Digital PR for AI Visibility: Earning the Mentions Models Repeat

By Muntazir · Published 10 October 2026

A regional trade publication runs 400 words on a family-owned glazing firm — a profile pegged to a local award, no link back to the firm’s site, just the name, the owner’s name, and the town. Eight months later, someone asks a chatbot to name glaziers in that town, and it answers with three names. The glazing firm is one of them. Nobody built a backlink. Nobody tracked the mention. The sentence that put the name in front of a model wasn’t optimized for anything — it was just true, published somewhere a model had read.

That’s roughly the shape of digital PR’s bet in AI search: coverage that never touches a ranking algorithm can still end up inside an answer, because the thing doing the work isn’t a link, it’s the name appearing in text a model was trained on or can retrieve.

What does digital PR mean once there’s no ranking to chase?

Digital PR used to mean pitching journalists for links — a data study, a survey, a quote, anything that earned a backlink from a domain with real authority. The link was the deliverable, and the article around it was mostly a courier.

That framing assumed the reader optimizing for was a search engine counting link equity. An AI system answering a question doesn’t click through a citation to check what else the linking page says about authority. It has already read the page, or a page like it, and it’s deciding what to say based on what the text said — including whether the text used a business’s actual name or described it generically. The deliverable quietly shifted from the link to the sentence.

Why does a link matter so much less than it used to?

Ahrefs ran a Spearman correlation across 75,000 brands in December 2025, measuring which off-site signals moved with AI visibility across ChatGPT, AI Mode and AI Overviews. Branded web mentions came out strongest — 0.656 to 0.709 depending on the system. Backlinks and URL rating showed correlations the study’s own authors described as very weak. Ahrefs is careful to say correlation isn’t causation, and that established brands tend to accumulate mentions, links and search demand together — so some of that number is just “big brands are big.” But even with that caveat, a backlink doing almost nothing while a plain-text mention does a lot is a reasonable description of how these systems actually seem to work.

One number from that gap is already visible in how this product’s own free scan is built: every audit it runs, paid or not, checks a domain’s Ahrefs Domain Rating — a backlink-authority score, because it’s cheap enough to check unconditionally. That’s a legacy-SEO metric, carried over because it’s useful for plenty of other things. It is not, on this evidence, a particularly good proxy for whether a model will say a business’s name back. It’s one of several off-site signals the AI SEO guide covers — most still matter, just not for this.

Does getting cited mean getting named?

Not reliably. A separate study — Semrush with Kevin Indig’s Growth Memo, published June 2026, covering 3,981 brand appearances across 115 prompts in 14 countries on ChatGPT, Gemini, Google AI Overviews and AI Mode — found that 74.9% of appearances carried a citation link, but only 38.3% named the brand in the answer text, and just 13.2% did both. The rest were links with no name, or — this is the part that matters for PR — a name with no link.

That means more than a third of the appearances the study counted were a model saying a business’s name without any citation trail a tool could find. A press mention that drives a “ghost” result like that — a name repeated, no link to track — is close to invisible to the usual measurement, and close to the entire point of the exercise at the same time. Coverage designed only to earn a link is optimizing for the 74.9% column. Coverage that gets the business’s name spelled out in running text is optimizing for the one that actually gets repeated back.

What can a business actually do differently?

A few things that sit inside normal PR practice, just aimed differently:

• Ask for the name, not just the link. A journalist happy to write “a local roofing firm” instead of “Dunholt Roofing” has, from a link-building brief, done nothing wrong — there was never a link to lose. For AI visibility, that’s the difference between showing up and not.
• Keep a consistent, checkable identity behind the name. A mention is more useful to a model trying to resolve who’s being talked about when the business’s name maps cleanly to one entity — the same facts, the same structured data, checkable through something like a schema checker, rather than three slightly different names across three write-ups.
• Target coverage that uses the name in running prose, not just in a photo caption or a linked logo. A trade write-up, a quoted comment in a local news piece, a mention inside a roundup — the format matters less than whether the name appears as text a model could plausibly have read. It’s close to the opposite of renting someone else’s domain to rank: earning a mention of the business’s own name, rather than borrowing a platform’s authority for content that isn’t about the business at all.
• Don’t expect to measure this the way link counts get measured. There’s no dashboard that reports brand-name frequency inside AI answers the way a backlink tool reports referring domains. The nearest substitute is asking a model the same few buyer-shaped questions periodically and noting, by hand, whether and how it answers — slow, manual, and still better than assuming coverage worked because it ran.

Where does this run out?

It runs out at measurement, mostly. A business can watch its own organic rankings move in response to a campaign because a rank tracker exists for that. Nothing comparable exists yet for “how often does a model say my name.” The ghost-citation numbers above are themselves closer to an industry snapshot than a tool anyone can point at their own brand and query on demand — which means a business doing this work is, for now, operating mostly on inference from someone else’s sample rather than a measurement of its own.

The glazing firm from the opening never found out a model was naming it. Someone else happened to ask and happened to mention it. Most coverage that works this way goes just as unnoticed by the business it helped — which is a strange position for a discipline built, historically, on being able to point at the result.