Thibault Besson-Magdelain - SEO Software founder

I'm an SEO and GEO expert. I spend my time on one question: how does a brand get found, on Google and inside the AI tools that are quietly replacing it.

Search isn't what it was two years ago

For twenty years, ranking meant one thing: show up on a results page, get the click, convert. That game still exists, but it's no longer the whole game.

Today a growing share of questions never reach a results page at all. They go straight into ChatGPT, Perplexity, Gemini or Google's AI Overviews, and come back as a direct answer, sometimes with a source, sometimes without one. If your brand isn't part of what the model was trained on, or isn't structured clearly enough to be cited in that answer, you don't just rank lower. You disappear from the conversation entirely.

That's the shift I work on. Not "AI SEO" as a marketing label, but the actual mechanics of it: what makes a page citable, what makes an entity recognizable across sources, and what makes a brand consistent enough that a model repeats it back correctly instead of guessing.

What GEO actually means

GEO stands for Generative Engine Optimization. People throw the term around loosely, so here's how I actually use it.

Classic SEO optimizes for a ranking algorithm that reads links, keywords and technical signals, then hands you a list of ten blue links. GEO optimizes for a language model that reads your content once, compresses it into an internal representation, and later reproduces a version of it in an answer. Those are different jobs.

A model doesn't care about your keyword density. It cares about:

1. Whether your content answers a question directly, in a self-contained passage it can lift and cite.
2. Whether your brand and your facts are consistent everywhere the model has seen them, because contradictions get discarded in favor of whatever version repeats most often.
3. Whether your site is technically accessible to AI crawlers in the first place, which sounds basic and is still wrong on most sites I check.
4. Whether third parties (reviews, directories, forums, other sites) describe you the same way you describe yourself, because a model trusts convergence more than self-promotion.

None of this replaces classic SEO. It sits on top of it. A page that doesn't rank on Google rarely gets crawled enough to be cited by AI either. The two systems feed each other now, and treating them as separate projects is how most sites end up invisible in both.

What I actually do about it

Day to day, this breaks down into a few concrete moves, repeated for every client, every site, every language:

Technical foundations first. Indexation, crawl access for AI bots specifically (GPTBot, ClaudeBot, PerplexityBot and the rest each behave differently), page speed, and a sitemap that actually reflects what's live. Nothing else matters if the crawler can't reach the page.

Structure the content to be lifted, not just read. Answer-first paragraphs, clear headings that map to real questions, definitions a model can quote without rewriting, and comparison tables where the format itself is the answer. A well-structured page gets cited. A well-written page that buries its point in paragraph four doesn't.

Keep the entity consistent. Same name, same description, same facts, in the same order, across every surface a model might have crawled. This is the part almost nobody does deliberately. Most brands say something slightly different on their site, their LinkedIn, their directory listings and their press mentions, and then wonder why an AI answer gets a detail wrong. It's not wrong. It's averaging across contradictory sources.

Internal linking and topical structure, done automatically. Which pages link to which, which topics a site should own, which keywords map to which page: this is exactly the kind of decision that should not depend on someone remembering to do it manually every week. I build this to run on its own.

Measure visibility inside AI answers, not just rankings. Whether a brand actually gets mentioned when someone asks a relevant question in ChatGPT or Perplexity is a different metric from a Google position, and almost nobody was tracking it a year ago. Now it's the number that matters.

Sorank

I put all of this into a piece of software: Sorank.

It measures and improves a brand's visibility across both Google and AI, on autopilot, instead of relying on a manual audit that's outdated a month after it's delivered. Targeting, internal linking and topic coverage run automatically once it's connected to a site. It's built for any sector, no exceptions: the same underlying problem (being findable and citable) shows up identically whether the site sells software, services or physical products.

One client went from 0 to 10'000 visitors in 5 months after putting this into practice. That's the kind of result I care about talking about: measurable, repeatable, not a screenshot of a vanity metric.

Who this is actually useful for

Mostly three kinds of people end up talking to me:

1. SEO agencies that need to prove their value to clients in an era where "we improved your ranking" isn't the full story anymore, and clients are starting to ask about AI visibility directly.
2. Independent SEO consultants who want the technical and GEO side handled by software instead of rebuilding the same audit from scratch for every client.
3. Large sites with enough pages that manual optimization stopped being realistic a long time ago, and where automation is the only way to keep structure and internal linking consistent at scale.

If none of those describe you, this still might be useful reading. If one of them does, worth a look.

A few things I keep coming back to

Not slogans, just what actually holds up after doing this long enough:

1. The people who don't build anything are the ones still consuming everyone else's advice about building.
2. A product people genuinely love creates its own army of users who explain it to others better than any ad could.
3. Information without execution is entertainment, not progress. Reading another guide doesn't move the number.
4. Systems beat motivation every single time it matters, because motivation runs out on exactly the days you need the work done most.
5. The goal was never "do SEO." The goal is to build a machine that keeps working without me standing over it.
6. Attention isn't the currency people think it is. Conversion is. A million views that convert nobody is worth less than a thousand that convert a few.
7. Human imperfection connects better than AI-generated polish. The paradox of this whole field: I build software to win visibility in AI systems, and the content that actually earns trust once someone lands on the page still has to sound like a person wrote it.

Follow along

I write about SEO and GEO, mostly in public, mostly about what's actually working right now rather than theory. If you're building something and want it to actually get found:

• Check Sorank.
• Get a free AI visibility check: sorank.com/call.
• Or just follow along here for the next thing I learn.