Updated on 19 October 2026, by João Leite, Founder of murmur.

Press relations help you appear in ChatGPT when the article runs in an outlet that the AI reads and uses as a source. In June 2026, news and media accounted for 8.3% of ChatGPT citations and 14.0% of Claude’s. The calculation is by Otterly, which sells AI monitoring, made with Press Ranger and with questions from the United States. The weight changes from one AI to another, and from month to month, and the press release barely shows up in a purchase question.

  • The press enters AI answers with a different weight in each one. Claude is the AI that cites news the most in Otterly’s count: 14.0% of citations in June 2026, in questions from the United States.
  • In Otterly’s count, the niche outlet appears more than the national press. It accounted for 75.8% of news citations, against 24.2%.
  • The press release barely shows up in a purchase question. In Loganix’s 100 purchase questions, no press release domain appeared. In Muck Rack’s overall calculation for February 2026, the sites that distribute press releases accounted for 1% of citations.
  • The article pays off more in the first weeks. In ChatGPT, the citation drops after about a week, and Claude is more likely to cite texts from the last ten weeks, according to Muck Rack.
  • The article can be checked in the sources of the answers. The same questions from your market, asked before and after it, show whether it became a source and whether the answer started citing the company.

murmur, a GEO agency in Brazil that places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents, studies each client’s sector, produces the content that fills the brand’s gaps and monitors the answers every week.

What is measured about press relations and AI

It is a measured fact that AI models cite the press. The 2026 counts find news sites among the sources of ChatGPT, Claude, Google AIO and Perplexity, in very different shares. Google AIO, or AI Overview, is the summary that Google’s AI writes at the top of many searches, before the links.

One study tracked the article distributed to outlets: the study by Stacker, which sells the distribution of brand stories to partner outlets. In March 2026, Stacker and Scrunch tracked 87 stories from 30 clients for 30 days, across more than 2,600 questions to 8 AI models. In their calculation, 97% of the distributed stories had at least one AI citation, against 82% of the content on the brand’s own site.

The mention of the company on other sites also goes together with the brand appearing in the answers. For this relationship, Ahrefs, which sells SEO tools, looked at 75 thousand brands in December 2025, and its calculation is in why AI recommends some companies and not others. The practical question is whether the article runs in an outlet that the AI uses as a source on your subject.

The press weighs differently in ChatGPT, Claude, Google AIO and Perplexity

The weight of the press changes from one AI to another: in Otterly’s June 2026 count, Claude was the one that cited news the most. Otterly counted, with Press Ranger, a public relations platform, 129.3 million citations from seven AI models in June 2026. The questions were from 16 sectors in the United States.

AIin favor of the press, with the dateagainst, or with a caveat, with the date
ChatGPTJune 2026: news and media in 8.3% of citations, and 48% more citations per cited page for the outlet with a licensing agreement with OpenAI (Otterly with Press Ranger)September 2026 study, with citations from July and August: news and media fall from 21.3% to 10.4% of citations after the changes of 8 and 14 August (Otterly); September 2026: among the 50 most cited domains, the Wall Street Journal, Forbes and Reuters take 0.10% to 0.15% each, and TechRadar, a technology site, 0.23% (Surfer)
ClaudeJune 2026: the AI that cites news the most, with 14.0% of citations (Otterly with Press Ranger); May 2026: it prefers smaller, niche outlets (Muck Rack)May 2026: it shows a source in 55% of answers, against 96% for ChatGPT (Muck Rack)
Google AIOJuly 2026 study, with citations from October 2025 to July 2026: the New York Times, which blocks AI bots, is still cited 8,007 times in AI Overviews (Goodie)June 2026: the licensing agreement gives the outlet no advantage (Otterly with Press Ranger)
PerplexitySeptember 2026 study, with citations from July and August: news and media go from 22.7% to 23.4% of citations (Otterly); July 2026 study, with citations from October 2025 to July 2026: the blocked New York Times is still cited 4,303 times (Goodie)June 2026: the licensing agreement gives no advantage (Otterly with Press Ranger); March 2026: no press release domain among 100 “melhor X em Y” (“best X in Y”) questions (Loganix)

