Search for the best learning-management platform for selling courses, and Google’s AI Overview will cite a page published by Oasis LMS. Then it will recommend Kajabi, Thinkific, and the other rivals named inside that very page. Oasis supplied the source. Its competitors took the sale.
This is not an edge case. Lily Ray’s analysis of 100 B2B “best [category]” queries found that when a brand’s own ranking listicle was cited as a source, the publishing brand was omitted from the recommendation itself 69% of the time. Citation is attention; recommendation is the sale. AI search has quietly severed the two, and a great deal of marketing budget is funding the wrong half.
For a decade, corporate communications rested on two assets: the channel you own and the placement you pay a premium to win. AI-led discovery has repriced both. Your own website is now close to the least trusted thing a model can find about you, and the marquee hit no longer does the work it once did.
Focus On: What the machine actually trusts
Strip away the tactics, and one finding dominates the research. Coverage from credible third parties is the strongest predictor of whether an AI system mentions you at all. The University of Toronto’s controlled study across ChatGPT, Perplexity, and Gemini found a systematic and overwhelming bias towards earned media — independent, authoritative sources — over brand-owned and social content, a sharp departure from Google’s more balanced mix. In its Canadian set the split ran roughly seven to three in favour of earned media, with social content almost entirely absent.
The correlation work points the same way. Ahrefs, across 75,000 brands, found branded web mentions predict AI visibility roughly three times more strongly than backlinks, and that ChatGPT shows little relationship with the domain-authority scores that ran the old search economy. A separate Muck Rack review of more than a million AI prompts put the earned-media share of citations above 85%. The signal isn’t the authority of your own domain; it’s how often, and in how many trusted places, the rest of the web is found discussing you.
None of this rewards self-assertion. The Princeton team that named this field tested nine ways to make content more visible to generative engines. An authoritative tone, unsupported by evidence, moved nothing. Pointing to verifiable proof — citing credible sources, showing the data — lifted a lower-ranked page’s visibility by more than 100%.
A word of caution on the chorus. The loudest voices declaring earned media the new sovereign are, cui bono, the PR and “GEO” firms that sell it. The independent evidence supports the direction; the precise figures, and the vendor enthusiasm around them, deserve the scepticism a young discipline has yet to earn its way out of.
When I argued in December, in an issue on AI-legible go-to-market, that authority would be demonstrated through clarity rather than asserted through budget, I was half right. Clarity is necessary; it is no longer sufficient. I also underplayed something the Toronto researchers named directly: an inherent big-brand bias. The systems that punish self-promotion still lean towards established names, because frequent mention reads to a model as credibility. The levelling many of us forecast for challengers is real at the margin and overstated at the centre. To borrow Lampedusa: everything has changed so that the largest incumbents can stay exactly where they are.
What this changes on Monday
Three things follow for anyone who owns the marketing and communications budget.
First, the metric. Reach and prestige were always proxies for influence; mention-density across credible, independent sources is now the thing itself. Track your share of recommendation — how often the machine names you when a buyer asks, including in queries about your rivals — not only the citation counts teams have started to chase.
Second, the budget. The single marquee placement and the high-volume owned blog are now the two least efficient lines in the plan, for opposite reasons: one is credible but solitary, the other voluminous but self-interested. Move spend towards distributed earned presence — trade press, independent reviewers, and the communities and video channels these models actually cite — and towards giving genuinely useful, verifiable material to people who will repeat it.
Third, the function. If a model reads your rivals’ comparison pages and recommends whoever the web names most, your communications team has a new remit: to make your firm the name others are found using, and to make your own pages the clear, sourced reference a model will quote rather than discount.
So the question for your next planning cycle is blunt. If visibility now depends on who corroborates you, what is the spend on channels only you control actually buying?
The model has already stopped taking your word for it. The only decision left is how long you keep paying to talk to yourself.
Follow me
That’s all for this week. To keep up with the latest in generative AI and its relevance to your digital transformation programs, follow me on LinkedIn or subscribe to this newsletter.
Disclaimer: The views and opinions expressed in Chronicles of Change and on my social media accounts are my own and do not necessarily reflect the official policy or position of S&P Global.