AI Search Doesn't Just Read Your Website. It Reads What Everyone Else Says About You.

Say you've done the on-page GEO work properly. Schema's in place, your entity signals are clean, the content is fresh and specific. By every checklist going, you should be showing up when a buyer asks an AI assistant to recommend something in your category. And a competitor is showing up instead.

The instinct is to go back over your own pages and look for what's missing. That's usually the wrong place to look, because the work you've already done only covers the half of AI visibility you control directly: your own site. AI search doesn't just read your website. It reads what everyone else says about you, too.

It's an easy thing to miss when the on-page checklist is legible and this isn't. You can run a GEO score, watch the number improve, and feel like you've done something. There's no equivalent single number for "how well does the rest of the internet describe you", so it’s rarer to make it onto the lit.

Of course the on-page work still matters, this is just the other half of it that you might not have considered.

What AI search is actually reading besides your website

When an AI engine puts together an answer, it pulls from more than your own pages: comparison articles, review sites, forum threads, industry roundups, and other companies' blog posts that happen to mention you. That's the same discovery layer that sits underneath generative engine optimisation more broadly, just from an angle most GEO checklists skip.

All of that gets read the same way your own content does.

• How a comparison post on someone else's site describes your product

• What a forum thread says when someone asks "has anyone tried X"

• The framing a competitor uses when they mention you by name

An AI engine can pull any of it into a generated answer.

And that happens whether or not the page in question ever links back to you. A backlink is one way an AI system might discover a description of you. It isn't the description itself, and it isn't the only route that description takes to reach the model.

The engines get there differently, which is worth knowing if you're deciding where to focus. ChatGPT leans on training data and Bing's live index, so it favours mentions that show up consistently across high-weight sources it's already seen. Perplexity crawls the live web and cites far more sources per answer than ChatGPT does, so freshness and volume both matter more there. Google AI Overviews draws on the Knowledge Graph, which means a mention on a page Google already trusts carries more weight than the same words on a page it doesn't. None of them are reading your site to find this. They're reading everyone else's.

Why this is a measurable input, not a vague idea

Ahrefs ran the numbers across 75,000 brands, filtering for Domain Rating above 40 and a minimum search volume of 800 to keep the sample comparable, and measured how strongly a set of signals correlated with whether a brand showed up in Google's AI Overviews.

Brand web mentions correlated at 0.664. Backlinks correlated at 0.218 — roughly three times weaker.

Other off-site signals landed in between: brand anchors at 0.527, brand search volume at 0.392. Domain Rating and referring domains, the two metrics most SEO reporting still leads with, both came in below mentions and anchors.

Ahrefs are upfront that this is correlation, not proof of causation. Brands with a strong existing presence are more likely to be mentioned a lot and cited a lot, so some of this could be the same underlying strength showing up twice. I think that's a fair caveat and worth taking seriously. It doesn't change the practical point, though: the signal everyone already tracks is measurably weaker than the one almost nobody tracks.

None of this makes backlinks worthless. It means the way you're described elsewhere deserves the same deliberate attention your on-page work already gets, rather than being left to chance.

Reviews are part of this too

Customer reviews are exactly the same kind of signal, and they're the most directly actionable part of this whole picture. A review on G2, Capterra, or Trustpilot is a third party describing your product in their own words, with a level of detail and impartiality your own homepage can't produce.

Trustmary's analysis of over 35,000 AI search measurements found that review platforms account for roughly 48% of third-party citations in brand-related searches, and for product-related queries specifically, Perplexity pulled a review into every response it measured. One caveat worth flagging: most of these engines can't crawl Google Reviews directly, so the platforms that matter here are G2, Capterra, Trustpilot and similar dedicated review sites, not whatever sits on your Google Business Profile.

This is the one part of "how does the web talk about you" that you can influence directly and quickly. You can't make a comparison-post author change their framing. You can ask a happy customer for a specific, detailed review this week.

What this looks like in practice

Picture two companies selling a similar product. Company A has a comprehensive, well-optimised site: clean schema, a strong FAQ section, fresh content. Company B's site is fine, nothing special. But when someone asks an AI assistant to recommend a tool in that category, Company B gets named and Company A doesn't.