In ChatGPT, news accounted for 8.3% of citations in June 2026. The outlet with a licensing agreement with OpenAI had 48% more citations per cited page than the outlet without an agreement, 10.2 against 6.9. At the turn of August, ChatGPT started citing Reddit at least 73% less, according to Otterly. Reddit and Quora are the subject of Reddit and Quora to appear in AI models. In another set of questions, Otterly tallied 6.1 million citations from seven AI models, across 17 sectors in the United States. The questions ran every day in July and August 2026. In that count, news and media fell from 21.3% to 10.4% of ChatGPT citations after the changes of 8 and 14 August.

In September 2026, the Wall Street Journal, Forbes and Reuters were among the 50 domains most cited by ChatGPT, with 0.15%, 0.13% and 0.10% of citations, and TechRadar, a technology site, in seventh, with 0.23%. The count is by Surfer, which sells an SEO tool. The 50 domains add up to only 8.4% of citations, across 1.2 million answers to questions from its users.

In Claude, news was 14.0% of citations in June 2026, about 1.7 times ChatGPT’s share. In May 2026, Muck Rack, which sells public relations software, wrote that Claude prefers smaller, niche and specialized outlets. By contrast, Claude shows a source in 55% of answers, against 96% for ChatGPT, in Muck Rack’s count. When it does show one, it cites on average 13 sources, against 5 for ChatGPT.

In Google AIO, the AI Overview at the top of Google searches, the press is still cited even when it blocks AI bots. AI bots (crawlers) are programs that read sites for the AI companies, such as OpenAI’s GPTBot. Goodie, which sells AI visibility software, counted 8,007 citations of the New York Times in AI Overviews, from October 2025 to July 2026. The newspaper blocks AI bots. In Otterly’s count, the licensing agreement gave the outlet no advantage in AI Overviews.

In Perplexity, Goodie counted 4,303 citations of the New York Times, the newspaper that blocks AI bots. In Otterly’s July and August count, the news and media share held firm in Perplexity. It went from 22.7% of citations before the ChatGPT change to 23.4% after. In Otterly’s June count, the licensing agreement also gave no advantage in Perplexity. What Perplexity cites beyond the press, and why, is the subject of how to appear in Perplexity.

In Brazil, the press share changes with who is counting, and the count by type of site is the subject of which sites most influence ChatGPT in Brazil. Ranqia, which sells a GEO service, counted editorial media in 10.6% of ChatGPT citations to unbranded questions, between December 2025 and July 2026. murmur measured 51 niche sectors in September 2026, with 25 purchase questions asked once in each AI. In those sectors, the press was 15.2% of Claude’s sources, 2.5% of Google AIO’s and 1.1% of those that ChatGPT showed with a link.

The weight of the press ranges from 1% to 27% of citations, depending on what you count

The weight of the press ranges from about 1% to 27% of AI citations, depending on what each study calls press. The 37 major news domains in the United States and the United Kingdom that Goodie tracks landed between 1.3% and 1.9% of citations each month. Otterly counted news and media in general, and reached 7.2% of the 129.3 million citations in June 2026.

Muck Rack counted journalism in 27% of the sources cited by ChatGPT, Claude and Gemini in May 2026. The sample is over 25 million links, across 17 sectors, and the journalism cited came from more than 20 thousand outlets. In the three editions of the study, since July 2025, that share was between 25% and 27%. The same Muck Rack puts 84% of citations in what it calls earned media, which others publish without the company paying, and the page does not say what went into that count.

Each press number above holds for the definition, the country, the AI models and the month of whoever counted it. For your market, the count that matters comes from the sources of the answers your buyer receives.

