Workstream · Content and semantics
SEO content: written to answer, not to fill
Topic clusters, search intents, pages that win the click and that generative engines can quote as they stand. One page per search intent, an answer in the first sentence, a named author on every piece. Written by our consultants, approved by you.
What is SEO content?
SEO content is the set of pages written for one precise search intent and structured to be found, read and quoted by traditional search engines and generative engines alike. At Botanik, every published piece answers in its first sentence and supports a commercial page named in advance.
It is neither volume nor keyword stuffing. A useful page answers the question asked in its first lines, then adds what makes that answer verifiable: dated figures, named sources, an identifiable author. That is also the form a generative engine can quote without having to rewrite it.
The bar moved in one year. In 2026, 38% of the sources in Google's top 10 are cited in AI Overviews, down from 76% in 2025. Ranking well no longer guarantees being quoted. A page now has to pass two selections: the ranking, then the citation. This service treats the second as seriously as the first.
Demand explains the urgency. AI queries grew by 494% between 2023 and 2026, and 48% of French people use generative AI in 2026. Google still records 84 billion visits a month versus 5.5 billion for ChatGPT, so traditional search is not going away. Generative answers stack on top of it. One page that serves both is the point of this service.
The four signals that bring companies to us
Several of your pages target the same intent. The engine alternates between them and none of them settles. The audit's merge and revision plan resolves it page by page.
Articles that point to no commercial page and receive no link from one. The traffic arrives, reads, and leaves without a next step.
Pages never revised since publication, on topics that have moved on. Their positions erode quietly until someone looks at the numbers.
Your brand appears in no generative answer on its own topics. The audit measures that state engine by engine before any writing starts.
Four signals trigger a content engagement at Botanik: cannibalization, an orphan blog, frozen content, and zero pickup in AI answers. The semantic audit starts from these four points, so you know which pages each signal touches before anything gets rewritten.
How do you structure content so generative engines can quote it?
We structure every page around four conversational intent families: discovery, comparison, local, and validation, the same four the Oracle corpus is built on. Each section answers one question in a self-contained block, so an engine can lift it without losing the subject, the figure, or the source.
Chunkability is the working principle. A generative engine rarely quotes a whole page: it retrieves passages and assembles an answer from them. So every block under a heading has to survive on its own. It names its subject instead of leaning on a pronoun, carries its own number, and dates its own source. A block that only makes sense in context gets summarized. A block that stands alone gets quoted.
The format is simple to state and hard to keep. The first sentence answers, the rest proves. Every heading asks a question a buyer actually asks. The paragraph under it stands on its own, and the figure sits early in the sentence rather than buried at the end. None of this is decoration: it is what retrieval selects for.
Conversational intents change what a page is for. Someone typing a keyword and someone asking an assistant a full question are not doing the same thing. We write for the question: discovery ("how to choose"), comparison ("which one is better"), local ("near me") and validation ("reviews of"), because those are the four families the Oracle corpus weights by real demand. The method behind that corpus is published and versioned on /en/oracle-method.
Structure also fails for technical reasons. Three barriers block generative pickup on otherwise sound pages: heavy JavaScript, a missing llms.txt file, and incomplete Schema.org markup. Only 3.2% of sites have deployed llms.txt. The audit flags all three and routes them to the technical workstream, so editorial work is never undone by a page an engine cannot read.
What we do, concretely
Clusters and internal linking: every published piece supports a commercial page named in advance. The blog stops being an island.
We start from the question asked, not the raw keyword. Two close queries do not always call for the same page, and forcing them onto one creates the cannibalization the audit later finds.
The first sentence answers, the rest proves. That is what makes a passage quotable as it stands, in a featured snippet or in a generated answer.
Named author, cited sources, verifiable experience. The standard engines demand on sensitive topics, applied to every page we publish.
Pages losing positions are revised when the tracking flags them, before they drop. No page waits for an annual content review to get fixed.
Titles and meta descriptions are tested on groups of pages large enough to be conclusive, against a control group. Opinions do not settle editorial debates; the test does.
Six workstreams make up the SEO content service: semantic architecture, intent mapping, answer-first formatting, applied E-E-A-T, tracking-driven refresh, and editorial testing. Each piece supports a commercial page named in advance, and every workstream is executed by our consultants rather than delivered as a recommendations document.
How we work
Three phases structure every content engagement: map, write, then link and track. The map decides which page answers which intent before anything is written. Writing follows the answer-first format. Linking and tracking make sure no piece is published in isolation or left to drift.
- 01MapBefore anything is written, we build the intent map and decide which page answers which query. Queries are grouped by intent, not by volume. Existing pages are classified: keep, merge, or rewrite. Every planned piece gets a commercial target named in advance, so no article is published without a page to support.Deliverable : An intent map with a keep, merge or rewrite decision for every existing page, and a named commercial target for every planned piece.
- 02WriteTexts are written and reviewed by writers, with an identifiable author in the byline. The answer sits at the top of the page and the proof follows. Sources are cited and dated in the body of the text. Subject-matter review happens before publication, never after.Deliverable : Published pages signed by a named author, with dated sources in the body.
