Blog · Third-party sources

Digital PR when the audience is a model

Published 10 July 2026 · Updated 8 October 2026 · 8 min read

Digital PR when the audience is a model
Photograph: Clo Art / unsplash

The short answer: Digital PR for answer engines is the work of getting one specific claim about you written, in someone else's words, on a page the engines can fetch and already retrieve. The link is a bonus. The sentence is the asset. If every piece of coverage you won this quarter carries your boilerplate instead of a reporter's own description of what you do, you have bought circulation and no corroboration.

That distinction sounds pedantic until you watch an answer get written. We have not found documentation from the engines we check describing how they weigh press coverage, so treat what follows as a working model rather than a documented one: an answer is assembled from passages, and a passage either contains a usable statement about you or it does not. Counting placements is a proxy for that, and a loose one.

The unit is a sentence, not a placement

A traditional PR report counts hits, reach and domain authority. None of those survive the trip into an answer. What survives is a short run of text that an engine can lift and stand behind: who you are, what you sell, who you serve, what you cost, what you were measured at.

So brief the claim, not the story. Decide the three sentences you want other people writing about you, keep them narrow enough to be checkable, and give journalists the material that makes writing them easy. A claim with a number, a date and a method behind it gives a writer something to restate. A claim built from adjectives does not, which is usually why it goes missing from the piece that runs.

This is also why the third-party sources AI engines keep citing matter more than the glossy placements. Pew Research Center's analysis of 68,879 Google searches made by 900 US adults during March 2025, with the result pages re-collected between 7 and 17 April 2025 and the piece published on 22 July 2025, found that Wikipedia, YouTube and Reddit together accounted for about 15% of the sources listed in the 12,593 searches that produced an AI summary, and that news sites accounted for about 5%. Government domains made up 6% of the sources cited in those summaries against 2% of the sources in standard results. That is one engine, one country and one month, and Pew notes the summary for a given search can change over time, so read the shape rather than the decimals before you spend a budget entirely on trade press.

Syndication looks like twenty sources and behaves like one

Here is the failure mode nobody puts in the coverage report. A press release picked up by twenty outlets is not twenty independent statements. It is one statement, copied. Collapsing near-duplicate passages is a standard technique in information retrieval, and the engines we check do not document whether their own pipelines apply it to syndicated text, so treat that as the likely mechanism and not a confirmed one. What you can say without guessing is narrower, and bad enough: you bought twenty mastheads and one sentence.

The check takes four minutes. Pull a distinctive twelve-word phrase from your last release, search it inside quotation marks, and count how many of the results are the same text under different mastheads. If that number is most of your coverage, your corroboration layer is thinner than your clipping book suggests.

The fix is unglamorous. Fewer releases, more briefings. One reporter who writes their own two paragraphs about you produces more usable material than fifteen outlets reprinting your third-quarter adjectives.

Independence is a property of the sentence, not the outlet

Wikipedia's notability guidance for organisations is the most explicit public statement of what independent coverage means, and it is useful well beyond Wikipedia. It discounts press releases, articles substantially rewritten from press releases, sponsored and contributor posts, routine funding or hiring briefs, inclusion in a list of a hundred companies, and quotations from company staff used as a news source. It also says plainly that one good independent source is almost never enough.

Read that list again as a PR brief. Much of what agencies deliver falls inside it. A quote from your founder inside an article about an industry trend is attributable to you, which means an engine that uses it is repeating a company statement, not a verified one. The same page with one line of the reporter's own description of your product is a different artefact entirely.

Two practical consequences. Ask for description, not just quotation: a sentence that says what you do, written by the journalist. And treat the trade listicle as what it is, cheap presence, not evidence. The same logic runs through review sites, G2 and what AI repeats about your product, where structured third-party text often outperforms a feature in a magazine.

