Press coverage works on AI search as a reputation instrument, not a freshness instrument. The intuition says the opposite: search-grounded models can read this week's coverage, so press should move searched answers first. In OppAlerts' data, news presence predicts memory answers (Spearman rho 0.169) more strongly than searched answers (0.139), and 12 of the 13 memory-only models out-predict every search-grounded model on this signal, despite never having read the current news cycle.From the news analysis in the OppAlerts study, summarized in What Actually Correlates with AI Search Visibility; method and tables in The Research Behind This Guide, part of the AI Search Visibility research.
Coverage compounds into durable model belief instead of expiring with the news cycle. That single finding reframes the PR budget: the placement you earn this quarter is training next year's model, and the payoff clock runs in quarters, not days.
How news moves the answer
In the same data, news works as both a threshold and a gradient. Appearing in your category's news cycle at all is worth about a 1.45x lift in recommendation strength, and unlike Wikipedia, where presence is nearly everything, additional coverage volume keeps adding signal after the threshold. Presence gets you counted; accumulation keeps paying. That makes PR one of the few authority signals where sustained investment has data behind it rather than just habit.
The mechanism fits the reputation reading. Training corpora accumulate coverage over years, so a brand with a long press trail is heavily represented in what the model learned, whichever week you measure. A single fresh hit, by contrast, changes searched answers only if the story ranks for the queries the AI product actually runs, and most press does not. Judge PR by the trail it builds, not the spike it causes.
The standard caveats apply: this is correlational, famous brands attract coverage for the same reasons models recommend them, and no controlled experiment on press coverage exists that I know of. The mechanism is why I treat sustained category press as a lever anyway, and it ranks mid-pack behind backlinks and Reddit in the independence tests.
Finding your gatekeepers empirically
When search-grounded models do credit third parties, the credited list is short and repetitive. News, review, and analyst sites get credited at two to four times their retrieval share, the inverse of Reddit's treatment. And one publisher is nearly universal: Forbes appears in the credited citations of all five search-grounded models, across 51 of 100 industries. A short list of Forbes-class publishers recurs so consistently that the study calls them gatekeepers.
Do not take the general list on faith; derive yours. Run your category's prompts with search enabled across the models you care about, collect every credited citation, and count domains. The publishers that recur are your category's gatekeepers, and that empirical list, usually a dozen domains or fewer, is the pitch target list. It rarely matches the media list a PR agency starts from.
Earning your way in is classic media relations with one adjustment: the gatekeepers worth the most are the ones that write category roundups and comparisons, because those pages get retrieved by recommendation prompts as well as credited. A profile piece flatters; a slot in the gatekeeper's "best X" coverage recommends. Links: The Signal That Refuses to Die covers why those pages are retrieval targets in their own right.
Citation-gap analysis
The highest-yield off-site exercise in this Part is a simple set difference. From the citation harvest above, list the sources the models credit when recommending your competitors. Mark the ones where you have no presence. That list is pre-qualified: every domain on it is already retrieved, already trusted, and already shaping answers in your category, so coverage there does not need a trust argument, only a pitch. Work it in order of how often each source is credited, and re-run the harvest quarterly, because the credited list drifts as the underlying indexes and models change. The competitive version of this exercise, including how to read a competitor's whole citation trail, is in Competitive Analysis.
Research-led PR: earn the link now, train the model later
The pitch that clears a gatekeeper's bar most reliably is original data, and Evidence, Expertise, and Original Research covers how to build those assets. What matters here is what the resulting coverage should contain, because not all coverage teaches a model equally.
Write and pitch for absorbable facts. A story that says "Acme's analysis of 40,000 service calls found repair costs rose 12% in two years" gives the model a fact with your name attached to your category's subject matter. A story that mentions your brand in passing, with no claim attached, is brand-name confetti: it still counts toward presence, and presence has real value at the threshold, but it teaches the model nothing to repeat. The difference between the two is set in the press release and the pitch, before any journalist writes a word. State the finding in one sentence, put the brand name inside that sentence, and make the category vocabulary explicit, because sentences are what both retrieval and training actually consume.
One more reason the gatekeepers matter more every quarter: AI summaries are absorbing the clicks that used to reach publishers directly. Pew found Google users clicked a result link in 8% of visits when an AI summary was present, against 15% without one.Pew Research Center, analysis of browsing data from 900 U.S. adults, March 2025. Coverage is becoming less about the referral traffic a placement sends and more about what it teaches the systems composing the answer. Budget for it on that basis, and measure it on the memory clock: the visibility effect of this quarter's coverage is properly judged after the next model generation ships, a lag Freshness, Cutoffs, and Temporal Accuracy explains in full.