For fifteen years earned media has been the awkward middle child of the marketing budget; loved in principle yet squeezed in practice. PR has at times been seen as the problem child of marketing in so far as it can be hard to measure and even harder to get right. In contrast to performance marketing, which had dashboards and tangible ROI points, earned had a coverage report and at times slightly defensive conversations about attribution and “reach”. Next came large language models which started answering buyers’ questions, and it turned out the machines had already decided which sources to believe. Their view is remarkably, and somewhat ironically old-fashioned: they believe the press.
When you follow what happens to a piece of coverage after it runs and you can see why.
Key takeaways
-
AI models have decided to trust the press. Licensing deals between AI labs and outlets like News Corp, the FT, AP and Reuters mean journalism sits in training data as privileged input – your own website content largely doesn’t.
-
Syndication is now a ranking signal, not redundancy. When coverage republishes across wires, aggregators and portals, that repetition is exactly what separates a “fact” from noise in a model’s eyes.
-
Recall outlasts coverage. Claims repeated across many credible sources harden into something models state confidently, long after the original story has faded from view.
-
Branded mentions beat backlinks. Ahrefs’ study of 75,000 brands found mentions across the web – not domain authority – best predicted visibility in AI answers.
-
This is already moving the needle on buyer behaviour. AI referral traffic has more than tripled in a year (Similarweb), and half of B2B buyers now start research with a chatbot, with most rating vendors more favourably when mentioned (G2).
-
Earned media is the only channel with memory. Paid stops the moment budget stops; earned keeps teaching models – and shaping buyer answers – years later.
-
“Citation share” is becoming the new share of voice. How often and how favourably AI engines reference you against competitors is the emerging metric PR teams need to own.
-
The craft hasn’t changed, but the mandate has. Judgement, journalist relationships and narrative-building still matter – but measurement now needs to stretch from “coverage” to “recall.”
-
Nothing is guaranteed. This is a probabilistic system; the goal is to stack the odds through substance, consistency and credible placement, not to chase guarantees.
Where the models learn to trust
Ultimately it all starts with where the models learn. The AI labs have spent the past couple of years signing licensing deals with a very particular kind of company; namely, news organisations. OpenAI alone has agreements with News Corp, Axel Springer, the Financial Times, the Associated Press, Reuters, Vox Media and The Atlantic. Google is paying Reddit for its conversation archive. Practically, that means journalism sits inside the training data as privileged, trusted input in a way a brand’s own blog simply does not. A story in the business press is teaching the models. Your press release, sitting unloved on the newsroom page on your website, mostly isn’t.
Why the story travels
Secondly, the story travels. Decent coverage doesn’t stay as one isolated URL; it syndicates, gets picked up by wires and aggregators, republished across finance portals and news apps, quoted and summarised. One article quietly becomes dozens of copies, each repeating the same facts about your brand. To a human, this may come across as redundancy and measured by the traditional way of reporting PR coverage, it may even seem unremarkable. To a model trained on a crawl of the web, however, it is further evidence that weighs heavily on its learning, perception and subsequently the facts and recommendations provided to the end user. Repetition across independent, credible sources is precisely the signal that separates a fact worth keeping from noise worth pruning.
And the pruning genuinely happens because models retrain on a rolling basis, and the process is ruthless with one-offs. A claim that appears once, in one place, tends to wash out quite quickly. However, a claim that appears consistently across many trusted sources hardens into something the model will confidently repeat a year later, unprompted, to a buyer you’ve never met. Coverage fades but recall persists. That distinction might be the most important new idea in our industry.
What the retrieval data shows
The retrieval side tells the same story. When researchers pull apart millions of citations from ChatGPT, Perplexity and Google’s AI Overviews, the sources the engines lean on are strikingly consistent: Reddit, Wikipedia, YouTube, LinkedIn, editorial press, review platforms. Places where third parties talk about you. Ahrefs studied 75,000 brands and found the strongest predictor of visibility in AI answers wasn’t backlinks or domain authority but branded mentions across the wider web, by a wide margin. From training data through to live citation, the whole architecture is biased towards exactly the thing PR people make – and are good at – making.
