization agency that publishes its measurement method
Botanik is a generative engine optimization agency: it measures where you stand in AI answers, on up to 5,000 demand-weighted prompts, and moves you from Absent to Recommended. The method is versioned, published and free to reuse under CC BY 4.0, so you can read exactly how you will be judged before you spend a euro.
Absent, Mentioned, Recommended: the three states of AI visibility
A brand is Absent on a given prompt when it is cited in fewer than 5% of that prompt's runs: for a buyer asking that question, the brand does not exist.
A brand is Mentioned when it appears in at least 5% of a prompt's runs without ever being the first recommendation: visible, but never the name the engine leads with.
A brand is Recommended when it is the first recommendation in at least 25% of a prompt's runs: the state every Botanik engagement is built to reach, engine by engine.
Every Botanik deliverable reads in three states. Absent: your brand appears in fewer than 5% of a prompt's runs. Mentioned: it appears in at least 5% of runs without leading. Recommended: it is the engine's first recommendation in at least 25% of runs. You understand a report in 30 seconds.
What is generative engine optimization?
Generative engine optimization, GEO, is the work of making your brand the answer AI engines give when buyers ask about your category. The shift is measurable: AI queries grew 494% between 2023 and 2026, and 48% of French people now use generative AI. Botanik does this work with a published, versioned method.
Generative engine optimization is the discipline of measuring and improving how often AI engines such as ChatGPT, Perplexity and Google AI cite or recommend a brand when users ask commercial questions in natural language.
Ranking on Google no longer guarantees being cited by AI. In 2026, 38% of top-10 Google sources are cited in AI Overviews, down from 76% in 2025. The two systems are drifting apart, which is why Botanik measures them separately: a full SEO audit ships inside every Oracle audit.
GEO does not replace search either. Google still records 84 billion visits per month, versus 5.5 billion for ChatGPT. The point is not to abandon one channel for the other: it is to be present, and measurable, on both.
Méthode Oracle v2.3: how Botanik measures AI visibility
Botanik measures AI visibility with Méthode Oracle v2.3, a versioned specification updated in July 2026 and reusable under CC BY 4.0. The spec fixes the corpus, up to 5,000 demand-weighted prompts, the engines, the 5 runs per prompt, the scoring formula, and the rule that separates signal from noise.
The corpus is built per market: commercial queries extracted from demand data, search and conversational, reformulated as natural prompts, then weighted by real demand volume. A heavily asked prompt weighs more than a niche one. Four intents are covered: discovery ("how to choose..."), comparison ("which is the best..."), local ("near me") and validation ("reviews of...").
The spec names three engines, queried with the same prompts, under the same conditions, via API, and measured separately: ChatGPT, Perplexity and Google AI. Each prompt runs 5 times per engine. Botanik reports the median score with a stability band, the interquartile spread, so you can tell movement from noise: a variation from one edition to the next is significant only if it exceeds the sum of the two bands. Below that, it is noise, and the reports say so.
Scoring uses Reciprocal Rank Fusion with k = 60: score(brand) = Σ 1 / (k + rank). An absence, treated as infinite rank, contributes zero. The formula rewards consistency: rank 1 contributes 0.0164, rank 4 contributes 0.0156, an absent brand contributes 0. One isolated first place is not enough to dominate.
The spec also lists what the method does not do, and those limits appear in every report: 5 runs reduce noise, they do not cancel it; an engine can cite an older state of your market, so every collection is dated; French prompts dominate the corpus, and multilingual coverage is partial; engines change without notice, so comparisons only hold at constant methodology, hence the versioning.
The method runs in the open. The protocol of the Baromètre GEO France, Botanik's public benchmark, was frozen on 1 August 2026 at 08:03 UTC, before the first API call, and fingerprinted with SHA-256; its 1,105 API calls are all timestamped and kept. That edition queried ChatGPT, ChatGPT Search and Google AI. Perplexity was not queried, its API key was not active during collection, and no value was extrapolated in its place: exclusions are written down, not papered over.
