We're getting asked more and more about "AI SEO", and I find myself constantly explaining the same thing: the crux of good, traditional SEO will, by its very nature, be doing most of the AI stuff as well. There are a few nuances, which is really what this piece is about, but that is the headline.
I spend a lot of time listening to podcasts from leading industry experts trying to get this exact point across to their clients (in the main, much larger than ours). So I figured, let's just write it down. I've covered most of the points I can think of. If you've got questions, reach out; if I don't know the answer I'll find out, because every day's a learning day. Hopefully it's useful if you're starting to fold AI into your day to day.
One caveat before we start. I'm leaning on core trusted sources here, Google in particular (who, I'm told, are quite big when it comes to search and AI), plus a few other respected folk I follow. But this is tech, and tech moves fast, so anything specific is worth a sense-check on the day you read it.
So here is the position, plainly: good SEO is the foundation of AI SEO. Put the usual, standard good practice in place and AI is very likely to pick up on it too. But "foundation" is doing real work in that sentence. A foundation is necessary. It is not sufficient. The interesting bit is knowing exactly where the foundation ends and the smaller new layer begins.
The case for continuity
The strongest evidence that AI SEO is mostly SEO comes from the one source with no incentive to sell you a new discipline. In May 2026, Google published official guidance on optimising for its generative AI features, and the headline was blunt: answer engine optimisation and generative engine optimisation are, in Google's framing, still SEO. The same guidance explicitly told site owners they do not need an llms.txt file, content chunking or special AI-specific schema to appear in AI Overviews or AI Mode. AI Overviews pull from the same index and apply the same quality signals as traditional organic results.
That matters because it cuts through a lot of vendor noise. If Google's own AI features draw from the organic index, then ranking in that index is not a legacy concern. It is the entry ticket. The wider AI search landscape tells the same story. Analyses through early 2026 consistently find that the large majority of URLs cited by AI systems already rank in the organic top 10, and for Google AI Overviews specifically, almost all citations come from pages on the first page. ChatGPT leans heavily on Bing's index, so your Bing visibility feeds directly into whether ChatGPT surfaces you at all.
The practical takeaway is unglamorous and reassuring in equal measure. Crawlability, indexation, site speed, internal linking, topical depth and demonstrable expertise are not things you do instead of AI optimisation. They are the conditions that make AI optimisation possible in the first place. An AI system cannot quote or recommend a page it has never successfully crawled and never deemed trustworthy. Without reliable, well-structured, authoritative content, these systems have nothing to summarise.
This is also why the doomier predictions deserve scepticism. The framing of AI as a replacement for SEO assumes the two are substitutes. They are not. AI search is, to a large degree, SEO's output being consumed through a different interface. The work that earns a top ranking is largely the same work that earns a citation, because both depend on the same underlying signals of relevance, clarity and trust.
Where the foundation ends
So far, so comforting. But if the story stopped there, this would be a reassurance piece rather than a useful one, and we would be ignoring the evidence that does not fit neatly.
The first and most important divergence is that ranking and being cited are not the same outcome. They overlap, but the overlap is partial and, by several measures, shrinking. Some 2026 analyses put the share of AI Overview citations that also rank in the organic top 10 at well under half, and trending down over the year. The other way round, a fair chunk of citations across the LLM platforms come from pages that barely register in Google's top 100 at all. Two discovery surfaces, partly overlapping, each with its own selection logic. Ranking number one is a strong predictor of citation. It is not a guarantee of it.
The second divergence is structural, and it is where a lot of otherwise excellent content quietly fails. Traditional SEO has long rewarded the narrative article: the piece that builds context, develops an argument and arrives at a conclusion. It signals depth, it earns dwell time, and it has served us well for years. AI extraction does not read that way. These systems favour answer-first content, the direct answer near the top with the detail underneath, rather than buried beneath four hundred words of throat-clearing. The same article that ranks in position one for a human reader can go entirely uncited because the answer is not where the model goes looking for it.
The fix is genuinely additive rather than a trade-off, which is the reassuring part. A concise answer near the top of a section, a real question as a heading with the answer straight after it, shorter paragraphs: none of this harms your traditional rankings. It is neutral-to-positive for classic SEO and materially helpful for AI extraction. But it is a deliberate choice you now have to make, and it is not one the old playbook necessarily prompted.
The FAQ case, and why it proves the point
The cleanest illustration of all this is the humble FAQ. It's worth dwelling on, because it captures the whole argument in one example.
Under pure traditional-SEO logic in 2026, FAQ schema looks close to pointless. Google removed FAQ rich results from standard search listings back in 2023, restricting them to a narrow set of government and health authority sites, and here's Google latest update on the state of FAQs. If your only goal is the blue-link SERP, adding FAQPage markup to a page earns you almost nothing visible. A classic SEO audit might reasonably flag it as effort better spent elsewhere.
And yet. AI platforms, including ChatGPT, Perplexity and Google's own AI features, actively crawl and extract FAQ-structured content. A question-and-answer block is, more or less, a pre-packaged extractable answer, which is exactly the format these systems like to lift and cite. The schema that lost its visibility in Google's links quietly gained value as a signal for generative search. Adoption is still low (structured data of any kind sits on only a small minority of domains), which makes it an opportunity rather than table stakes.
Here is the subtlety worth naming. There are pages where an FAQ section would not arise naturally from the content. A focused service page, a sharp opinion piece, a tightly argued case study: bolting questions onto these can feel forced if you are only thinking about the human reading top to bottom. The traditional instinct is to leave them off. The AI-aware instinct is to ask whether a small, genuine FAQ block would hand an extraction engine a clean answer to a real query the page deserves to win. Sometimes the answer is no, and a forced FAQ helps nobody. But the point stands: AI SEO will occasionally ask you to add something that classic on-page logic told you to skip. That is a real, if narrow, divergence from "just do good SEO."
The genuinely new layer
There is one area with no real traditional-SEO equivalent at all, and it is worth being honest that it sits slightly outside the foundation metaphor. AI systems appear to weight how often, and in what contexts, they encounter a brand across the wider web: third-party coverage, mentions, the general density of credible references to you that the model has seen. This is not a ranking lever you pull on your own site. It edges into territory that looks more like digital PR and reputation than classic technical or on-page SEO. You can have impeccable on-site optimisation and still be thinly represented in the places models learn to trust a name.
Add to that the practical reality of measurement. AI answers are non-deterministic. Ask the same question ten times and you can get ten different responses, with different sources. The platforms fragment, too: Google AI Overviews draw on Google's index, ChatGPT leans on Bing, Perplexity runs its own crawler. There is no single rank-tracker equivalent that tells you cleanly where you stand. This is the part of AI SEO that is genuinely immature, and anyone claiming a settled, repeatable playbook for it is, in my view, overstating the case.
So what should you actually do
The conclusion is a calm one. Don't tear up the SEO playbook, and don't treat AI optimisation as a separate department with its own budget and its own mysterious rules. The overwhelming majority of the work is the work you already know: technical health, genuine topical authority, demonstrable expertise, clean structured content. Get that wrong and no amount of AI-specific tinkering will save you, because you won't be in the candidate set the models draw from in the first place.
Then layer deliberately on top. Lead sections with a direct answer. Use real questions as headings where they fit. Add structured data, FAQ markup included, where a genuine question-and-answer would serve an extraction engine, even on pages where the old logic said not to bother. Make your claims verifiable, with named sources and specifics rather than vague assertions, because that is what these systems prefer to cite. And start paying attention to how your brand turns up off your own site, because that is the one lever the foundation metaphor does not cover.
