Mentions and digital PR: what a mention does for AI visibility, and how to earn one

Why does an assistant care whether anyone else names you?

Because your own site is a claim and a second page is a check. An assistant deciding whether to put a business in an answer has read that business's homepage, which says it is excellent, and has read the same sentence on every competitor's homepage. What separates them is what other pages say: that the business exists at that address, does that work, has been written about, is listed by the association that licenses the trade. A mention is evidence from someone with no reason to flatter you.

This is different from a backlink, though the two often travel together. A link is a pointer a search engine counts; a mention is a statement a reader, human or machine, can understand without following anything. A newspaper that names your business and states what you did has told an assistant something it can use even if the paper never linked. We keep the two apart in backlinks versus mentions; here the subject is the statement.

What the audit shows about who gets named

One finding from our audit of 9,483 small business websites (40 trades, 90 metros, September 2026) bears directly on this. For 9,240 of those businesses we put one real buyer question to Claude with web retrieval on and read every name it gave back. 99.4% of the names were actual businesses. 0.6% were directories or aggregators. Google's first page for the same questions is mostly aggregators; the assistant's answer was mostly not.

Read that carefully, because it says two things. The aggregator is no longer between you and the buyer in an assistant's answer, which is good news for a business that spent a decade unable to outrank one. And the pages that put a business into the answer are pages about that business: its own site first, then the places that describe it. 14.9% of the businesses were named at all, 17.7% among the 5,928 answers that gave a differentiated list rather than a hedge. One assistant, one question each, one day, so a direction and not a rate. The businesses that were named were, on inspection, the ones whose sites had the answers and the consistent identity, and whose names appeared elsewhere in the same form.

What the research says about being cited

The one controlled study we cite, Aggarwal et al., "GEO: Generative Engine Optimization" (KDD 2024), looked at the other side of the same coin: what a source page can do to be cited more often by a generative engine. On its 10,000-query benchmark, adding citations, quotations and statistics to a source raised its visibility in the generated answers by up to about 40% for the best methods. The study measured the effect, not the reason for it, and no operator documents one.

Turn that around and it is the case for mentions. A page about you that carries a quotation from you, a figure you measured or a fact it checked is the kind of source the study found more visible, and it names you. Being quoted in someone else's well-sourced page is worth more than a bare link from a page that says nothing. The practical consequence for a small business is that the thing to give a journalist or an association is a fact, not a slogan.

Where a small business earns mentions

Every row has the same rule in its middle column: the mention states facts that match your site. 43.1% of the audited sites were inconsistent about who and where they are across their own three pages, and a business that cannot agree with itself is asking a directory to guess. Settle the one form of your name, address and phone before you ask anyone to repeat it.

Digital PR that a machine can use

Digital PR for a small business is not a press release about a rebrand. It is giving people who write about your trade or your town something true and specific they did not have: a measurement you took, a price list you published when nobody else does, a project with a before and an after, a season's numbers from your own books. The story is the fact; your name travels with it.

1. Decide the one fact you can stand behind and would be happy to see quoted next to your name for a year.
2. Write it up on your own site first, dated, under a named author, with how you know it. That page is what the coverage will cite, and what an assistant will read when it checks the coverage.
3. Offer it to the three people who cover your trade or your area, by name, with the fact in the first sentence and a quote they can use. No attachments, no embargoes, no slogan.
4. When it runs, check the piece for your name, address and the fact, and ask for a correction if any is wrong. A mention with a wrong address is a contradiction you paid for.
5. Add the coverage to your site's press page as a plain list with dates, so the mentions are readable from the page an assistant fetches first.

We do this ourselves. The audit this article keeps citing exists so that there is a fact next to our name, and the page you are reading is the version of it we would want quoted.

What not to do

Do not buy mentions dressed as coverage. A paid placement is an advertisement; Google's spam policies treat undisclosed paid links as link spam, and a reader who discovers the arrangement discounts everything else the page says about you. Do not mass-produce guest articles for the link; the same policies name scaled content abuse, and the pieces say nothing an assistant can use anyway. Do not invent awards, memberships or press quotes; an assistant that finds no page behind the claim has found a claim it cannot verify, and the ones it can verify are the competitor's.

Do not seed reviews or write them yourself. Google's review snippet rules exclude fake reviews and incentivized ones that do not say so, and the platforms remove what they detect; how to put the real ones where a machine reads them is in reviews as an AI signal. And do not submit your business to fifty directories in an afternoon with details typed from memory. Ten listings that agree beat fifty that contradict, because the contradiction is what a parser notices.

How to tell whether the mentions changed anything

Measure the answer, not the coverage. Keep a fixed panel of the questions your buyers ask, run it against the four assistants monthly, and watch three things: whether your name enters the answers to the problem questions, whether the facts the assistant states about you now match the coverage, and which other businesses are named beside you. A mention that worked shows up as a name in an answer where there was none, holding across runs. How to measure AI visibility sets out the method; our own measurement is a narrower form of it: one assistant, Claude, asked the same twelve questions at the baseline, again once the fixes are live and every month with monitoring, with every answer kept word for word.