Niche outlets and articles that help people choose appear more than the news of the day

In Otterly’s count, niche outlets appear more than the national press. Articles that help people choose appear more than the news of the day. In June 2026, niche and sector outlets took 75.8% of news citations, against 24.2% for the national press. They won in 15 of 16 sectors. In outlets with a licensing deal, 58.8% of citations went to text that stays useful for months: best-of lists, buying guides and product tests. Dated news took 5.1%.

The type of question also changes how much the AI cites journalism. At Muck Rack, the question about a sector trend pulled journalism citations more than twice as often as the “how to” question. For anyone planning the editorial calendar, this points to articles with sector data, comparison and selection criteria, in an outlet the buyer reads. Proprietary data gives the article that angle, and the path is in the guide on proprietary research to appear in ChatGPT.

Press releases and press release distribution appear little in AI answers

The press release, a statement the company writes for the press, appears little in AI answers, and almost only in questions about a sector trend. In the February 2026 update, Muck Rack counted the sites that distribute press releases (PR Newswire, Business Wire and GlobeNewswire) in 1% of citations. In the July 2025 study, it was 0.2%. In the May 2026 edition, the press release appeared 3.5 times more in answers about a sector trend than in “best” questions.

In a buying question, the press release almost never appears. Loganix asked 100 “best X in Y” questions, from 10 sectors, to Perplexity, ChatGPT and Gemini in March 2026, and no press release domain appeared. The recommendations came from editorial lists, review sites and niche comparison sites, such as G2 and Clutch. The weight of reviews is in the guide on reviews and Reclame Aqui in AI answers. Loganix, which sells SEO and link building, writes that the press release helps the AI answer about the company itself, not recommend it. That is its explanation, with no count to support it.

An article pays off more in the first weeks, and Claude cites it for longer

An article pays off more in the first weeks after publication, and Claude keeps citing it for longer than ChatGPT. In Muck Rack’s February 2026 figures, the citation rate was higher in the first week after publication. More than half of citations were of texts from the last 12 months. In May, Muck Rack wrote that ChatGPT drops off after about a week, and that Claude’s peak is the last ten weeks. Claude has about three times ChatGPT’s chance of citing a text from two to four weeks ago.

With citations that age in weeks, press presence calls for frequent new articles. The timeline for a new page, from going live to the first citation, is the topic of how long it takes to appear in ChatGPT.

How to choose the outlet: the ones the AI already cites in your market and can read

Choose the outlet among those the AI models already cite in your market, and check whether its site lets AI crawlers read. Put your buyer’s questions to ChatGPT, Claude and Google AIO, open the sources and note the outlets that repeat. The script for that study, question by question, is in when the AI cites your competitor and not your company.

The list of outlets that comes out of AI answers may be different from the press office’s. In December 2025, Muck Rack compared the journalists who receive the most pitches through its platform with those the AI models cite the most for each brand. The overlap was 2%, on average.

Blocking an AI’s search crawler removes the outlet from its answers, and blocking the training crawler does not. In 31 million citations, Goodie found the New York Times zero times in ChatGPT and Gemini, and once in Claude. The newspaper blocks the crawlers of those three AI models. In Brazil, blocking is rare: 93% of 4,025 journalism sites had no specific rule against AI crawlers. The number is from the study The Protocol Gap, reported by Núcleo Jornalismo in March 2026. What each crawler does is in the guide on the OpenAI and Anthropic AI crawlers.

How to know whether the article entered ChatGPT’s answers

To know whether the article entered, put your market’s questions to ChatGPT, Claude and Google AIO before and after it. Open the sources of each answer and see whether the article’s URL appears and whether the answer text includes the company name. An article can become a source without the answer citing the company, and the two things are counted separately.

