Why AI answers lean on publications, and how your brand gets into them

When a client asks ChatGPT, Gemini or Perplexity which firm to trust with a decision, the answer is put together in seconds from a handful of sources the assistant finds at that moment. Whether the answer is right depends mostly on which sources it finds. In a 14-day test of six AI chatbots on 2,100 questions drawn from same-day BBC reporting, researchers led from Stanford University found that “retrieval, not reasoning, failures drive over 70% of all errors”. “When models retrieve a correct source, they often extract the correct answer,” they wrote; “the problem is to land on the right source in the first place”[1].

So the question for any brand is which sources AI systems land on, and why. Our guide on how AI answer engines decide which sources to cite covers what the platforms say about access and ranking. This guide looks at why publications keep winning, and what that means for your brand.

What AI answers cite

A short list of established outlets. Kai-Cheng Yang of Northeastern University studied 366,087 citations from real conversations on AI Search Arena, a site where users compare AI search answers. News made up 9% of all citations, and 19.1% for OpenAI’s models. When the models cited news, a few outlets took most of it: Reuters accounted for 22.8% of OpenAI’s news citations and AP for 12.2%, and the 20 most cited outlets for 67.3%. Low-quality sources were rarely cited: 96.2% of the news sources OpenAI’s models cited were rated high quality, against 92.2% for Google’s and 89.7% for Perplexity’s[2]. The data covers seven weeks in 2025 and a self-selected group of users, so read it as a snapshot.

More so when the question is about the news. The UK think tank IPPR ran 100 news queries through four AI tools in late 2025 and found that “journalism accounted for nearly 40 per cent of citations”, drawn from “a narrow range of prominent news brands”. The Guardian was a source in 58% of ChatGPT’s answers, and Google’s AI Overviews used the BBC for 52.5% of the news queries they answered[3].

Earned, not paid. In Muck Rack’s May 2026 analysis of the links cited by ChatGPT, Claude and Gemini, about 84% of citations came from earned media and about 27% from journalism, while “paid and advertorial content accounts for just 0.3% of all citations”[4]. Muck Rack sells PR software, so the finding suits its business, but its three editions since July 2025 have told the same story. An advert does not buy a place in the answer.

Why AI companies pay for journalism

The clearest sign of what AI companies value is what they pay for, and their announcements use the same words again and again:

• OpenAI’s deal with Axel Springer was to add “recent and authoritative content” to ChatGPT, with “attribution and links to the full articles”[5].
• The Financial Times’s chief executive said “it’s clearly in the interests of users that these products contain reliable sources”[6].
• Announcing its deal with The Washington Post, OpenAI said it was investing in “high-quality journalism” so that users “get timely, trustworthy information when they need it”[7].
• Perplexity says “our ability to provide high-quality answers hinges on trusted, accurate sources”, and shares revenue with publishers whose content its answers reference[8].
• Microsoft, launching a marketplace for licensing publisher content in February 2026, said its own testing with Copilot “shows that premium content meaningfully improves response quality”[9].
• Google said on 15 October 2026 that more than 400 publications now take part in its News AI programme, through which “we pay participating news publications”[10].

These are the companies’ own words, and they do not prove that licensed sources rank higher. IPPR found that ChatGPT’s top news citations often matched OpenAI’s publisher deals, but said that without transparency on how sources are ranked it could not tell how much of an advantage those publications get[3]. OpenAI says “any website or publisher can choose to appear in ChatGPT search”[11], so a licence is not the price of being cited.

The same preference runs through what the models learned from. In C4, a widely used training corpus built from the web, researchers found Wikipedia and news sites such as The New York Times, the Los Angeles Times and The Guardian among the most represented websites[12]. The News/Media Alliance, a publishers’ trade body that argues this use of journalism infringes copyright, found that curated training sets “overweight publisher content by a factor ranging from over 5 to almost 100” compared with the open web[13].

Why publications are easier to trust

AI gets things wrong, and borrows credibility. When Columbia University’s Tow Center asked eight AI search tools to identify the source of excerpts from news articles, they “provided incorrect answers to more than 60 percent of queries”. The researchers noted that “the credibility of the publishers is often used to boost the trustworthiness of a chatbot’s brand”[14]. In a study led by the European Broadcasting Union and the BBC, journalists from 22 public service media organisations found that 45% of AI answers about the news had at least one significant issue. Sourcing was the biggest cause, with significant problems in 31% of answers[15]. An assistant that cites a weak source pays for it in accuracy; one that cites a strong source borrows its reputation.

Platforms say they weigh reputation. Microsoft’s explanation of how Bing ranks results says “a well-known news site is considered to have a higher reputation than a brand-new blog”[16]. Its search team wrote in May 2026 that AI answers need “discrete, supportable facts with clear provenance”[17]. Perplexity says its answers are “supported by citations from reputable news organizations, academic publications, and established content sources”[18].

