What is AI reading now? Analysing Muck Rack's October 2026 findings
The latest edition of Muck Rack's ‘What Is AI Reading?’ study (October 2026) has been released, and the headline figure has hardly moved: about 84% of AI citations come from earned media and roughly 99% from non-paid sources.
Elsewhere, plenty has changed. Journalism's share has fallen, the four main AI providers draw on very different sources and the same question asked twice on the same day can return a different set of links, so there is no single AI platform for brands to optimise for.
The study is produced with Muck Rack’s own Generative Pulse platform, and draws on more than 40 million AI responses. Below, I dive into the research and provide some opinions of what it means for B2B tech and enterprise businesses.
Key takeaways
Earned media still accounts for ~84% of AI citations, but journalism's share fell from ~27% in May 2026 to 22.2%
Each provider favours different sources: government and NGO pages lead ChatGPT (34.6%), owned media leads Gemini (34.4%), journalism leads Perplexity (27.8%) and Claude splits almost evenly between third-party corporate content (23.5%) and journalism (23.22%)
Only 212 domains were cited by all four providers
Question type shapes what gets cited: practical guides take 68% of citations on how-to prompts and buyer's guides 45% on best-of prompts
Claude, Gemini and Perplexity cite much smaller news outlets than ChatGPT, strengthening the case for trade and regional coverage
Asked the same question twice on the same day, cited domains overlapped by between 41% (ChatGPT) and 94% (Perplexity), so one-off checks are unreliable
Earned media still leads, but journalism's share is shrinking
Muck Rack's definition of earned media is broad. Alongside journalism, it includes academic research, government and NGO sources, encyclopaedic sites such as Wikipedia, social platforms and third-party corporate content. Here’s how the main categories are split:
Third-party corporate blogs and content: 25.1%
Journalistic: 22.2%
Aggregators and encyclopaedic: 18.3%
Owned media: 15.2%
Government and NGO: 9.6%
Academic and research: 4.4%
Social and UGC: 3.1%
Press releases: 1%
Paid and advertorial: 0.3%
Read the 84% figure with that in mind. Third-party corporate content outweighs journalism, so press coverage is only one part of what counts as earned. Owned and paid content together make up about 17%, roughly one percentage point higher than in May.
Journalism fell from about 27% of citations in May to just over 22%. Those five points mostly moved to other earned categories, so the overall earned share held steady.
ChatGPT led that fall, with its weekly share of journalism citations dropping from 32.9% in January to 22% by September, with sharp falls in April and July. For Claude, it was a drop from 27.6% to 25.1%, and Gemini from 28.1% to 23.6%.
Through 2026, twelve major publishers fell to near zero in ChatGPT's citations for a period, in sudden waves:
January: Reuters
March: Financial Times
May: The Verge, Business Insider, Wall Street Journal, Investopedia, People, Barron's and Politico
June: CNBC
July: AP and Axios
Some have since returned, with AP and Axios back in ChatGPT's most-cited list, while Harvard Business Review, PBS and NBC News gained citations over the same period. Most affected sites allowed OpenAI’s search crawler, and many were still cited by other providers. Muck Rack suggests that this shift reflects ChatGPT’s own sourcing decisions.
What this means is that a placement in a major national title can vanish from an AI engine's answers overnight, for reasons that have nothing to do with you or the story. So a coverage plan built on a handful of top-tier titles carries a risk.
Four AI providers, four different reading lists
Muck Rack sums this up as the four main providers barely reading the same internet. Only 212 domains were cited by all four, led by forbes.com, linkedin.com, energy.gov, epa.gov and fda.gov, and Forbes is the only news outlet in every provider's top ten.
Muck Rack also re-ran the same 102 questions five months apart. Between April and September 2026, ChatGPT's share of government and NGO citations rose from 10% to 35% and owned media from 14% to 20%.
Perplexity's use of third-party corporate content fell from 24% to 20%, while Gemini's rose from 14% to 17%. ChatGPT and Gemini changed model versions in between, so model updates are a factor as well as time.
