The AI SEO agency that publishes its method, its prices, and its limits

An AI SEO agency measures how often AI engines cite your brand in their answers, then produces the content and citations that change those answers. The market moved fast: AI queries grew by 494% between 2023 and 2026, and 48% of French people use generative AI in 2026. Botanik does this work under a published, versioned method.

An AI SEO agency, also called a GEO agency (Generative Engine Optimization), is a provider that measures a brand's presence in AI-generated answers and executes the content, technical and citation work that improves that presence.

The shift is measurable. In 2025, 76% of Google's top 10 sources were cited in AI Overviews; in 2026, 38% are. Ranking well no longer guarantees being cited. At the same time, classic search has not collapsed: Google still records 84 billion visits per month, against 5.5 billion for ChatGPT. The two channels coexist, and a serious agency works both.

Botanik reports visibility in three states, engine by engine. Absent: your brand is cited in fewer than 5% of a prompt's runs. Mentioned: it appears in at least 5% of runs without being the first recommendation. Recommended: it is the first recommendation in at least 25% of the prompt's runs. The thresholds are published, so two people reading the same report reach the same conclusion.

One conviction runs through the work: visibility is cultivated, not hacked. No trick forces an AI engine to recommend you. There is a method, applied over months, and it is published.

AI SEO costs from €65/month for a self-serve tracking tool to €15,000/month for enterprise programs. A serious retainer that includes implementation sits between €1,780 and €8,000/month in 2026. Botanik publishes its full grid on a single €890 day rate, with a one-off audit from €2,670 and retainers from €1,780/month.

Botanik works on a single day rate of €890, whatever your volume, the same day rate as Junto, which distributes Botanik's services. Engagements are sold as multi-day packages calibrated after the audit, never as single days, and raising intensity never raises the day rate. All prices are excluding VAT.

The entry point is Graine, the seed: a one-off audit from €2,670. If you continue into Pousse, the sprout, or Canopée, the canopy, within 60 days, the audit fee is deducted from your first month of the retainer.

One market practice deserves naming: quote-only pricing is the first sign of rent-seeking. A serious agency publishes three things before selling anything: its method of measurement, its prices, and the indicator it agrees to be judged on. This page is Botanik doing exactly that.

Méthode Oracle v2.3, updated in July 2026 and reusable under CC BY 4.0, is Botanik's published measurement method. For an audit, it builds a corpus of up to 5,000 prompts weighted by real demand, runs each prompt 5 times per engine, and reports the median with a stability band.

The method applies to three Botanik products: the Oracle audit, the Baromètre GEO France and the free scan. All three report the same three states, Absent, Mentioned and Recommended. The corpus and scoring described below are the audit's; the Baromètre publishes its own frozen protocol on its own page.

The corpus starts from commercial queries extracted from demand data, search and conversational, reformulated as natural prompts and weighted by real demand volume: a heavily asked prompt weighs more than a niche one. Four intents structure it: discovery (how to choose), comparison (which is the best), local (near me) and validation (reviews of).

Each prompt runs 5 times per engine, and Botanik reports the median score with a stability band, the interquartile range, because a single AI answer is a sample, not a truth. A variation between two editions is significant only when it exceeds the sum of the two bands. Below that, it is noise, and the reports say so in those words.

Aggregation uses Reciprocal Rank Fusion with k = 60: score(brand) = Σ 1 / (k + rank). An absence counts as rank infinity and contributes zero, and a lone first place is not enough to dominate. In practice, rank 1 contributes 0.0164 and rank 4 contributes 0.0156, so consistent presence beats occasional wins.

The method also publishes its 4 limits, in every report. 5 runs reduce noise, they do not cancel it. An engine can cite an old state of your market, so every collection is dated. French prompts dominate the corpus, and multilingual coverage is partial. Engines evolve without notice, so comparisons hold only at constant methodology, which is why the method is versioned.

The method is authored by Franck Neuenschwander, Botanik's founder, with fourteen years on the advertiser side. Anyone can reuse it with attribution: "Méthode Oracle v2.3", Botanik, July 2026, botanik.ai/methode-oracle, CC BY 4.0.

Baromètre GEO France, edition T3 2026 (third quarter), is Botanik's public benchmark of AI visibility in France: 1,105 API calls, 1,104 usable answers, 480 scored responses across 4 published markets, all collected on 1 August 2026 under a protocol frozen before the first call. Reuse is free under CC BY 4.0.

The protocol was frozen on 1 August 2026 at 08:03 UTC, before the first call, with a published SHA-256 fingerprint (6a5bc90fbb59…). 48 prompts, 8 per market, were locked in advance and are published word for word: this 48-prompt corpus is the Baromètre's own, separate from the audit's corpus of up to 5,000 prompts, and the Baromètre's scoring formula is printed on its page. Collection ran 1 discovery pass then 5 measurement passes in 21 minutes, across three engines: ChatGPT (gpt-5.3-chat-latest), ChatGPT Search (gpt-5-mini) and Google AI (gemini-2.5-flash).

The exclusions are named too. Perplexity was not queried: its API key was not active in the collection environment. Mistral was not queried at all. No value is extrapolated in place of an absent engine.

On 120 observations per market, BoursoBank leads French online banking at 98 ±2.0, ahead of Fortuneo at 82 ±2.1, a 15.0-point gap that is significant. HubSpot leads CRM for SMBs at 80 ±3.6, ahead of Pipedrive at 65 ±4.0. Two leads are published as not significant, in plain text: Henrri at 67 ±4.6 over Indy at 58 ±5.4, and Qare at 94 ±3.6 over Livi at 90 ±2.8.

The engines disagree more than most buyers expect: in 2 of the 4 published markets, the three engines do not name the same number 1. In invoicing software for very small businesses, ChatGPT's number 1 is Indy (81.2), ChatGPT Search's is Sage (70), and Google AI's is Tiime (81.2). Measuring one engine and calling it AI visibility is not measurement.

Two of the six collected markets failed the coherence test and were not published: real-estate estimation (13.7% average overlap, 2 isolated prompts out of 8) and CPF professional training (6.3%, 7 isolated prompts out of 8). Their 240 answers stay traced and available on request. A prompt counts as isolated when the brands it makes engines cite overlap, on average, with fewer than 10% of the brands cited by the other prompts of the same market.

On the classic SEO side, a product-page mission for Ebay delivered +95% organic traffic, +38% clicks to products and +22% average order value. The Swile case, an SEO acquisition strategy with results in 3 months, publishes no figures, and Botanik says so rather than inventing them.

You do not need an AI SEO agency when an in-house team can execute, when your horizon is shorter than 6 months, or when AI answers barely move your category yet. Botanik writes these cases down because a provider that never says no is selling to everyone.

If you have a team that can execute, take a tracking tool from €65/month. Botanik's own pricing page says it plainly: take the tool if you have a team to act on it. Otherwise, you pay to watch the problem.

If you cannot hold 6 months, do not sign a growth retainer, with Botanik or anyone else. Engines, classic and AI, take 3 to 4 months to re-crawl, re-cite and re-rank. Below 6 months, you pay to sow without ever harvesting.

If AI answers barely touch your category, start smaller. Google still records 84 billion visits per month against 5.5 billion for ChatGPT. The free scan answers the first question, whether AIs recommend you at all, in 60 seconds, without an account or a card. If nothing is moving, keep your money.

Fit matters too. Botanik's priority GEO targets are consulting, IT services companies (ESN), business services, health, industry and premium retail. If your market sits far from that list, the free 30-minute call will tell you quickly, and it costs nothing.