For most of the last two decades, a backlink meant one thing: a vote of confidence from one site to another, and the more of those votes a domain collected, the higher it tended to rank. That premise hasn't disappeared, but a second, separate signal has grown large enough to outweigh it in a specific, measurable way — whether AI systems trust a brand enough to name it in an answer, whether or not that answer includes a link at all.
An Ahrefs analysis across 75,000 brands found that brand mentions correlate with AI search visibility roughly three times more strongly than backlinks do — a 0.664 correlation for mentions versus 0.218 for links. Digital PR, unusually, produces both signals from the same placement, which is part of why 34% of SEOs now rank it as their single best-performing link-building method, nearly double guest posting's 18%.
Why entities matter more than links right now
Part of what's driving this is a change on Google's side, not just the AI platforms'. Google's February 2026 Discover update tightened how the search engine evaluates entity-level authority, now assessing expertise by content cluster rather than raw domain metrics — a site with deep, consistent coverage of one specific topic can outperform a broader, higher-authority site that only touches it in passing. AI systems apply a version of the same logic, with a sharper edge: they need to confidently identify what a brand actually does before they'll cite it, and a Wikipedia entry, a Wikidata record and complete schema markup can close that gap in days rather than the months a traditional authority-building campaign takes.
Where AI systems are actually finding brands to cite
The citation sources aren't evenly spread, and the split matters for where PR effort should actually go. The same Involve Digital analysis found that 68% of AI citations pull from third-party sources rather than a brand's own site, citing research across 500+ brands — inside that third-party share, Reddit accounted for roughly 22%, Wikipedia contributed about 19%, and review platforms like G2 and Capterra made up a meaningful slice of the rest. A separate synthesis of 680 million citations across six studies by 5WPR put Reddit specifically closer to 40% of citations sitewide — the single largest source across every major AI platform it tracked. The exact number moves depending on methodology and what's being measured, but every study lands in the same place: brand-owned content is a minority of what AI systems actually cite from.
It isn't one unified system, and that's the hard part
The complication most guides skip over: these platforms don't agree with each other. An analysis by Averi found only 11% of sites get cited by both ChatGPT and Perplexity for the same queries, and 40-60% of citations shift from one month to the next. Earning a citation in a publication ChatGPT trusts doesn't guarantee Perplexity or Gemini will ever surface the same piece. A digital PR program built around one placement type, or optimized for a single AI platform's apparent preferences, is building on ground that moves every few weeks.
What this actually looks like as a program
Muck Rack's State of AI in PR 2026 report found earned media accounts for roughly 84% of all AI citations, with paid and advertorial content responsible for well under 1% by comparison. Measurement has had to catch up with that shift: 66.2% of digital PR practitioners now track AI citations as a KPI, a metric that barely existed in last year's version of the same survey.
The mechanism behind most of this is unglamorous and repeatable: original data. Journalists need numbers to back up a story, and a business running any real volume of client or product work is usually already sitting on data worth publishing — response times, conversion patterns, seasonal trends, anything with a genuine methodology behind it. It's why 95.9% of digital PR practitioners now use data-led campaigns as their primary approach rather than pitching opinion pieces. Package the data once, pitch it to the right handful of publications, and the citation tends to follow the coverage rather than the other way around. When there isn't fresh proprietary data to hand, journalist-request platforms like Connectively, Qwoted and Featured are a lower-effort starting point — the modern, faster-moving successors to what HARO used to be.
This is the same logic behind our own digital PR work, and it overlaps more with GEO than either discipline used to acknowledge on its own. A link earns Google's trust. A mention earns the AI's. Increasingly, a single well-placed piece of coverage does both at once, which is exactly why the budget is moving here.
None of this requires Fortune-500 resources or a research team. It requires picking one genuinely interesting number from data you already have, being willing to publish it even if it isn't flattering, and treating the resulting coverage as infrastructure rather than a one-off campaign. The brands showing up in AI answers a year from now are mostly the ones that started that habit today, not the ones with the largest backlink count right now.