ChatGPT | Claude | Gemini | Perplexity | |
Largest source category | Government/NGO (34.6%) | Corporate blogs and content (23.5%) | Owned media (34.4%) | Journalistic (27.8%) |
Most-cited page type | Product/service pages (21%) | Buyer's guides (21%) and expert analysis (20%) | Product/service pages (50%) | Practical guides (26%) |
Answers with any citation, by question type | 27% to 96% | 34% to 97% | 92% to 99% | 100% |
Distinct domains cited (1,020 prompts) | 2,059 | 4,237 | 9,206 | 11,543 |
Median monthly visitors of cited news outlets | 6.51M | 0.76M | 0.12M | 0.59M |
Top news outlet | Axios (12 citations) | Forbes (19) | Forbes (44) | The New York Times (47) |
Domain overlap, same prompt run twice | 41% | 56% | 45% | 94% |
Judging by these stats, each provider rewards a different kind of presence. ChatGPT leans on government and NGO pages, Gemini on brands' own sites, and Perplexity and Claude on journalism and third-party content. Given how quickly these mixes move, any GEO audit dates fast.
Why trade and regional titles matter for AI visibility
The median monthly audience of the news outlets ChatGPT cites is 6.51 million unique visitors. For Claude, Gemini and Perplexity it is under a million.
Muck Rack highlights Claude's readiness to cite small trade and regional outlets as a pitching opportunity many PR professionals miss. For B2B sectors such as deep tech, space or enterprise software, the specialist titles buyers already read may count for more in AI answers than their audience size suggests.
Crawler access matters, especially for Claude and Gemini
In August 2026, Muck Rack compared 66 selected major publishers' robots.txt settings with their citations:
81% of publishers blocking Claude-SearchBot received almost no Claude citations, compared with 12% of those that allowed it
For Gemini, 67% of publishers blocking Google-Extended received almost no citations, compared with none of those that allowed it
ChatGPT broke the pattern: 56% of blockers were rarely cited, but so were 24% of publishers that allowed OAI-SearchBot, and some blocked sites still appeared through other routes
This is an association across a selected group of publishers, not proof of cause. But still, it's a good idea to check the robots.txt file of an outlet to see whether it blocks AI crawlers, before treating it as a GEO priority. While you’re at it, check your own site's settings too.
Each question type rewards different content
Muck Rack maps six common question types, including how-to, best-of or industry trends, to the formats cited most:
Question type | Formats cited most often | What communicators can do |
Problem discovery | Practical guides 28%, product pages 23% | Explain the problem clearly and connect it to a solution |
Best-of | Buyer's guides and reviews 45%, product pages 26% | Provide verifiable product information and pursue review coverage |
Industry trends | Expert analysis 41%, recent news 18%, research 14% | Publish original analysis and offer useful expert commentary |
How-to | Practical guides 68%, product pages 16% | Create clear, detailed instructional content |
Comparative | Evergreen reference 37%, buyer's guides 21% | Develop transparent comparisons and support third-party evaluation |
Risk and due diligence | Practical guides 36%, evergreen reference 17%, expert analysis 14% | Publish evidence-rich guidance and credible expert analysis |
For B2B brands, the industry-trend row is the most relevant. Expert analysis is the content a PR programme produces through bylines, commentary and original data.
Some answers cite nothing at all
Whether a provider cites anything at all also varies by question. Perplexity cited sources every time, while ChatGPT ranged from 27% on comparison questions to 96% on industry-trend prompts.
That means the prompt mix in any visibility report can swing its results. It also means that when an answer carries no links, the only way to appear is to be named in the text, so brand mentions need tracking separately from citations.
Journalism and owned media do different jobs
Nearly half of cited journalism (48%) covers breaking or recent news, and another 30% is buyer's guides and reviews. Owned media citations are led by product and service pages (58%) and practical guides (24%).
AI engines go to journalists for timely developments and independent evaluation, and to brands for product information and guidance. That mirrors how PR and owned content should already work together: coverage supplies the third-party validation and your own site supplies accurate product and service pages that help.
Product and service pages make up 50% of Gemini's citations, so a thin or outdated website costs more there than anywhere else.
