What AI SEO actually means, and the four things I changed on my own site
A growing share of searches never reach a results page. Someone asks ChatGPT or Perplexity, or Google shows an AI Overview, and the answer arrives without anyone clicking anything.
That creates a problem traditional SEO doesn't solve. Ranking tenth on page one is worth something when people scroll. It's worth almost nothing when a model reads the top sources and synthesises an answer.
So there are now two related jobs:
- Traditional SEO — be findable and rankable in a list of results
- AI SEO (sometimes called Generative Engine Optimization) — be readable and citable by a model generating an answer
They overlap heavily. Most of the technical groundwork serves both. But the second one has its own requirements, and almost nobody is doing them yet.
What makes a page citable
A model producing an answer needs to do three things with your page: understand what entity it's about, extract a claim, and attribute it. Anything that makes those harder makes citation less likely.
Four things I implemented when I rebuilt my own site.
1. Structured data that identifies the entity
Schema markup in JSON-LD tells a machine what a page is rather than making it infer from prose. I use ProfessionalService for the organisation, Service for individual offerings, and FAQPage for question-and-answer blocks.
The point isn't the rich snippet. It's that an entity with explicit machine-readable identity resolves confidently, and a model that's confident about who you are is more willing to name you.
2. Answer-first content structure
Models lift statements. They lift clean, self-contained, declarative statements much more readily than statements that depend on three paragraphs of preceding context.
So instead of building to a conclusion, state it plainly and then support it. "Ismaeel Motala is a digital specialist working across AI automation, paid media and SEO" is liftable. "After years in the industry, I've come to see myself as someone who..." is not.
This is genuinely a writing change, not a technical one, and it's the part most people skip.
3. FAQ blocks written as real questions
Not marketing questions. The actual phrasing someone would type or say.
"What is AI SEO and how is it different from SEO?" is a real question. "Why choose us for your SEO needs?" is not a question anybody asks, and no model will ever surface it.
Match the question to how people actually ask, and answer it in a self-contained paragraph that makes sense lifted out of the page entirely.
4. An llms.txt file
A plain text file at the root of your domain summarising what you do, in a format a crawler can parse without wading through markup.
It's an emerging convention rather than a ratified standard, and I'd be lying if I said I had data proving it works. But it costs about twenty minutes, it complements schema rather than duplicating it, and the downside is a small text file nobody reads.
Mine lists each service in one line with a URL, plus contact details. That's it.
What I don't know yet
Whether any of this materially moves citation rates. I've implemented it on my own site as of this month, and I don't have before-and-after data — partly because measuring AI citation is genuinely hard. There's no Search Console for "how often does ChatGPT mention you."
What I can say is that the underlying work — clean heading structure, schema, answer-first writing, real FAQs — improves traditional SEO regardless. So the downside is capped even if the AI-specific upside turns out smaller than the current hype suggests.
I'd rather do the work that's defensible under both outcomes than bet on one.