SEO and AI visibility are not the same job

Every week or so someone asks some version of "so this is just SEO for AI, right?" I get why — it's the easiest bucket to put it in. But no. The difference isn't cosmetic. It's mechanical, and understanding it matters more than the vocabulary does.

Traditional SEO is, at its core, a negotiation with a crawler. Google sends a bot to your site, the bot reads your HTML, follows your links, notices how other sites link back to you, and eventually decides where your page belongs in a ranked list of results for a given query. Everything about SEO — backlinks, keyword density, page speed, schema markup — exists because it changes some input to that ranking algorithm. You're optimizing a page to win a competition against other pages, judged by a system built to rank pages.

An AI model asked "best accountant for freelancers in Rotterdam" is not ranking pages. It's not crawling anything in that moment. Depending on the setup, it's drawing on what it absorbed during training, or running a live search and reading a handful of results, or both — and then doing something a search engine never had to do: writing an answer. Picking one name, maybe three, saying them out loud with a reason attached. There's no results page for the user to scroll and form their own opinion. The model already formed one.

That's the part people miss when they say "AI SEO." SEO's job is to win a ranking. This job is to win a sentence. Those aren't solved the same way.

I watched this happen firsthand building Foreground. Sites with genuinely strong SEO — good domain authority, clean technical health, real backlinks — got skipped entirely in AI answers. A smaller, scrappier competitor with a plainly-written "about" page got named instead. Not because Google would have ranked the smaller site higher. It wouldn't have. But the model wasn't ranking pages. It was looking for a clear, specific, quotable answer to "who does this and why should I pick them" — and one site handed that over on a plate while the other buried it under three paragraphs of mission-statement language.

That's the actual skill this requires, and it has almost nothing to do with backlinks: say, in plain sentences, exactly what you do, who it's for, and what makes you different — in text a model can lift and repeat without interpreting or guessing. Vague copy is invisible to a model in a way it was never quite as invisible to a human reader. A human reader fills in the gaps with context and trust signals — a nice logo, a professional photo — that a language model mostly can't use.

There's also a freshness difference worth naming. A page you wrote in 2019 can still rank on Google today if nothing better has come along. An AI answer is more likely to lean on whatever a live search just pulled back, or on how recently your name showed up somewhere the model trusts — a directory, a review site, your own updated copy. Stale but well-optimized content ages better in Google's world than it does in this one.

None of this means SEO stops mattering. A site still needs to exist, be crawlable, and rank well enough that a model's live search pulls it up as a candidate in the first place. GEO doesn't replace SEO — it sits on top of it. But treating them as the same discipline, solved by the same checklist, is how a lot of businesses are going to spend the next year doing more of what already isn't moving the number that's starting to matter.

If I had to leave someone with one sentence, it's this: SEO gets you into the pile of candidates. What happens after that — whether the model actually says your name — is a different job, with different rules, and right now almost nobody is doing it on purpose.

Next: What to actually do this week if AI doesn't know you exist — a free, 40-minute checklist to start fixing it.

— Marc. I build Foreground, which measures the "does the model say your name" part specifically. Try it free.