Also count brand presence. Presence is how many answers the brand appears in, divided by the total number of answers. Ask each question in a new conversation and more than once, because the answer changes each time. The spreadsheet and the measurement frequency are explained in how to measure brand presence in ChatGPT.

murmur writes the buyer’s pages and shows what changed in the market’s answers

murmur writes the pages that answer each client’s buyer’s questions. The pages sit on a subdomain with the client’s brand. The content is all the client’s, even after the contract ends. From the diagnostic onward, the same person follows the brand and studies its sector. Four clients have a published case study with their name: Mindo, a motion design studio from São Paulo; Visio, operations software for retail; FotoRegistro, a photo printing e-commerce; and Fly Vet, a veterinary marketing agency.

At murmur, ChatGPT, Claude, Perplexity and Google AIO receive the same questions from the client’s market every week. The dashboard included in the service puts the brand’s presence and the competitors’ side by side. This ongoing monitoring shows what changed in the answers of the company’s market, whatever the reason for the change. It is what is missing when someone publishes an article and does not rerun the questions.

Frequently asked questions

Are a press office and digital PR the same thing?

Press relations and digital PR overlap, and the name changes depending on who is selling. Press relations puts the company in front of journalists and outlets, with a story pitch, a press release and an interview. Digital PR is the name marketing gives to getting the brand onto portals and websites, often alongside link building. For AI models, in both cases, what counts is the text published on a site they read.

Does a press release help you appear in ChatGPT?

A press release helps little with appearing in ChatGPT. In Muck Rack’s February 2026 count, sites that distribute releases accounted for 1% of citations. In Loganix’s 100 buying questions, from March 2026, no release domain appeared. Wikipedia also does not consider a press release an independent source. What that changes for an entry is in the guide on whether Wikipedia helps you appear in ChatGPT. The release serves to get the article written, and Loganix’s recommendations came from editorial lists, review sites and niche comparison sites.

Does sponsored content count for AI?

Sponsored content counts little for AI and, without identification, puts the site at risk. Paid content and advertorials, paid text published to look like an article, added up to 0.3% of citations in Muck Rack’s May 2026 count. For Google, a paid link without the sponsorship marking is spam, and the CONAR code asks that paid coverage be identified. Where the line falls between what is allowed and what is not is in the guide on manipulating what ChatGPT recommends.

A mention of the company name on other sites goes together with appearing in AI answers. In Ahrefs’ study of 75 thousand brands, from December 2025, mentions on other sites accompanied appearance in ChatGPT and in AI Overview. Ahrefs does not say whether the mention it counted carries a link, and the number of links to the site accompanied far less. Links to the site, backlinks, have their weight counted AI by AI in the guide on backlinks for appearing in AI. The mention has to be real.

Is it better to appear in a big outlet or a niche one?

In Otterly’s June 2026 count, the niche outlet in the sector appeared more in AI answers than the national outlet. It was 75.8% of news citations, against 24.2%, in questions from the United States. Muck Rack writes that Claude prefers smaller, specialized outlets. The best outlet for your company is the one AI models already cite in your buyer’s questions.

Where to start

Start with the list of outlets that AI models already cite in your market’s questions, before choosing the story pitch or the press relations agency. To see what AI models already answer in your market, use murmur’s free diagnostic. It asks only for the site address, with no sign-up, and shows the result in less than two minutes. murmur’s sector studies show, in one round, what AI models answered to the buying questions of each niche sector. Most also include the sites that AI models used as a source.