The publisher's robots.txt is now part of your media list

This one surprises people. Your placement is only as retrievable as the site hosting it. OpenAI's crawler documentation is direct about the consequence: sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though the same page says they can still appear as navigational links. That rule applies to the publisher, not to you. It is also OpenAI's rule for OpenAI's crawler: other engines document their crawlers separately and we have not found an identically worded commitment from each of them, so check the one you care about rather than assuming the policy is shared. A hard-won feature on a title that blocks OAI-SearchBot, or puts the piece behind a wall, is not going to reach ChatGPT search answers no matter how good the sentence is.

So add two lines to the vetting you already do on a target outlet. Fetch https://<publisher>/robots.txt and look at how it treats OAI-SearchBot, GPTBot, Google-Extended and the Bing and Anthropic agents. Then check whether the article itself is reachable without a login. Google's own guidance on AI features says a page has to be indexed and eligible to be shown in Google Search with a snippet before it can appear as a supporting link, that there is no special markup or AI text file that buys the slot, and that meeting every requirement still does not guarantee the page is crawled, indexed or served. Access first, everything else after.

None of this means a paywalled outlet is worthless. It means the value is human attention, which is real, and you should stop counting it as machine-readable corroboration.

What to brief, and how to check it worked

Write a one-page fact sheet with the claims you want repeated, each with a date and a source anyone can open. Put the same facts on your own site in the same words, because an engine that reaches two versions of your pricing or your founding year has nothing to choose between them with, and the answers we see in that situation tend to hedge or skip the number. We have not measured how often that happens, so take it as a reason to keep the figures identical rather than as a quantified risk. Offer data a reporter can use without asking you twice.

Then measure the way you would measure anything noisy. Take ten questions a buyer actually asks, run each several times across more than one engine, and record three things: were you named, was the statement true, and which sources were cited. Run the set before the campaign and again three to six weeks after the coverage lands. Ten questions run a handful of times is a sample and not a census, so write down the dates and expect the noise. Watch the cited sources, not only the mentions, because the first thing successful digital PR changes is which documents the engine reaches for. Our self-check lists ten of these tests if you want a starting point that costs nothing.

One warning about reading the result. If your numbers move sharply in a week with no new coverage, suspect the engine before you credit your agency, and read what a model update does to your visibility first.

Honest limits

This post does not establish that the engines collapse syndicated coverage, only that near-duplicate collapsing is a standard retrieval technique and that we cannot see inside those pipelines to confirm it is applied to your press release. The opt-out consequence quoted above is OpenAI's, for OpenAI's crawler, and we have not verified an equivalent published statement for every other engine, so it is not a law of publishers. Attribution is the second gap: these surfaces mostly send no referrer we can match, so connecting a sentence in an answer to a pipeline number stays an argument rather than a measurement. We also do not know how long a corroborating article keeps paying once it ages out of an index's freshness preference, which is why we re-run our own audit prompt set, about 45 prompts across five engines, each month and watch the results move: that is one studio's question list, a sample rather than a census of anything. The uncomfortable limit is that coverage can carry an error, and a widely syndicated mistake about your pricing or your headcount is hard to retract, so if that has already happened to you start with when the model makes something up about you.

FAQ

Does a backlink still matter for AI visibility? It helps, mostly indirectly, because links influence what gets indexed and retrieved. The sentence around the link does more work than the link itself, so a nofollow mention with an accurate description beats a followed link inside a reprinted release.

How much coverage is enough? We have not measured a threshold and we are not aware of a published one. Our working view, formed from audit work rather than a controlled test, is that a few genuinely independent and descriptive articles do more than a large volume of syndicated material, and that the claim being consistent across them matters more than the count.

Should we pay for sponsored placements? They can be worth it for human readers. Treat them as advertising when you plan your visibility work, because they are usually labelled, excluded by Wikipedia's guidance from what counts as independent coverage, and rarely written in a reporter's voice. How any given engine weighs them is not something we can see.

How long until coverage shows up in answers? Indexing can happen within days. Whether an engine starts preferring that document is slower and less predictable, and we would be suspicious of anyone quoting a date.

Sources

Related reading

The third-party sources AI engines keep citingReview sites, G2 and what AI repeats about your productWhat a model update does to your visibilityWhen the model makes something up about youForums, communities and user content in AI answersHow to get a wrong fact about your brand corrected in AI answers