The method is authored by Franck Neuenschwander, co-founder, fourteen years on the advertiser side. Cite it as: "Méthode Oracle v2.3", Botanik, July 2026, botanik.ai/methode-oracle. Free reuse with attribution (CC BY 4.0). Ask any agency you evaluate for the same document.
Generative engine optimization services in 4 steps: Measure, Structure, Place, Prove
Every Botanik engagement follows the same 4-step sequence: measure where you stand, structure what engines read, place your brand in the sources they cite, and prove movement engine by engine. First citations move in 6 to 10 weeks; a durable shift from Mentioned to Recommended plays out over 4 to 6 months.
- 0101 · MeasureThe Oracle audit measures your visibility on up to 5,000 prompts, engine by engine, and sets your baseline: a starting point measured by a third-party tool and signed before any action. Internal rule: no engagement starts without a signed baseline.Deliverable : Oracle audit scored Absent / Mentioned / Recommended per engine, plus a signed baseline. Delivered in 3 weeks.
- 0202 · StructureContent designed to be cited, plus the removal of three classic technical barriers: heavy JavaScript, a missing llms.txt file, incomplete Schema.org markup. Only 3.2% of sites have deployed an llms.txt file.Deliverable : Published content and structured data, with the three technical barriers lifted.
- 0303 · PlaceEngines cite sources, so Botanik places your brand in them: comparison pages, press coverage including regional dailies, authority directories, Wikipedia, Reddit, LinkedIn.Deliverable : Placement under way in the sources engines cite.
- 0404 · ProveEach month, Botanik replays the same prompt panel, up to 5,000, and records your state per engine and per query against the signed baseline, with a third-party tool doing the measuring.Deliverable : A monthly per-engine report you read in 30 seconds: Absent, Mentioned or Recommended.
How much does a generative engine optimization agency cost?
Between €1,780 and €8,000 per month: that is the GEO specialist band on the 2026 market, per Botanik's published buying guide. Tracking tools alone start at €65 per month; enterprise programs reach €15,000 per month. Botanik sits inside the specialist band, on a public grid built from a single €890 day rate.
| Provider type | 2026 price range | What you get | Who does the work |
|---|---|---|---|
| Tracking tool alone | From €65/month | Monitoring dashboards, no implementation | You |
| Generalist agency with a GEO option | €890 to €1,780/month | Often a single engine tracked | A project lead, frequently a junior |
| GEO specialist | €1,780 to €8,000/month | Multi-engine measurement, implementation, placement in cited sources, per-engine reporting | A dedicated senior consultant |
| Enterprise and multi-market | €8,000 to €15,000/month | Several markets and languages, BI integration, governance | A multidisciplinary team |
| Botanik | Graine (the seed) from €2,670 one-off · Pousse (the sprout) from €1,780/month · Canopée (the canopy) from €2,670/month | Published method, executed work, 15 to 25% of fees indexed on results | 100% senior, never subcontracted |
Botanik pricing: one day rate, three plans
From 3 days of audit at the €890 day rate. Your visibility measured on up to 5,000 AI prompts from your market, an Absent / Mentioned / Recommended score engine by engine, a full SEO audit, and a prioritized roadmap. Delivered in 3 weeks. Implementation is not included. Continue with Pousse or Canopée within 60 days and the audit amount is deducted from your first month of support.
From 2 days per month, with a 6-month commitment. Continuous AI monitoring, implementation by the consultants themselves, content designed to be cited, placement in the sources engines cite, and monthly reporting that tracks you from Absent to Recommended. The 6-month commitment is not a sales tactic: engines, classic and AI alike, take 3 to 4 months to re-crawl, re-cite and re-rank. Below 6 months, you pay to sow without ever harvesting. On Botanik's grid, Pousse carries the Recommended badge.
From 3 days per month. Everything in Pousse, across several markets. Local GEO with geogrid measurement zone by zone, 62 cities covered with a dedicated page per market, PR and citations, and a dedicated engagement director.