Publications carry signals machines can check. Google’s policies for news sources ask for “clear dates and bylines”, information about the authors, the publication and the publisher, information about the company behind the content, and contact information[19]. These are Google News policies, not a formula for AI answers, but they describe what an edited publication already provides and many brand blogs do not: a named author, a date, sources, corrections and a publisher who answers for the page.

Readers check the source too. In the Reuters Institute’s 2025 survey of six countries, 54% had seen an AI-generated answer to a search in the past week, and many said they verify what they read, “especially for health or politics, treating AI as a first pass before checking non-AI sources”[20]. In its 2026 Digital News Report, a third of people who use chatbots for news said they ask them “to assess the reliability of a news source”[21]. The publication behind a citation is where the reader decides whether to believe it.

What does not work on its own

One big placement. Seer Interactive, a marketing agency, found almost no relationship (a correlation of 0.07) between a brand’s coverage on big news sites and how often ChatGPT mentioned it from its training data, without web search. “Simply being featured in Forbes or CNBC doesn’t guarantee your brand appears in ChatGPT’s answers,” it concluded[22]. The test predates live search, but the lesson holds: one article in a famous title is not a switch.

Links without mentions. In Ahrefs’ study of 75,000 brands, mentions of a brand across the web correlated far more strongly with its visibility in Google’s AI Overviews (0.664) than backlinks did (0.218)[23]. A follow-up found branded web mentions correlated at 0.66 to 0.71 with visibility in ChatGPT, AI Mode and AI Overviews[24]. Ahrefs sells the tool it measured with, and correlation is not cause, but the direction is consistent: be discussed, widely and accurately, in the places AI systems read.

How to get your brand into the publications AI reads

1. Ask the questions your clients ask

Type the questions your clients would ask into ChatGPT, Gemini, Perplexity and Google, and note which publications the answers cite. Each assistant has its own habits: in Muck Rack’s analysis, ChatGPT’s most cited domain was Wikipedia, Claude’s was PubMed Central and Gemini’s was Reddit[4]. The publications that keep appearing for your sector are your target list.

2. Give journalists something worth citing

Coverage starts with something new: original data, a named expert who can explain a change, a clear view on the news of the week. Our guides on how to become the expert your industry’s media quotes and how to write a press release that journalists and AI assistants pick up set out how.

3. Publish under a named expert in your sector’s titles

A contributed piece puts your expert’s thinking inside a publication that already carries the signals machines check: an author page, a date, linked sources and a correction policy. Choose titles that source every figure, because a claim with its source attached is what an AI system can use.

4. Give interviews that become a written record

A long interview published with a written feature or a transcript gives AI systems text to cite, under the publication’s name rather than yours. Keep it regular: Muck Rack found that 57% of journalism citations with a known date were published within the previous 12 months[4], so a steady rhythm of coverage beats one big feature every few years.

5. Keep your facts the same everywhere

AI systems assemble answers from several sources and do not always tell good from bad. Finland’s public broadcaster Yle, one of the organisations in the EBU study, called it “a concerning trend” that the assistants “treat all sources as equally reliable”[15]. Make sure your legal name, registration and licence details, founders, locations and key figures match across your website, your directory profiles and every article about you. Our guide on how to build a website that earns trust at first glance covers your own site.

6. Label paid content, and make it worth citing

Labelled partner content puts your facts, reviewed by you, on a trusted publication’s domain. On its own it is a small share of what AI cites, so treat it as a complement to coverage, interviews and expert bylines, not a substitute. Choose publications that hold partner pieces to the same factual standards as their journalism, and make sure the label is clear every time. Our guide on what regulators and search engines now expect as proof of trust sets out the rules.

7. Check again every month

Answers change as models and sources change. Re-run your questions every month, note which publications are cited and whether your brand appears, and adjust where you pitch and publish. Before you buy a placement, ask the publisher the questions in our guide on what zero-click search means for brands buying media.

A checklist

1. List the questions your clients ask an AI assistant about your sector.
2. Run them in ChatGPT, Gemini, Perplexity and Google, and record the publications cited.
3. Make those publications your target list.
4. Offer them something new: data, a named expert, a clear view.
5. Publish under your expert’s name in titles that source every figure.
6. Give long interviews that are published as text as well as video.
7. Keep coverage regular, because recent articles are cited most.
8. Make your legal name, licences, people and figures identical everywhere.
9. Use labelled partner content alongside coverage, never instead of it.
10. Choose publications with bylines, dates, sources and a corrections policy.
11. Re-run your questions every month and adjust.