Blog · Third-party sources

News sites, publishers and the citation economy

Published 27 July 2026 · Updated 8 October 2026 · 7 min read

News sites, publishers and the citation economy
Photograph: Utsav Srestha / unsplash

The short answer: News and publisher pages are raw material for answer engines, not traffic partners. Engines lift a passage, name a source, and send almost nobody onward. For a brand, that makes a press placement useful only when the outlet is still readable by the engine's search crawler and the sentence naming you is quotable on its own. For a publisher, it makes the page itself the product, and the referral a rounding error.

Two groups read this question differently. Publishers want to know where the clicks went. Brands want to know whether PR still buys anything in AI answers. Both answers sit in the same mechanism, so it is worth looking at the mechanism before the tactics.

What the trade actually is

An answer engine needs text it can stand behind. Reported articles supply that: dated, edited, attributed, usually written in short declarative paragraphs. So engines retrieve them heavily and quote them happily. What they do not do is pass the reader along.

Pew Research Center's March 2025 study of 68,879 Google searches from 900 US adults found that users who saw an AI summary clicked a traditional result on 8% of visits, against 15% for visits with no summary. Clicks on the links inside the summary itself ran at 1%. Those numbers are from Google's summaries specifically, not from every assistant, and they are from a panel rather than from publisher server logs. Treat them as a direction, not a rate card.

Here is the part most coverage of that study skips. In the same dataset, news sites made up about 5% of the sources appearing in AI summaries, roughly the same share they held in the standard results underneath. News is not being promoted by the summary layer. It is being read, compressed, and left behind at the same rate it was before, with the click removed.

News loses, publishing wins, and they are not the same thing

Lump "publishers" together and the picture goes blurry. Separate them and it sharpens.

Dated reporting ages out of usefulness within days. A comparison page, a definition page, a best-of list or a reviews round-up does not, and those are the pages engines reach for when somebody asks a buying question. The trade outlets in a narrow sector tend to own that evergreen material, which is why a specialist title with 40,000 readers can out-cite a national masthead on a product question while losing to it on everything else.

That has a direct consequence for anyone planning coverage. A launch story is a one-week asset in the citation economy. A line about your product inside an outlet's long-lived comparison page is a multi-year asset, and it behaves much more like a review site entry than like press. The same logic explains why the third-party sources engines keep citing skew so heavily towards reference material rather than news.

The crawler split that decides whether a placement counts

This is the check almost nobody runs before pitching, and it is cheap.

The engines do not use one crawler. They use several, with different jobs, and a publisher can be opted out of one while fully open to another. OpenAI documents four user agents. GPTBot gathers content for model training. ChatGPT-User fetches a page because a person asked for it in the moment. OAI-SearchBot is the one that matters here, and OpenAI's own documentation is blunt about the consequence: sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Perplexity draws the same line, between PerplexityBot, which builds the index, and Perplexity-User, which fetches on a person's instruction.

So a newsroom that decided in 2023 to block AI crawlers on principle may now sit in one of three states. Blocked for training only, and still fully citable. Blocked for search, and effectively invisible in that engine's answers no matter how good the article is. Or blocked for both, in which case the article you worked three weeks to place cannot be retrieved when it would have mattered.

Run the check before the pitch, not after the placement:

  1. Open https://<outlet-domain>/robots.txt.
  2. Look for OAI-SearchBot, PerplexityBot, Google-Extended and ClaudeBot by name.
  3. Note which are disallowed and on which paths. Section-level blocks are common, so a

    4. general Disallow: / is rarer than a block on /archive/ or a paywalled subdirectory.

  4. Ask the engine a question the article should answer, and see whether that outlet appears

    5. at all. If the domain never surfaces for any question in its own subject area, the robots file usually explains it.

We have run this on pitch lists for a handful of clients and the split is never what the PR team expects. Outlets that publish loud opinions about AI are sometimes wide open; quiet trade titles are sometimes shut. The file is public, so there is no reason to guess.

If you are the publisher

Three controls are genuinely yours, and one of them is underused.

Preview length is a control. Google's documentation on AI features states that pages must be indexed and eligible for a snippet to appear in AI Overviews or AI Mode, and that nosnippet, data-nosnippet and max-snippet govern what can be previewed. Setting max-snippet:0 across an archive is a real opt-out of that surface. Many sites have it set on templates nobody has reviewed since 2019.

Markup is a control, and NewsArticle carries properties Article does not: dateline, printEdition, and backstory, which schema.org defines as a short explanation of why and how the piece was made, including methods and data sources. For investigations and data journalism, backstory is the cheapest way to state provenance in a form a machine can read without parsing your prose.

Original reporting is the third, and it is the only one that compounds. A number nobody else has is the one thing an engine cannot synthesise from four competing summaries, which is the argument in original data as a citation magnet. The economics of making peace with the click loss are a longer story, covered in AEO for publishers losing clicks.

If you are the brand

Shift the brief. Ask for the sentence, not the story.

Publicists are trained to win the headline. Engines do not read headlines preferentially, they read passages, and the passage that helps you is a plain sentence that names your company, says what it does, and attaches one verifiable attribute. "Citrel AI, a Bengaluru answer-engine-optimisation studio, audits brand visibility across five engines" survives extraction. "The firm is riding a wave of interest in AI search" does not, because it names nothing a retrieval system can match against a query.

Then check the outlet, as above. A quotable sentence inside a domain the search crawler cannot read is a sentence nobody will ever retrieve. Ten of the checks worth doing on your own properties are in the self-check, and this crawler test is the same idea pointed at somebody else's domain.

What we are not sure about

Licensing complicates all of this and we cannot measure it from outside. Several AI companies have signed content agreements with publisher groups, and the plausible effect is that a licensed archive is retrieved more readily on that company's surfaces. Nobody outside those companies can separate that effect from the fact that the publishers who sign deals are usually large publishers with deep archives, which engines favour anyway. We describe this as unresolved rather than quoting a premium.

The second gap is attribution. Most of these surfaces send no usable referrer, so a brand cannot connect a named mention to a pipeline entry, and a publisher cannot price what it lost. Anyone quoting a precise revenue figure for AI citation is modelling, not measuring.

FAQ

Is a press placement still worth anything for AI visibility? Yes, with two conditions. The outlet has to be readable by the engine's search crawler, and the sentence naming you has to stand alone without the surrounding story. A placement that fails either test is brand marketing, which is fine, but it is not AEO.

Should a publisher block AI crawlers? It depends which one. Blocking a training crawler and blocking a search crawler are separate decisions with separate consequences, and the documentation for both OpenAI and Perplexity says so plainly. Blocking search removes you from answers in that engine.

Why do trade publications get cited more than national ones on product questions? Because they hold the evergreen comparison and review material that buying questions retrieve. National titles win on events. Specialists win on "which one should I buy".

Does being quoted in the news make an engine repeat the claim? Sometimes, and not reliably. Corroboration across several independent sources moves an engine far more than one strong placement, which is why the slow work of consistent facts across many third-party pages outperforms a single campaign.

Sources

Related reading

The third-party sources AI engines keep citingReview sites, G2 and what AI repeats about your productAEO for publishers losing clicksOriginal data gives an engine something only you haveWhy Reddit gets cited so much, and what to do about it