The question remains whether any of this matters yet? The answer is increasingly, yes. Similarweb has tracked AI referral traffic to websites more than tripling in a year. G2’s research found half of B2B software buyers now start their research with an AI chatbot, and 85% view a vendor more favourably when the chatbot mentions it. The industry has noticed too; in Muck Rack’s latest measurement research, 61% of PR professionals were tracking or planning to track their brand’s mentions in AI answers. Citation share, how often and how favourably the engines reference you against your competitors, is emerging as the AI-era version of share of voice.
There is an inversion, however, that is at the crux of how marketers and PR folk need to think. The marketing mix has been organised for two decades around the assumption that paid is dependable and earned is nice to have. A paid ad stops existing the moment the budget stops. It never enters the corpus, never teaches anything, never compounds. Contrast this with a piece of earned coverage that has the potential to continue working for years, feeding every future model, shaping answers for buyers who weren’t even in the market when it ran. Earned media has become the only channel with a memory.
What this means
For campaigns, that stretches the brief. The old question was how many people will see this, yet the new companion question is “what will LLMs remember about us in a year”. Which favours substance, because models retain specifics: numbers, firsts, named experts, original research. It favours the outlets that actually teach the models, licensed and heavily syndicated press over guest posts. Moreover, it favours consistency, because a story that varies wildly across interviews and announcements feeds the crawl contradictions, and contradictions wash out.
For PR professionals, credibility with journalists, the judgement to know what’s actually a story, the patience to build a narrative over years: the craft didn’t change, the distribution did. The measurement mandate changes with it. Too often, most PR ends at coverage; the pipeline ends at recall, and teams who can talk about citation share and what the engines actually say about their clients will find themselves in strategy conversations that used to happen without them.
One caution, and regular readers will know we feel strongly about it: this is a probabilistic system. Nobody can guarantee your story survives the crawl, and anyone promising you a citation is selling something you shouldn’t buy. What you can do is stack the odds, persistently, and with rigour, in the one channel the machines have decided to trust.
The story that runs today is still working next year, in answers you’ll never see, for buyers you haven’t met. We believe that no other channel in marketing can say that in earnest.
FAQs
-
What does it mean that AI models “trust the press”?
AI labs have signed licensing agreements with major news organisations, which means their content is treated as reliable training input. A brand’s own blog or press release, by contrast, rarely carries that same weight unless it’s picked up and repeated by credible third parties. -
Why does syndicated coverage matter more than a single placement? Models are trained to weigh consistency across independent sources. When a fact about your brand appears repeatedly across trusted, unrelated outlets, it’s more likely to “stick” and be repeated by the model later. A single, isolated mention is more likely to wash out during retraining.
-
What is “citation share”? It’s how often – and how favourably – AI engines like ChatGPT, Perplexity or Google’s AI Overviews reference your brand compared to competitors when answering relevant buyer questions. It’s being positioned as the AI-era equivalent of share of voice.
-
How is this different from traditional SEO? Traditional SEO optimises for search engine rankings and backlinks. This is about optimising for inclusion and favourable treatment in AI-generated answers, where the strongest predictor of visibility isn’t backlinks but the volume and consistency of branded mentions across the web.
-
Does paid media have any of this “memory” effect? No. Paid media stops working the moment the budget stops; it isn’t part of the web content that trains models, so it doesn’t compound over time. Earned coverage, once published and syndicated, can keep influencing AI answers indefinitely.
-
What should PR teams actually change about how they work? The core craft – building real stories, working with credible journalists, being patient – doesn’t change. What changes is the measurement conversation: teams need to track not just coverage volume but whether their brand is being cited, and how favourably, in AI answers.
-
Can you guarantee a story will be picked up and remembered by AI models?
No – this is a probabilistic system, and anyone promising guaranteed citations isn’t being straight with you. The realistic goal is to consistently improve your odds through substance, credible placement and message consistency.
About the author
Simarin Tandon | Junior Digital Account Director
Having worked with brands across the Beauty & Wellness, FMCG, FinTech, and Home & Lifestyle sectors, Simarin focuses on driving acquisition and growth, whilst managing the Digital team at brandnation.
A curious marketer, Simarin’s finger is always on the pulse when it comes to performance and digital updates across both paid and organic platforms.