Trace it back and the difference usually sits in a handful of comparison posts, review pages and forum threads that describe Company B clearly and specifically, and either don't mention Company A or describe it vaguely (or worse, negatively). The AI is reading what everyone else said about Company B, not Company A's on-page schema, and Company A simply wasn't part of that conversation.

It also compounds in a way on-page work doesn't. A page you control stays wherever you leave it until you change it. A mention on someone else's site gets read again every time that page gets crawled, and it sits alongside whatever else that same page says about the category. If a widely-read comparison post consistently ranks Company B ahead of Company A, that framing shows up in every answer an AI system builds from it, not just once.

How to actually see this instead of guessing

Agreeing this matters is the easy part. The hard part is that most of it is invisible by default. You don't get a notification when a competitor picks up a mention you didn't, or when someone describes you inaccurately on a page you'll never think to check.

This is what Frase's Backlinks feature is actually for, and it's more than "a backlinks tool" in the old sense. It shows the domains linking to competitors but not to you, so you can see where the conversation in your category is already happening without you in it. It flags unlinked mentions: places your brand gets named with nothing to click, which is exactly the kind of signal the Ahrefs data says matters more than the link itself. And it tracks sentiment and context on existing mentions, so you can tell whether the way you're being described is actually working in your favour.

Essentially it's an audit of how the web actually talks about you right now, and helps draft the outreach to places that are linking to competitors and not to you to help increase your visibility.

Where to start this week

You don't need a six-month campaign to start on this. A few things worth doing in the next week:

• Run a competitor gap scan and see which domains link to them but not you.

• Check your own unlinked mentions, so you know where you're already being talked about with nothing to show for it.

• Look at the sentiment on your two or three most-referenced review platforms. Is the detail in those reviews actually specific, or generic enough that an AI system has nothing to lift from it?

• Pick two or three target sites based on where your competitors are already getting mentioned, and start there rather than everywhere at once.

None of this replaces the on-page GEO work. It's the half that was missing from it.

Frequently asked questions

Do backlinks still matter for AI search?

Yes. They're still one of the ways an AI system discovers a mention of you, and they still count for a meaningful chunk of the correlation Ahrefs measured. What's changed is that they're no longer the strongest signal in that set, and they were probably never the sole mechanism behind a citation on their own. This is additive to backlink strategy, not a replacement for it.

Is this the same as digital PR?

It overlaps a lot with what digital PR teams already do, and if you have that function, this is a reason to loop them into your GEO work rather than treating them as separate teams. The difference is the lens: digital PR usually optimises for coverage and links. This adds "how does this page describe us, and would an AI system read it as a clear, favourable answer" as a second thing to check.

How specifically do customer reviews affect AI citations?

Review platforms account for roughly 48% of third-party citations in brand-related searches, per Trustmary's analysis, and for product queries specifically, engines like Perplexity draw on reviews in the large majority of responses. Reviews carry two things at once: sentiment, and specific detail, both of which are useful raw material for a generated answer.

How is this different from entity optimisation?

Entity optimisation is about making your own pages unambiguous: clean schema, a definitional opening sentence, the right structured data. That's the on-page half. This is the off-page half: what other people's pages say about you, which you can influence but never fully control. You need both, and they're usually owned by different people on a team.

How long before this shows up in AI citations?

Expect it to move slower than on-page changes, because you're waiting on other people's content to get crawled, re-crawled, and in some cases retrained on. A newly earned mention on a fast-moving site like a forum or a fresh review can show up in weeks. A comparison article's framing changing, or enough volume to shift a correlation like the one Ahrefs measured, is a multi-month project, not a single sprint.

The web's description of you is part of your AI visibility now

This is a case for finishing the on-page GEO work, not abandoning it. You've made your own pages legible to AI systems. The other half of the job is knowing what the rest of the internet is telling those same systems about you, and right now, for most teams, that's a total blind spot.

The starting point is checking what's already true, not launching a bigger campaign: run a page through the GEO Score Checker if you haven't audited the on-page side yet, or go straight to the Backlinks feature if you want to see your own mention gap first. Either way, you'll know something about your AI visibility today that you didn't know yesterday. What does your own gap actually look like?