- 03Link and trackAn isolated piece serves no one: internal linking and tracking are part of the delivery, not an option. Internal links go live at publication, not later. Positions and clicks are tracked page by page. Revisions are planned for the pages that slip, before they drop out of view.Deliverable : Internal links live on day one, page-level tracking, and a revision queue driven by the data.
What an audit finding looks like
The semantic section of the Graine report (Graine is our diagnostic tier) delivers findings in a fixed format: a figure, a severity, a fix, a priority, an effort estimate. The two rows below illustrate the report template. They are examples, not client data: yours come out of your own audit.
| Finding | Severity | Measured value | Fix |
|---|---|---|---|
| Pages in cannibalization | High | 34% of audited pages | Merge the duplicates, set canonicals, redirect internal links to the master page |
| Articles with no target page | Moderate | 2 out of 3 articles | Commercial target assigned to every article, contextual links placed in the body |
Editorial work we have already delivered
Botanik restructured the navigation and the category pages to maximize product visibility. Results: +120% organic sessions, +39% product pages viewed, +25% in sales from SEO.
A premium netlinking strategy combined with expert content creation to strengthen Mirakl's positioning. The case study is published without figures.
Two published case studies involve editorial and structural content work: Wobz, an e-commerce catalog restructuring that produced +120% organic sessions, and Mirakl, an authority program built on expert content and premium links. Both are documented as case studies on this site.
How long before engines quote the new content?
First citations move in 6 to 10 weeks once the structured content is live. A lasting move from Mentioned to Recommended plays out over 4 to 6 months. Those are the timelines published across the site, and the content workstream works to the same clock.
Engines do not read your pages the same way. ChatGPT leans heavily on its own corpus and on Bing search, Gemini on the Google index and its SEO signals. A page with its answer up top and its sources visible serves both without being rewritten for either. That is the point of writing it that way once.
Measurement is continuous, not a one-off screenshot. Each month during an ongoing plan, we replay the same panel of prompts, up to 5,000, and record your state per engine and per query: Absent, Mentioned or Recommended. The baseline is signed at the start and a third-party tool does the measuring. How those states are defined, scored and versioned is documented on /en/oracle-method. What the tracking looks like in practice, report by report, is on /en/llm-seo.
If you want a reading before committing to anything, the free scan returns a verdict per engine in 60 seconds, with no account and no card. And for the full method, tiers and prices in one place, the category page is /en/generative-engine-optimization-agency.
Why don't my articles rank?
Three causes show up in audits: the page does not answer the intent actually behind the query, it competes with another page of the same site, or it is linked to nothing. The audit tells you which one applies to you, page by page.
Is the content written by AI?
No. Every published text is written and reviewed by writers, and signed by a named author who has a consultant page on this site. We use tools for research and formatting, never to produce the published text.
How much content do you produce per month?
Ongoing work starts at 2 days per month with Pousse, our monthly growth tier. Volume depends on the consulting days in your plan, not on a character quota. The grid is public: see /en/what-a-geo-agency-costs.
Can ChatGPT quote an SEO page?
Yes, when the answer sits at the top of the text with its sources visible. A generative engine builds its reply from passages it has retrieved, and a page structured that way can be quoted without reformulation. We measure that pickup engine by engine in the Oracle audit, and /en/llm-seo shows how it is tracked over time.
How much does SEO content cost?
€2,670 is the published price of the Graine audit, our diagnostic tier: it maps your intents, your duplicates, and what engines already quote from you. Ongoing content work is billed in consulting days at €890/day, sold in multi-day packages, and raising intensity never raises the day rate. Full tiers and prices: /en/what-a-geo-agency-costs.
One page per intent. Not one more.
The Graine audit maps your intents, your duplicates, and what AI engines already quote from you. Published price: €2,670, from 3 days of work, delivered in 3 weeks. If you continue into Pousse or Canopée, our ongoing tiers, within 60 days, the audit price is deducted from your first month.
You leave with three deliverables: the intent map with duplicates included, the prioritized merge and revision plan, and your Absent, Mentioned or Recommended score per engine.
Start with the audit, or run the free scan first to see your current state. Either way, the prices on this page are the prices you pay: the grid is public and the same for everyone.
Case studies, measured results
How Pennylane strengthened its SEO presence on key accounting and financial queries
With Botanik, Pennylane structured a comprehensive SEO strategy combining technical optimization, educational content, and semantic targeting to strengthen its position in the financial management tools ecosystem.
Franck Neuenschwander · founder, author of the Oracle method
The method you just read, Organic Inference™, is mine: published, versioned, with its limits in writing. Fourteen years on the advertiser side, and the only agency I have ever worked at is the one I run: if you want to discuss it, you discuss it with me, not with a salesperson.
Thirty minutes with a consultant
Not a salesperson. We tell you which lever fits your case, and if none does, we say that too.
Are you absent, mentioned or recommended?