Muck Rack’s “What Is AI Reading?” Report Review: What Its 25 Million Links Can Show

What is the report?

Muck Rack, a PR software company, publishes the report from data gathered for its Generative Pulse product. The first edition appeared in July 2025, with updates in December 2025 and May 2026. For the May 2026 edition, Muck Rack says it built a large, varied set of realistic consumer prompts, some naming companies and some not, ran them on web-enabled versions of ChatGPT, Claude and Gemini, and recorded every URL each model cited. It then sorted the links into categories: journalism, third-party corporate content, a company's own content, press releases, academic research, government and non-profit sources, social and user-generated content, encyclopedic sites, and tech platforms.

What are the headline findings?

What does it show that is useful?

The most practical findings are about differences, not totals:

• Each assistant reads different sources. Wikipedia tops ChatGPT's list and PubMed Central tops Claude's, while Gemini's top domains lean towards user-generated and Q&A platforms, YouTube and Quora among them. Muck Rack calls them three effectively separate information environments, and no journalism outlet ranks in the top five for all three.
• Question type changes the source mix. Industry trend questions cite journalism in 46% of responses, more than twice the rate of how-to and comparison questions, which lean on reference and brand-owned content instead.
• Recency matters for news. Journalism citations peak in the first month after publication and fall sharply by month six, with a long tail of older articles.
• Industries differ. Media and entertainment questions cite journalism most, at 36%; education and industrial questions least, at about 19%, where AI substitutes rankings platforms, research and company content.
• Some platform habits stand out. ChatGPT cites Google Maps results heavily for local “best of” questions, with about 188 Maps citations per 1,000 such queries. Claude cited no YouTube videos at all.
• One outlet breaks through. Axios appears in ChatGPT's top three cited domains in 13 of 17 industries; no other outlet does anything similar.

Each industry section also names the most-cited domains and journalists, which gives PR teams a starting list for their sector.

What can it not show?

• “Earned” is a wide net. The 84% includes Wikipedia, social platforms, academic papers, government sites and other companies' content, not only press coverage. Journalism alone is about 27%. Quoting “84% of AI citations come from earned media” as proof that PR drives AI visibility overstates what the number measures.
• Link counts favour heavy citers. Claude averages about 13 links per citing answer against ChatGPT's 5, so in pooled totals, sources Claude likes count for more than their share of answers.
• The prompts are not published. We cannot tell how many prompts there were, how they were phrased, or how they split between industries and question types. Totals depend heavily on that mix.
• Some assistants are missing. Perplexity, Microsoft Copilot and Google's AI Overviews and AI Mode are not included, and results from web-enabled model versions may differ from what people see in the consumer apps.
• Citation is not influence. A link in the source list does not show which sentence it supported, or whether the answer named your brand.
• The publisher sells PR software. That does not make the data wrong, but the framing favours earned media.

Muck Rack itself notes that AI systems are opaque and that behaviour may shift as models change.

How should you use it?

1. Start with the assistant your audience uses. The source lists differ so much that a single “AI” strategy is misleading. Wikipedia matters for ChatGPT and Claude far more than for Gemini.
2. Match effort to question type. If buyers ask about industry trends, journalism and press coverage matter. If they ask how to do something or compare products, reference pages, your own documentation and third-party reviews carry more weight.
3. Use your industry section as a shortlist of outlets and writers to research, not a list to pitch blindly.
4. Keep coverage fresh. For news-driven questions, a steady flow of coverage beats one big hit, as covered in how long earned media stays useful in AI answers.
5. Check against your own prompts. Run 20 to 30 of your buyers' questions in each assistant and list the cited domains. That shows which sources shape your category, as described in how to find frequently cited sources for outreach.

How the report's figures bear on paid versus earned placements is discussed in whether organic AI citations carry more trust than paid placements.

How does it compare with other citation studies?

It is one of the largest public citation datasets, and one of the few built around PR categories such as journalism and press releases. SEO-focused studies, such as Ahrefs' analyses of ChatGPT citations, look instead at rankings, page types and domains, and often cover one assistant. They answer different questions, so conflicting headline numbers usually reflect different prompts and categories rather than one study being wrong. Ahrefs' work is summarised in what Ahrefs' study of ChatGPT citations found.

We have not reproduced Muck Rack's analysis. This review is based on the May 2026 report and Muck Rack's published summaries of earlier editions.

What might this look like in practice?

Consider an illustrative energy storage start-up, Hasketh, planning its first PR budget. The report's energy section showed ChatGPT citing Axios heavily, Claude and Gemini leaning on government data such as the EIA, and trend questions drawing on recent journalism. Hasketh's own tests confirmed that “where is home battery storage heading” answers cited news and government data, while “how to size a home battery” answers cited installers' guides. It split its effort: pitching trade and business press on its field data, and publishing a detailed sizing guide of its own. Three months later, both question types occasionally cited Hasketh's sources.

What do people ask about Muck Rack's AI citation report?

Is the report free?

Yes. Muck Rack publishes it as a free download.

Does 84% earned media mean PR drives AI answers?

Not on its own. “Earned” includes Wikipedia, research, government and social content; journalism is about 27%.

Which AI assistants does it cover?

ChatGPT, Claude and Gemini. Perplexity, Copilot and Google's AI Overviews are not included.

How often is it updated?

It has appeared roughly every five to six months since July 2025, in July 2025, December 2025 and May 2026.