Volver al Blog

GEO

Mentions are the new backlinks. A zero-budget PR sprint for small brands.

Ahrefs found brand mentions track AI visibility 3 times more tightly than backlinks. Muck Rack found 84% of AI citations are earned media. Here is a zero-budget, four-part sprint to get your brand named on the pages the engines actually read.

Por WhoCanFindMe5 minutos de lectura
Sketch comparing a crossed-out chain link with a brand name repeated across a newspaper, a forum thread, a directory listing and a top-ten list.

If you are searching for how to get mentioned by AI, the honest answer starts with letting go of an old habit. For two decades the currency of being found online was the backlink: a link from someone else's website to yours. Two large studies of how AI systems cite the web suggest AI visibility without backlinks is not just possible. It may be the normal case. When ChatGPT, Perplexity, Gemini or Claude decides which businesses to name in an answer, the strongest external signal is being mentioned, in plain text, on pages the engines read. That is unusually good news for small brands, because mentions can be earned without a budget.

The two studies, in plain English

Ahrefs: mentions beat backlinks 3 to 1

Ahrefs studied 75,000 brands to see which signals move together with visibility in AI answers. Branded web mentions correlated with AI visibility at 0.664. Backlinks managed 0.218. A correlation is a measure of how tightly two things rise together, on a scale where 1 is lockstep and 0 is no relationship at all. So mentions tracked AI visibility roughly 3 times more tightly than links did.

The honest caveat: correlation is not cause. The engines do not publish their workings, and nobody outside them can prove what earns a citation. But this is the largest external dataset we have seen on the question, and it points in one direction.

Muck Rack: 84% of AI citations are earned media

Muck Rack's Generative Pulse (May 2026) analysed more than 25 million links cited by AI systems. 84% of those citations were earned media: coverage a journalist or writer chose to give you, not coverage you paid for. Press-release wire syndication, the classic paid shortcut where a service blasts your announcement across hundreds of sites, accounted for 0.04% of citations.

One more finding matters for planning. More than half of journalism citations were under 12 months old. Coverage decays. A story from 3 years ago is doing less for you than you think.

Why this is good news for small brands

Backlinks became a market. Agencies sell them, marketplaces price them, and a small brand bidding against funded competitors loses on budget alone.

Mentions work differently. A local paper naming your bakery, a community thread where someone recommends your service, a trade roundup that lists you: none of these is bought, and all of them are winnable by being genuinely useful and easy to describe. The field is not level, but it is far more level than link building ever was.

We think about it with our usual pair of questions. Readiness asks: is the door unlocked, can the engines read your site at all? Visibility asks: is anyone walking through it? Mentions live on the visibility side. They happen off your site, on pages the engines already trust. (If the door itself might be locked, start with what AI search engines actually read on your site.)

How to get mentioned by AI: the four source families

This is the zero-budget sprint. Four families of sources, one concrete action each. Expect it to take weeks, not days, and expect variance: no engine guarantees anything, and neither do we.

1. Local or trade press

Journalists cite specifics. Find the one reporter or trade blog that covers your area or your category, and offer a single checkable fact only you can supply: a number from your own operations, a local trend you can document, a before-and-after with figures. And remember the freshness finding: with more than half of journalism citations under a year old, press is not a one-off task. One useful story a year is roughly the minimum pulse.

2. Community threads, honestly

Engines read Reddit and forums, and they cite them. Participate honestly: answer real questions in the communities where your buyers ask them, say who you are, and name your business only when it is the genuine answer. Do not astroturf, which means posting fake praise for yourself under names that are not yours. It gets removed by moderators, and one exposed fake account costs more trust than months of honest replies earn.

3. Directories and review sites

Directories are unglamorous and heavily read. The action here is consistency, not volume: make your name, category and description identical on every directory and review site that covers your industry or your area. This matters more than it sounds, because engines confuse similar names easily. We have written about how only one of the three ways a brand appears in an AI answer actually counts, and inconsistent listings are how you end up in the worthless two.

4. One authoritative listicle

For most buyer questions, engines lean on roundups: "best X in Y" articles from sources they trust. Ask the four engines your own category question and note which roundup articles they cite. That is your target list. Pick the one that appears most often, find the author, and make a short factual case for inclusion. Sometimes the answer is no. It is still the highest-leverage single pitch available to you.

Give writers a sentence they can lift

Here is the tactic that makes the other four cheaper: put a quotable claim on your own site, high on the page, in plain text.

Writers on deadline lift sentences. Engines summarising your business lift the same sentences. If your about page opens with "We are a family-run roastery in Hull, roasting single-origin coffee since 2011", that exact framing will follow you into coverage and into AI answers. If your homepage opens with "Unlock synergistic solutions", there is nothing liftable, and every writer has to invent their own description of you. Some will invent it wrong.

One sentence. Who you are, what category you are in, where you are, one concrete number if you have one. Put it where a skim reader lands first.

What we would skip

In the spirit of showing our working, two things we would not spend even free effort on.

Press-release wires. They account for 0.04% of citations in the Muck Rack data. The maths does not support the habit.

Exotic technical fixes ahead of the basics. We have already published the honest maths on llms.txt. A mention on a page engines already read moves more than a manifest file most crawlers never fetch.

Measure before you sprint

A caution from our methodology: only real mentions count. An engine parroting your name back, or talking about a similarly named company, is worth zero, and a before-and-after only means something if you classify answers properly. A mention is also not the same as a recommendation, a distinction we covered in AI mentions vs recommendations.

So take a baseline first. Run a free scan at whocanfindme.com. No signup. Takes about ten seconds. Then run the sprint, give it a couple of months, and scan again. If the mentions land, you will see it. If they do not, you will know that too, which is the point of measuring.

Siguiente

GEO, Measurement

One question, ten hidden ones

When a buyer asks an AI one question, the engine quietly turns it into about ten. Seer Interactive measured 10.7 hidden sub-queries per prompt across 501 prompts on Gemini 3. Most sites answer the headline question and none of the ten. Here is how to find your gaps.

Leer más
Measurement

A confidence interval on one AI run is fiction

The same question, same engine, same day can produce different answers. That is the nature of AI search, not a bug. The 3 rules we use to avoid showing you fake statistics, and how to read a visibility score without fooling yourself.

Leer más