There are two loud camps in the AI search conversation right now, and both are wrong.
The first says nothing has changed. AI search is just SEO with a new coat of paint, so keep doing what you have always done and you will be fine. The second says SEO is dead, throw out the playbook, and buy a shiny new “GEO” tool that will get you cited by ChatGPT.
The truth sits in the uncomfortable middle, and it matters because getting it wrong is expensive. Lean too far toward “nothing changed” and you will quietly disappear from the answers your customers now read instead of clicking. Lean too far toward “SEO is dead” and you will waste budget chasing tactics with no foundation under them.
So let us settle it properly.
The part that really is just fundamentals
Start with how these systems actually work, because it dismantles half the hype on its own.
ChatGPT, Perplexity, Google’s AI Overviews and AI Mode, Gemini, and the rest do not conjure answers from thin air. They pull live from the web, and they lean on the same index and the same authority signals that traditional search has always used. If a crawler cannot find your page, render it, and trust it, a language model cannot cite it. Full stop.
That means the boring foundations still decide most of the outcome. Technical health, fast and crawlable pages, clean site architecture, genuine topical authority, and quality links. The pages that already rank well are, more often than not, the same ones getting pulled into AI answers. Google has said as much in its own guidance: there is no separate “AI ranking” lever, and the best preparation for AI features is the same quality and relevance work that has always mattered.
So anyone selling generative engine optimisation as a completely separate discipline that ignores crawlability and content quality is selling you snake oil. Roughly 70 to 80 percent of your AI visibility is earned through SEO fundamentals done properly. If the fundamentals are broken, no amount of AI-specific tinkering will save you.
That is the myth-confirming half. Now for the half that the “nothing changed” camp gets badly wrong.
The part that is genuinely new
There are real techniques that move AI visibility specifically, above and beyond classic SEO, and over the past year the evidence behind them has firmed up considerably. Four are worth your attention.
Structure content for extraction, not just ranking. Language models lift sentences, not pages. They want a clean, self-contained passage they can drop into an answer. That changes how you write. Lead with a crisp, direct answer or a plain definition near the top of the page. Use question-style headings that mirror how people actually ask. Write passages that still make sense when pulled out of their surrounding context. This is not cosmetic. The widely cited Princeton study on generative engine optimisation found that adding citations, statistics, and quotations lifted a source’s visibility in AI answers by roughly 30 to 40 percent over unoptimised content. Same facts, better structure, more citations.
Publish original data and analysis. This is the biggest genuine departure from old-school SEO, and the one most businesses ignore. Models reward primary sources, not summaries of what everyone else already said. Analyses of millions of AI citations keep landing on the same finding: trends pieces and content containing unique, first-party data get cited far more often than generic explainer content. If your blog is the fortieth “what is X” article on the web, you are invisible. If you run a survey of your customers, publish a benchmark, or share numbers only you have, you become citable. Being the source beats being the summary.
Build entity presence off your own site. Here is the shift that trips up people who think in terms of backlinks. It is less about links pointing at you and more about your brand, your people, and your products being mentioned consistently, and described consistently, across the sources these models trust most. Reddit and LinkedIn punch enormously above their weight in AI citations. So do Wikipedia, reputable industry press, and established review platforms. Language models build entity graphs, so consistent naming of your brand, products, and services across the web helps them attribute the right things to you. Getting talked about in the right places is now one of the highest-leverage activities you can invest in.
Measure citations, not just rankings. You cannot manage what you cannot see. The newest muscle to build is tracking your share of voice inside AI answers: which prompts cite you, against which competitors, and how that changes over time. A category of tools has appeared specifically to do this, and it did not meaningfully exist two years ago. Rank tracking alone no longer tells you the full story, because a customer can get their answer, form a preference, and never see a blue link at all.
A couple of things to stay sceptical about
Being on top of AI search also means knowing what to ignore.
The proposed “llms.txt” file, a manifest that supposedly tells models how to read your site, is talked up constantly. There is still no solid confirmation the major engines actually consume it. It is cheap to add, so no harm done, but do not treat it as a growth lever.
And treat the eye-watering conversion statistics floating around with caution. You will see claims that AI referrals convert many times better than organic. Some of that is real, because someone arriving after an AI has effectively pre-qualified them is warmer. But a lot of these numbers come from vendors with an obvious interest in the figures looking impressive. Trust your own analytics over someone else’s marketing chart.
So, what should you actually do?
Stop thinking of SEO and AI search as a fork in the road. They are converging, not splitting apart. The work is additive. You still need to be findable and trusted the traditional way, and now you also need to be quotable and consistently present in the places models draw from.
Practically, that looks like this. Get the fundamentals genuinely right first, because everything else builds on them. Then restructure your most important pages so a model can lift a clean answer straight off them. Invest in original data and real points of view rather than rehashed explainers. Get your brand mentioned, accurately and consistently, across the platforms these engines trust. And put a measurement system in place so you can see whether any of it is working.
The businesses winning in AI search are not the ones who bought a magic tool or the ones who buried their heads and carried on. They are the ones who kept doing the fundamentals properly and layered genuinely new work on top.
That is the whole myth, busted. It was never “SEO or something new.” It was always both.