Read also

References

  • Otterly, Rick Tousseyn, “News Publishers With OpenAI Deals Get +48% More AI Citations on ChatGPT”, updated on 21 August 2026, produced with Press Ranger. It covers 129.3 million citations from June 2026 across seven AI models (ChatGPT, Google AI Overviews, AI Mode, Perplexity, Copilot, Gemini and Claude), with questions from 16 sectors in the United States and 91 licensing deals between AI companies and news outlets. Otterly sells AI search monitoring, and Press Ranger is a public relations platform. https://otterly.ai/blog/ai-news-licensing-deals/ (accessed on 6 October 2026).
  • Otterly, Thaylise Nakamoto, “ChatGPT cut Reddit citations by at least 73% in August 2026. Here is what replaced it.”, August 2026. It covers 16 brand reports from 14 sectors in the United States market, comparing citations from 6 to 13 August with those from 14 to 17 August 2026. https://otterly.ai/blog/chatgpt-reddit-citations/ (accessed on 6 October 2026).
  • Otterly, Rick Tousseyn, “The AI Search Index: What ChatGPT, Gemini, Perplexity and Google Actually Cite in 17 Industries”, updated on 8 September 2026. It covers 6,095,520 citations across seven AI models, with the same buyer questions asked every day in July and August 2026, in 17 sectors in the United States; the calculation compares before and after the ChatGPT changes of 8 and 14 August. Otterly sells AI search monitoring. https://otterly.ai/blog/ai-search-index/ (accessed on 7 October 2026).
  • Muck Rack, “Earned media still drives 84% of AI citations. Here’s what that means for PR.”, 7 May 2026. More than 25 million links cited by ChatGPT, Claude and Gemini, across 17 sectors, in the third edition of the study “What Is AI Reading?”; the page gives neither the country nor the number of questions and does not define earned media. Muck Rack sells public relations software. https://muckrack.com/blog/what-is-ai-reading-may-2026 (accessed on 6 October 2026).
  • Muck Rack, “Updated for 2026: ‘What Is AI Reading?’ a study of 1 million+ links cited by AI models”, 16 February 2026. More than 1 million links cited by AI models; the page compares it with the July 2025 study. https://muckrack.com/blog/what-is-ai-reading-new-insights (accessed on 6 October 2026).
  • Muck Rack, “3 new insights on how Anthropic’s Claude cites media from Muck Rack’s ‘What is AI Reading?’ report”, 6 May 2026. The page gives neither the size of the dataset nor the method. https://muckrack.com/blog/how-claude-cites-media (accessed on 6 October 2026).
  • Muck Rack, press release “Earned Media Still Drives Generative AI Citations as Press Release Visibility Grows”, Miami, 2 December 2025, distributed by GlobeNewswire and read in the copy published by the Manila Times on 3 December 2025. More than 1 million links cited by ChatGPT, Claude, Gemini and Perplexity, from July to December 2025; the release does not say how many brands are included in the journalists’ comparison. https://www.manilatimes.net/2025/12/03/tmt-newswire/globenewswire/earned-media-still-drives-generative-ai-citations-as-press-release-visibility-grows/2235613 (accessed on 6 October 2026).
  • Surfer, “The 50 Most-Cited Domains in ChatGPT (September 2026)”, updated on 6 October 2026. It covers 47,213 Surfer user questions and 1,229,618 ChatGPT answers, from 1 to 28 September 2026, with about 5.73 million citations; the Wall Street Journal has 0.15% of citations, Forbes 0.13% and Reuters 0.10%, and TechRadar, a technology site, appears in seventh place with 0.23%. The page notes that, because they come from Surfer users, the questions skew toward marketing, SaaS and a few Polish domains. Surfer sells an SEO tool. https://surferseo.com/blog/most-cited-domains-chatgpt/ (accessed on 6 October 2026).
  • Goodie, Mostafa ElBermawy, “AI Citations & News Publishers: 2026 Study”, 28 July 2026. It covers 31 million citations across 11 AI surfaces, from October 2025 to July 2026, with 37 major news domains from 105 outlets in the United States and the United Kingdom, and robots.txt read in July 2026. Goodie sells AI visibility software. https://higoodie.com/blog/publishers-and-ai-search-study/ (accessed on 6 October 2026).