Continuous Oracle monitoring. 100% senior staffing, no juniors: your account is never handed to a junior or subcontracted. Execution rather than recommendations. A published methodology. Reporting you read in 30 seconds. Full reversibility: your content, your access, your data.
Botanik prices its work in days of support: a single day rate of €890 excluding VAT, whatever your volume, sold as multi-day packages calibrated after the audit. Scaling up intensity never raises the day rate. Three plans: Graine from €2,670 one-off, Pousse from €1,780 per month, Canopée from €2,670 per month.
A guarantee with written conditions, and fees indexed on results
Botanik guarantees +30% local visibility on your zone, or the variable share of its fees is not due. Between 15 and 25% of Botanik's fees are indexed on a result indicator written into the contract. These are two distinct mechanisms, and each only binds because its conditions are written down.
The local guarantee has three written conditions: the indicator is defined contractually at the start, the measurement is done by a third-party tool, not by Botanik, and the measurement window is 1 month. If all three are met and the threshold is not reached, the variable share is waived.
The general variable share is a separate mechanism with its own clock: 15 to 25% of fees, an indicator defined in black and white in the contract, measurement by a third-party tool plus a baseline signed at the start, and a quarterly calculation window from the 4th month.
Result measurement is always the same: a baseline signed at the start, a third-party tool doing the measuring, the same one used in the audit, and monthly reporting on your contractual indicator.
"We raise the probability that your business gets chosen every time someone searches your category. And we agree to be paid, in part, based on whether that is true." Franck Neuenschwander, co-founder.
What we won't do
Hidden prices serve the seller, never the buyer. Botanik's grid is public and the same for everyone: €890 per day, Graine from €2,670 one-off, Pousse from €1,780 per month, Canopée from €2,670 per month.
A "+30% guaranteed" with no contractual indicator, no signed baseline and no measurement window is a sales promise, not a guarantee. Botanik's local guarantee carries all three conditions in the contract, and no engagement starts without a signed baseline.
Your account is never handed to a junior or subcontracted. 100% senior, no juniors, at every level of engagement.
Botanik does not deliver recommendations: it executes them. A dashboard without execution means paying a subscription at consultant price, and watching the problem instead of fixing it.
Full reversibility: your content, your access, your data, everything belongs to you. If you leave, they follow you, without penalty and without a closed tool.
A number without a method is an opinion: every figure Botanik publishes carries its measurement tool and its period, and illustrative lines are labeled as examples, in plain sight. The results register has a "Partial" status, published at the same rank as "Achieved".
Refusals are easier to verify than promises, so Botanik writes its own down. No hidden pricing, no guarantee without written conditions, no juniors on your account, no dashboard sold as a deliverable, no lock-in, no numbers without a method. Delivery is 100% senior and never subcontracted.
What do I have in hand at 30, 60 and 90 days?
30 days is the deadline by which a GEO agency owes you a deliverable, not a discovery phase. Botanik's calendar is public: a scored audit and a signed baseline at day 30, published assets and active placement at day 60, per-engine reporting and a continue, adjust or stop decision at day 90.
- 01Day 30Your score per engine, Absent, Mentioned or Recommended, a baseline signed and measured by a third-party tool, and a prioritized roadmap.Deliverable : Scored audit + signed baseline + prioritized roadmap.
- 02Day 60Content and structured data published, placement under way in the sources engines cite, and a first dated competitive comparison.Deliverable : Published assets + a dated competitor snapshot.
- 03Day 90Per-engine reporting against the baseline, your evolution measured against 3 competitors, and a decision on the table: continue, adjust, or stop.Deliverable : Reporting versus baseline + a documented continue / adjust / stop decision.
