Building authority signals for the AI search era
The same authority signals that helped you rank in classic SEO now also drive citation in AI search. But the bar is higher. Here is what is genuinely working in 2025 — and what is not.
Tom BarberA year on from AI Overviews going broadly live, the picture is settling. Generative search has not killed classic SEO — but it has raised the bar significantly on the kind of authority signal that gets you cited. After a year of helping clients adapt, here is what is genuinely working.
Named authorship is now table stakes
Anonymous agency content struggles in AI search. Named authors with bios, headshots, expertise statements and links to credible profiles get cited far more often than equivalent content under a generic byline. This is not subtle — across our client data, the gap is large.
Practical implication: every blog post needs a real author, and every author needs a real profile page. A LinkedIn link in the bio is the bare minimum; relevant credentials, named published work elsewhere, and Person schema markup add measurable lift.
Third-party mentions are doing more work than backlinks
In classic SEO, a backlink from a reputable site was the gold standard authority signal. In generative search, a mention — even without a link — appears to count. AI engines crawl far more broadly than Googlebot historically did, and they aggregate mentions across directories, podcasts, press, partner content, and community sites.
The takeaway is not to abandon link building, but to broaden the definition of "being mentioned". A guest article on a respected industry blog that does not link to you is more valuable than it used to be. A podcast appearance where your name is consistently mentioned counts.
Depth on the topic, not breadth of topics
Pages on sites with significant topical depth — fifteen or twenty articles around a single subject — get cited substantially more often than pages on sites with one article on each of fifteen subjects. The signal Google's AI seems to use is consistency: "is this site demonstrably expert on this thing, or are they a generalist?"
For a small business, the lesson is to be intentional. Picking one or two topical areas to genuinely own beats spreading content thinly across everything you do.
llms.txt and structured data are catching up
A year ago, llms.txt was a curiosity. Today, the major AI engines — ChatGPT, Perplexity, Gemini, Copilot — all read it. Add one to your site. It does not replace any other signal, but it makes life easier for the bots and removes ambiguity about what your site is.
Structured data continues to matter, with one notable shift: schemas that describe people, organisations and credentials now do more work than they used to. Person nodes for authors, hasCredential on organisations, AggregateRating on local businesses. The schemas that help AI engines confidently identify "who is this and what are they an expert in" are punching above their weight.
What is not working
- Generic, well-optimised but undifferentiated content. The pages that fared worst in our analysis are the ones that looked technically perfect but had no original perspective or specific claim.
- Article farms designed for keyword coverage. They were already losing in classic SEO; AI search has accelerated that.
- Anonymous "team" bylines. Even where the content is genuinely good, no named author is a measurable disadvantage.
A short and unfashionable summary: the same things that have always worked are still working, but the floor has risen. Decent content under an anonymous byline used to get by. It no longer does.