AI citations still come overwhelmingly from earned media — more than 84% of what ChatGPT, Claude, Gemini, and Perplexity cite when they answer a question, and that number hasn't budged in a year. What has moved is which earned sources make up that 84%. Journalism's slice fell from 27% to 22% since May, per Muck Rack's October 2026 "What Is AI Reading?" study, and a chunk of that has gone to third-party corporate content and aggregators instead. If you're trying to show up in AI answers, that shift changes who you need to be talking to.
What Muck Rack's October 2026 AI Citations Report Found#
Muck Rack and its Generative Pulse data arm analyzed more than 40 million AI responses and 207 million citations across 2026, plus a fresh September run of 1,020 identical prompts across the four big models. The September citation mix breaks down like this:
- Earned and third-party sources combined: over 84% (journalism, government/NGO sites, academic research, encyclopedic sources, third-party corporate content)
- Journalism specifically: 22%, down from 27% in May — the largest single shift Muck Rack has tracked across four editions
- Third-party corporate content: 24–25%
- Aggregators and encyclopedic sources: 17–18%
- Owned media (a business's own site): 13.7–15.2%
- Paid and advertorial content: just 0.3%, unchanged from May
The report also found something we think matters more than the headline number: only 212 domains got cited by all four major AI providers in that September run. Each model is reading its own, mostly separate slice of the internet.
Does This Matter If You Don't Have a PR Team?#
If you've never pitched a journalist and have no plans to, you might assume this is someone else's problem. It isn't, and here's why. Third-party corporate content and industry aggregators sit at roughly 24-25% and 17-18% of citations — bigger than journalism now. That's analyst write-ups, directory listings, comparison sites, and partner mentions. None of that requires a press contact. It requires someone, somewhere, writing about your business in a way an AI model trusts enough to quote.
So no, you don't need a PR agency to benefit from this data. You need to show up in the kind of third-party content that's already winning the citation share journalism is losing.
The Part Nobody's Selling You: Crawler Access Decides Who Gets Cited at All#
Buried in the report is a detail we found more useful than the topline stat. Among publishers that block Claude's search crawler outright, 81% got almost no Claude citations back. Among publishers that allow it, only 12% saw that same near-zero outcome. That's not a subtle correlation.
The honest trade-off: allowing AI crawlers means you're handing your content to a model for free, with no guarantee of a click back, and some publishers block them for exactly that reason. We get the instinct. But the data says blocking the crawler doesn't protect your content — it just removes you from the answer entirely. There's no middle option where you're cited without being crawled.
How We've Already Adjusted Our Own Setup#
We don't run a newsroom, but this blog has been built around getting cited since we rebuilt our AEO structure months ago — every post here ships with a direct-answer summary, structured FAQs, and real citations, specifically so a model skimming this page has something clean to lift. This report is the first data that tells us whether that bet paid off or not, and we're treating the 212-domain overlap figure as our baseline to beat.
This week we also pulled crawler-access settings across the client sites where we handle SEO and web development, mostly GCC and European service businesses, and found two still blocking Claude's bot from a staging-era robots.txt nobody had touched since launch. Fixed both. If a prompt-consistency number like "41% of ChatGPT's citations change between two runs on the same day" sounds unreliable to test against, it is — which is the whole reason we stopped trusting one-off AI Overview checks a while back and started tracking visibility the same way we'd track Search Console data: over time, across models, not off a single screenshot.
What I'd Tell a Client Asking About This This Week#
Check your robots.txt for GPTBot, ClaudeBot, Google-Extended, and PerplexityBot before you do anything else — that's the five-minute fix that matters most in this whole report. After that, don't chase a single AI tool's answer as proof of anything; Gemini and ChatGPT reshuffled 55-59% of their own citations between two runs the same day, so one screenshot tells you almost nothing. If you want third-party content working for you, a handful of genuine comparison or directory mentions will do more right now than a press release will.
If you want to know whether your own site is even crawlable by the models that matter, send me the domain and I'll check it myself — two minutes, no pitch. cal.com/webepex/growth-review