Six criteria to vet any GEO agency in one hour
Six criteria separate a serious generative engine optimization agency from a rebranded retainer, and you can check all six in one hour, without a sales call. They come from Botanik's published buying guide, updated 31 July 2026, and they apply to Botanik as much as to anyone else.
| Criterion | The question to ask | Périmètre réaliste | Horizon de résultat |
|---|---|---|---|
| A published, versioned method | Where is your measurement methodology, and which version are you using? | ||
| An explicit engine scope | Which engines do you track, and which are excluded? | ||
| Dated deliverables | What do I have in hand at 30, 60 and 90 days? | ||
| Third-party measurement and a signed baseline | Who measures the result, and what is the recorded starting point? | ||
| Named people | Who will work on my account, and is it the person I'm talking to? | ||
| Reversibility | If I stop in six months, what do I keep? |
Four red flags that disqualify a GEO provider
The clearest disqualification signal of all. If the measurement cannot be read, versioned and challenged, every number that follows is unverifiable by design.
"+30% guaranteed" with no contractual indicator, no signed baseline and no measurement window is a commercial promise, not a commitment.
The first signal of rent-seeking. A provider that hides its grid keeps the information advantage on its side of the table.
You end up paying a subscription at the price of a consultant. Access to a tool is not implementation, and implementation is what moves you from Absent to Recommended.
Four signals disqualify a GEO provider, and each one suffices on its own to remove a name from your shortlist: no published methodology, a guarantee without written conditions, prices only on quote, and a dashboard in place of a deliverable. Walk away on any one of the four.
What is included in the Oracle audit?
From €2,670, the Oracle audit includes your visibility measured on up to 5,000 real AI prompts from your market, an Absent / Mentioned / Recommended score engine by engine, a full SEO audit, and a prioritized roadmap. It starts at 3 days of audit work and is delivered in 3 weeks. If you continue with Pousse or Canopée within 60 days, the audit amount is deducted from your first month of support.
How long does generative engine optimization take to work?
First citations move in 6 to 10 weeks. A durable shift from Mentioned to Recommended plays out over 4 to 6 months, because engines, classic and AI alike, take 3 to 4 months to re-crawl, re-cite and re-rank. That is the whole reason Pousse carries a 6-month commitment: below 6 months, you pay to sow without ever harvesting.
Why hire an agency instead of a €65/month tracking tool?
Take the €65/month tool if you have a team ready to act on what it shows. If not, you are paying to watch the problem. A tool measures; it does not produce content designed to be cited, place your brand in the sources engines cite, or tie 15 to 25% of its fees to your result.
Does GEO replace SEO?
No. Google still records 84 billion visits per month, versus 5.5 billion for ChatGPT. But ranking no longer guarantees citation: 38% of top-10 Google sources are cited in AI Overviews in 2026, down from 76% in 2025. The engines differ from each other too: ChatGPT leans on its corpus and Bing search, Google AI, built on Gemini, on the Google index and its SEO signals. Botanik measures both worlds: every Oracle audit includes a full SEO audit.
How does the +30% local visibility guarantee work?
Three conditions are written into the contract: the indicator is defined at the start, the measurement is done by a third-party tool, not by Botanik, and the measurement window is 1 month. If all three are met and the threshold is not reached, the variable share of the fees, 15 to 25%, is not due.
Can I reuse the Oracle methodology?
Yes, freely, under CC BY 4.0. Cite it as: "Méthode Oracle v2.3", Botanik, July 2026, botanik.ai/methode-oracle. Reuse is free with attribution. The version matters: engines change without notice, so comparisons only hold at constant methodology.
How does the free scan work?
In 60 seconds, no account, no card: the scan tells you where you stand, with a verdict per engine, Absent, Mentioned or Recommended. Pick your sector, run it, and read your states. It is the fastest way to know whether you have a GEO problem before spending anything.
Which businesses should invest in GEO first?
Botanik's priority GEO targets are consulting, IT services firms (ESN), business services, healthcare, industry, and premium retail: categories where buyers ask AI engines for shortlists. With 48% of French people using generative AI in 2026, that list of categories keeps widening.
Thirty minutes with a consultant
Not a salesperson. We tell you which lever fits your case, and if none does, we say that too.