Press releases, newswires and the Forbes question
Muck Rack classes press releases as paid media, and they account for about 1% of AI citations. But that’s not a reason to dismiss press releases as worthless. Their GEO (Generative Engine Optimisation) value lies in the coverage they generate.
Where cited releases are hosted also differs. ChatGPT links to the company's own site in about 85% of cases, a wire service in 14% and republishers in 2%. For Claude the split is 58%, 37% and 5%. Muck Rack adds that owned newsroom pages appear more often for discovery questions, while wires matter more when someone asks about a brand by name.
So host every release on a proper newsroom page, and treat wire distribution as useful for named-brand queries rather than a substitute for coverage.
The study also looks at Forbes Councils, the pay-to-play contributor programme many PR teams ask about. In ChatGPT's Forbes citations, editorial beats Councils on most question types, including industry-trend prompts (41.1% to 33.2%).
Risk and due diligence is the exception, where Councils lead 35.1% to 28.3%. Affiliate content such as Forbes Advisor often outranks both, taking 47.9% of ChatGPT's Forbes citations on how-to prompts.
If you are weighing a paid Councils membership for AI visibility, the data points to earned editorial coverage on most query types, at least on ChatGPT.
One check on one chatbot proves nothing | Muck Rack recommendations
Run the same prompt twice on the same day and the cited domains change, sometimes by more than half, even though each provider cites at roughly the same rate and volume.
A screenshot of one answer is an anecdote. Muck Rack recommends:
Use a stable set of prompts, organised by question intent
Run each prompt more than once
Report results by provider rather than as one blended score
Use ranges and rolling trends instead of one-time appearances
Record model and provider changes, since updates can shift citations quickly
Track citations and brand mentions separately, because a brand can be named without being linked
Where to focus
Most of this is familiar PR work. Here’s my advice for improving your chances of being cited, mentioned and recommended in AI search
Spread your coverage. Mix national, trade and regional outlets, since each engine draws on a different set and three of the four lean towards smaller titles
Match content to buyer questions. Expert analysis for trend questions, practical guides for how-to questions, honest comparisons for evaluation questions
Keep your own site in good order. Clear product and service pages, useful guides and a newsroom hosting every release. You can’t guarantee media coverage, but this is entirely in your control
Check crawler access. For target outlets and your own site
Measure at scale and by provider. Create your own audit process, prompts and schedule. Gather your own data across all providers and use it to inform your communications work moving forward.
Source: Muck Rack and Generative Pulse, What Is AI Reading?, October 2026. The full report is available at generativepulse.ai/report.
Author
Rob Phillimore is a freelance public relations consultant based in Cornwall UK, with 10 years experience across PR, marketing and communications. He runs GEO audits that show which AI engines cite and name your brand, which outlets they draw on in your sector and where the gaps are. Curious how ChatGPT, Claude, Gemini and Perplexity describe your business today? Email rob@robphillimore.com
Frequently asked questions
What is Muck Rack's What Is AI Reading? study? A recurring analysis of the sources AI providers cite when answering realistic questions, produced with Muck Rack's Generative Pulse platform. The October 2026 edition covers more than 40 million responses across ChatGPT, Claude, Gemini and Perplexity.
Does earned media still matter for AI search? Yes. About 84% of AI citations come from earned media. Journalism's share has fallen to 22.2%, but third-party sources of all kinds still dominate what AI engines cite.
Which AI platform should my brand prioritise? The ones your buyers use, and that is usually more than one. Each provider favours different source types and outlets, so visibility on one engine says little about the others.
Are press releases worth it for GEO? Releases make up about 1% of AI citations, so the coverage they earn matters more than the release itself. Wire versions help most when people ask about your brand by name.
How often should we check our AI visibility? Regularly, with the same prompts each time, and each one more than once. Sources vary even between same-day runs, so trends over time tell you far more than any single answer.
Should B2B brands still pitch trade press? Yes. Claude, Gemini and Perplexity cite news outlets with median audiences under a million, and expert analysis leads citations on industry-trend questions. Both point towards specialist titles.
Comments