  • Stacker, “New Stacker Research: Earned Media Distribution Triples AI Search Visibility, Delivers 239% Median Lift in Brand Citations”, GlobeNewswire, 16 March 2026. It covers 87 stories from 30 clients, more than 2,600 questions and 8 AI models, over 30 days, with Scrunch; the difference from 97% to 82% comes with p < 0.006, the median citations of distributed stories were 239% above those of brand-only content, and the note records that the previous study, from December 2025, with 8 stories, did not have the scale for broad conclusions. Stacker sells the distribution of brand stories to news outlets, and Scrunch is an AI platform for customer experience. https://www.globenewswire.com/news-release/2026/03/16/3256365/0/en/New-Stacker-Research-Earned-Media-Distribution-Triples-AI-Search-Visibility-Delivers-239-Median-Lift-in-Brand-Citations.html (accessed on 6 October 2026).
  • Loganix, press release “Zero Press Release Domains Appeared in Loganix’s 100-Query AI Category Test”, 24 March 2026, on Yahoo Finance. It covers 100 “melhor X em Y” (“best X in Y”) questions from 10 sectors, asked to Perplexity, ChatGPT and Gemini in March 2026. Loganix sells SEO and link building. https://finance.yahoo.com/sectors/technology/articles/zero-press-release-domains-appeared-221800810.html (accessed on 6 October 2026).
  • Ranqia, “Quais sites influenciam o ChatGPT no Brasil? Reddit e Wikipedia pesam mais que toda a mídia”, 20 July 2026. It covers 195 days, from 19 December 2025 to 1 July 2026, and 60,723 domains, in ChatGPT answers to unbranded questions that triggered search, in Brazil. Ranqia sells a GEO service. https://www.ranqia.ai/blog/sites-que-influenciam-o-chatgpt-no-brasil/ (accessed on 6 October 2026).
  • Núcleo Jornalismo, “93% dos sites de notícias no Brasil estão expostos à coleta de dados por IA”, 12 March 2026, about the study “The Protocol Gap”, by the Journalism Relay Project, Momentum and the International Fund for Public Interest Media, with data from Atlas da Notícia, across 4,025 journalism sites in Brazil. https://nucleo.jor.br/curtas/2026-03-12-93-dos-sites-de-noticias-no-brasil-nao-tem-politicas-de-ia/ (accessed on 6 October 2026).
  • Ahrefs, “Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied)”, Louise Linehan, 12 December 2025, updated on 12 August 2026. Ahrefs sells SEO and AI tracking tools. https://ahrefs.com/blog/ai-brand-visibility-correlations/ (accessed on 25 September 2026).
  • Google Search Central, “Políticas de spam para a Pesquisa Google na Web”, last updated on 16 September 2026, link spam section. https://developers.google.com/search/docs/essentials/spam-policies?hl=pt-br (accessed on 1 October 2026).
  • CONAR, Código Brasileiro de Autorregulamentação Publicitária, article 30. https://conar.wpenginepowered.com/wp-content/uploads/2026/08/Codigo_CONAR_2026.pdf (accessed on 1 October 2026).
  • Wikipédia, “Wikipédia:Critérios de notoriedade”, recommendation approved by the community, last edited on 15 March 2025: press releases are among the examples of sources that are not independent. https://pt.wikipedia.org/wiki/Wikip%C3%A9dia:Crit%C3%A9rios_de_notoriedade (accessed on 6 October 2026).
  • murmur, sector studies from September 2026: 51 niche sectors, 25 purchase questions per sector, without company names, asked once in ChatGPT, Claude and Google AIO, from 8 to 29 September 2026; the calculation by type of site considers 3,372 sources with a link from ChatGPT, 1,933 from Claude and 7,869 from Google AIO. https://murmur.marketing/estudos/ (accessed on 6 October 2026). Case studies published with client authorization: https://murmur.marketing/en/cases/mindo/, https://murmur.marketing/en/cases/visio/, https://murmur.marketing/en/cases/fotoregistro/ and https://murmur.marketing/en/cases/fly-vet/. Free diagnostic: https://murmur.marketing/en/diagnostic/.