AI models overwhelmingly cite earned media, not your website. Around 84% of the links ChatGPT, Claude and Gemini use to build their answers come from third-party coverage rather than brand-owned content. Which means the agencies who have spent decades getting clients written about are sitting on the single most valuable asset in AI search.
Most of them have not noticed yet. That is the opportunity.
Earned media is the real GEO cheat code, and PR agencies are sitting on it. Let me get the eye-rolling out of the way first.
GEO, or Generative Engine Optimisation, arrived with the full LinkedIn circus. New acronym, new gurus, new "frameworks" drawn on napkins, the lot. Half of what you read about it is a rebadged SEO deck with the dates changed.
But underneath the noise there is a boring, important fact. When ChatGPT answers a question about a product, a market or a company, it does not pull that answer out of the brand's carefully polished homepage. It pulls it from what other people have said. Reviews. News articles. Forum threads. Comparison pieces. The stuff you cannot buy and cannot fake.
Muck Rack analysed more than 25 million links across ChatGPT, Claude and Gemini responses spanning 17 industries. Earned media accounted for 84% of citations. Paid and advertorial content accounted for 0.3%. Read that again. You could spend your entire Q4 budget on sponsored placements and, in the eyes of the machine, be effectively invisible.
This is not a rounding error. It is the whole game.
Where AI answers actually get their facts
The data has stopped being a hunch and started being a pattern. Here is what the machine is really reading when it builds an answer.
| Source type | Share of AI citations | What it tells you |
|---|---|---|
| Earned media (news, reviews, third-party coverage) | ~84% | The machine trusts what others say about you, not what you say about yourself |
| Non-paid sources overall | ~94% | Credibility is the currency, and it is not for sale |
| Journalism specifically | ~27% | Trade and news coverage does a lot of the heavy lifting |
| Paid and advertorial | 0.3% | Your ad budget is close to invisible in AI answers |
Why do AI models trust earned media so much?
Because they are built to. A large language model is, at heart, a very expensive pattern-matcher trying not to get sued. When it assembles an answer it leans on sources that look credible, corroborated and independent. A brand saying "we are the market leader" is marketing. Three respected publications saying the same thing is a fact. The model treats the second one as truth and repeats it.
The research backs this up rather than just sounding clever at a conference. Ahrefs ran a study across 75,000 brands and found brand mentions correlated with AI visibility at 0.664. Backlinks, the old SEO currency, managed a limp 0.218. Being talked about beats being linked to, by a factor of three.
The academics got there first. The Princeton-led GEO study presented at KDD 2024 found that adding statistics lifted a page's visibility in generative answers by 41%, and adding quotations from credible sources by 28%. Lower-ranked pages that cited external authorities saw boosts of up to 115%. Citations, quotes, third-party validation. The exact raw material a decent PR team produces before its second coffee.
So which brands are actually winning?
The honest answer is that the named case studies are still thin, because everyone is learning this in public. But the shape of the winners is clear.
Vercel, the developer platform, now gets roughly 10% of its new sign-ups referred from ChatGPT. Six months earlier that figure was 1%. They did not achieve that by stuffing keywords. They did it by being the thing developers write about, reference and recommend across the open web, which the model then reads and repeats.
The second pattern is distribution. Stacker found that spreading a single story across multiple publications rather than parking it on your own site lifted AI citations by up to 325%. One press release, placed well in fifteen outlets, does not add up. It multiplies.
The model reads Axios, which topped ChatGPT's citation rankings in 13 of 17 industries. It reads Reddit, PubMed and Wikipedia depending on the platform. It has its own taste, and it did not consult you.
What should PR teams and agencies actually do?
Here is the part the hype merchants skip, because it involves work rather than webinars.
The GEO earned-media checklist
- Map where your client already gets cited. Run their brand and their competitors through a GEO analytics platform and see which outlets, threads and pages the models actually lift from. That is your real target list, not the one in your CRM.
- Seed the facts, not the adjectives. Models cite specifics. Get clean statistics and quotable lines into credible coverage, because "a leading provider" gets ignored while "cut onboarding time by 40%" gets quoted word for word.
- Go wide on distribution. Syndicate the same story across multiple trusted outlets rather than hoarding it on the newsroom page nobody visits.
- Chase the sources the model trusts, not the ones you know. If Axios and a couple of niche trade titles are doing the citing, that is where the pitch goes, however unglamorous.
- Measure citations, not clippings. The old coverage report counted eyeballs. The new one counts whether the machine repeats you. Report that to the client and watch their face change.
None of this is magic. It is PR, pointed at a new reader. The reader just happens to answer a billion questions a day and never forgets a source.
Turn earned media into AI citations
Our GEO for PR Agencies module shows agency leadership, account teams and strategists how to run a GEO analysis and turn citation data into client media plans you can sell next week. Half a day, hands-on, no napkin frameworks.
The uncomfortable opportunity
Agencies love telling clients they are ahead of the curve. Here is a rare chance to actually be. SEO consultancies are scrambling to learn earned media from scratch. Media agencies are trying to explain why their paid placements generate 0.3% of citations. And PR, the function everyone spent a decade calling unmeasurable, turns out to own the exact lever that moves AI visibility.
You will not have this advantage for long. Everyone reads the same studies eventually. That gap is exactly what our GEO for PR Agencies module is built to close, and it pairs neatly with our Introduction to GEO for teams starting from scratch.
The machine is already quoting somebody in your client's market. The only question is whether it is them.
Frequently asked questions
Do AI models really prefer earned media over a brand's own website?
Yes, and by a wide margin. Muck Rack's analysis of more than 25 million AI citations found around 84% came from earned media, a figure that has held between 82% and 89% across three editions of the study. Brand-owned content and paid placements barely register.
Is GEO just SEO with a new name?
No. Traditional SEO rewards backlinks and on-page optimisation. GEO rewards being mentioned, quoted and corroborated across credible third-party sources. Ahrefs found brand mentions correlate with AI visibility about three times more strongly than backlinks do.
Why are PR agencies well placed to lead on GEO?
Because earned media is their core product. The skill AI rewards, getting clients credibly talked about across trusted outlets, is the skill PR has practised for decades. Most agencies just have not repointed it at AI models yet.
Does paid media help with AI citations?
Almost not at all. Paid and advertorial content accounted for just 0.3% of AI citations in Muck Rack's study. For AI visibility, credibility cannot be bought.
How quickly can a PR team get up to speed on GEO?
Our GEO for PR Agencies module is a half day, three hours, and hands-on. Most teams leave able to run a GEO analysis and turn it into a client media plan the same week.