AI assistants cite pages they can extract clear, specific, attributable claims from — and they favour sources other people already treat as authoritative.
Here's what moves the needle, in rough order of impact. The ordering matters more than the list: most effort gets spent near the bottom.
1. Mentions on sites you don't own
The highest-leverage factor, and the least comfortable.
Assistants answering "best running shoes for marathon training" lean heavily on magazine roundups, review sites, and forum threads. Being in the roundup everyone cites beats any change to your own site — and no amount of on-site work substitutes for it.
This is why an AEO programme usually looks more like PR and outreach than like technical SEO.
2. Expert quotes and named attribution
Assistants prefer text they can attribute. A claim from a named person with a stated role is more quotable than the same claim in anonymous marketing prose.
Practically: put a name and a credential on your opinions.
3. Specific numbers and statistics
"Reduces impact by 23% over 40 kilometres" gets cited. "Superior cushioning technology" doesn't. Concrete figures are extractable; adjectives aren't.
Original data is the strongest version of this, because it makes you the only possible source.
4. Answering the question directly, immediately
Put the answer in the first two sentences of the relevant section. Assistants extract passages, and a page that buries its answer after four paragraphs of context often gets skipped in favour of one that leads with it.
Clear headings that match real questions help, as do tables, lists, and short FAQ-style sections — not because of any markup, but because they isolate an answer into a liftable chunk.
5. Freshness, where it matters
In categories where products, prices, and recommendations change, recency counts. A visible last-updated date and genuinely current content help. Where the subject is stable, this matters much less.
6. Consistent naming
Describe your brand and products the same way everywhere. If you're "Nike Running" in one place and "Nike Athletics" in another, you're harder to recognise as a single entity — and being recognised as one entity is what accumulates authority.
What's overrated
Structured data. Worth having as hygiene — it helps machines parse your pages. But assistants extract from visible text, and large studies haven't found schema markup driving citation lift. If someone is selling schema as an AI visibility strategy, they're selling you the least effective part of the list.
Keyword density. Actively counterproductive. Stuffing measurably reduces the chance of being cited, because it produces exactly the kind of text an assistant skips.
Publishing volume. Ten thin pages lose to one page with a real number in it.
Where to start
Improve before you create. A page that already ranks for the topic and just needs a statistic, a quote, and its answer moved to the top will beat a new page competing from zero — and it's a fraction of the work.
Then look at what's actually cited for the questions you're losing. If third-party pages dominate, your work is off-site. If your own pages are cited but you're still not recommended, the work is on the page. Those are different projects, and the citation data tells you which one you have.
See What is answer engine optimization (AEO)? for the broader picture.
