82% of Journalists Now Research Stories in AI. Here's What Happens When Your Brand Isn't in the Answer.
Muck Rack's 2026 State of Journalism report says 82% of surveyed journalists use AI tools. This brief separates journalist AI adoption, AI citation evidence, earned-media sourcing, and practical media-targeting audits.
There is a new operating risk in media targeting: journalist research habits, AI retrieval, earned coverage, and buyer discovery are now connected, but they are not the same claim.
Muck Rack's 2026 State of Journalism report says 82% of surveyed journalists use AI tools. ChatGPT leads at 47% adoption among respondents, with Google Gemini at 22% (Muck Rack, 2026). The measured unit is respondent-reported tool adoption and use cases. Boundary to preserve: the survey shows many journalists report using AI for work such as research and support tasks; it does not establish that an absent brand is excluded from sourcing, coverage, consideration, recommendation, or revenue.
That still matters for PR teams. If journalists are using AI systems as one input while researching stories, then media lists built only around human reach are incomplete. The task is not to assume a deterministic loop. The task is to audit whether the publications and reporters you pitch also appear in the AI answers that shape category research.
The 2% overlap problem
The clearest targeting warning comes from Muck Rack's Generative Pulse analysis of more than one million AI-generated citations across ChatGPT, Claude, Gemini, and Perplexity. Muck Rack reported that the journalists most frequently pitched by PR professionals and the journalists most frequently cited by AI engines overlapped by just 2% (Muck Rack Generative Pulse, via GlobeNewswire, March 30, 2026).
The same release reported that earned media accounted for nearly 25% of citations in that observed citation sample and that non-paid sources represented 94% of AI-cited links. The measured unit is Muck Rack's observed sample and taxonomy. Boundary to preserve: Muck Rack Generative Pulse does not establish a universal media-targeting mechanism, the largest causal input to AI recommendation, a deterministic budget prescription, or proof that every high-citation journalist should outrank every other media target.
My read on this: the 2% figure exposes a media-list blind spot, not a law. Teams often build pitch lists around beat coverage, audience size, recent responsiveness, and relationship history. Those are still useful. But they should be checked against a second map: which publications and journalists are actually cited in AI answers for the category.
Ahrefs' 75,000-brand study reported that brand web mentions correlated with AI Overview visibility at 0.664, compared with 0.218 for backlinks (Ahrefs, 2025). The measured unit is an observed correlation in that Ahrefs study. Boundary to preserve: Ahrefs does not establish three-times causality, a provider mechanism, or proof that web mentions, earned placements, or backlinks by themselves cause AI visibility.
The Fullintel-UConn study presented at IPRRC in March 2026 — 400 prompts across 10 personas, run on a single platform against a single health topic — reported that journalistic sources made up 47% of sampled AI citations, with corporate, university and health-network sites making up 48% (Fullintel, 2026). The measured unit is sampled source composition. Boundary to preserve: Fullintel-UConn is not peer-reviewed and has no public paper; it does not establish a universal earned-media mechanism, source-selection primacy, guaranteed citation, recommendation lift, forecast, business outcome, or media-budget rule.
Together, these sources support a practical audit question: does your media list include publications and journalists that appear in AI answers for the categories your buyers and reporters research? They do not prove that earned media automatically creates AI recommendation, that AI systems index any placement on a fixed timeline, or that a reporter's source list is pre-filtered by an AI answer.
How to think about the loop without overstating it
The useful model has four distinct parts. Keep them separate when you plan, measure, and brief executives.
| Layer | What the cited evidence can support | What it does not prove |
|---|---|---|
| 1. Journalist AI adoption | Surveyed journalists report using AI tools, including ChatGPT and Gemini | That every reporter uses AI for source selection or excludes brands absent from an answer |
| 2. AI retrieval and citation | Observed AI answers cite some publications, journalists, and non-paid sources more often than others in sampled datasets | A universal provider mechanism, fixed indexing speed, or guaranteed citation after a placement |
| 3. Sourcing consideration | AI answers can be one research input among many for reporters and buyers | Deterministic story assignment, source-list pre-filtering, earned placement, or exclusion |
| 4. Brand visibility and commercial impact | Visibility can be audited across answers, citations, publications, and category queries | Recommendation, revenue, pipeline, or market-share outcomes without direct measurement |
A journalist at a trade publication may open ChatGPT, Gemini, Perplexity, or another tool to research which companies are active in a category. The answer may include publications, executives, studies, or companies already present in the machine-readable record. If your brand is absent, that is a signal to inspect the record. It is not proof that the journalist will omit you or that one article will reset future answers.
This is where concentration data adds urgency but still needs a boundary. The Authoritas study tracking 143 digital marketing experts reported that, between December 2025 and February 2026, the top 10 captured 59.5% of citability across ChatGPT, Gemini, and Perplexity, up from 30.9% two months prior. It also reported a 293% rise in citation concentration over that panel and time window (Authoritas, 2026). The measured unit is a bounded panel over a bounded interval. Boundary to preserve: Authoritas does not prove universal accelerating compounding, a narrowing entry window for every category, or a fixed outcome for brands outside the observed panel.
The three-step media targeting audit
If your team runs earned media and has not mapped targeting to AI citation behavior, start with this audit. It takes one afternoon and tells you whether your pitch list is aligned with the publications AI systems cite for your category.
Step 1: Map your current pitch list against AI citation sources. Pull your active media list. For the top 20 targets, run category prompts across ChatGPT, Google AI Mode, Gemini, and Perplexity. Record whether each publication appears in the cited or referenced sources. If few of your top targets appear, label it as a visibility gap to investigate, not as proof the list is wrong.
Step 2: Identify the publications AI engines actually cite for your category. Run five category queries across at least three engines. Record every cited publication, article, and journalist when the engine provides them. Build a frequency-ranked list. The publications that appear repeatedly are candidates for earned authority work, but frequency is an observed retrieval signal, not a guarantee that future coverage will be cited.
AT's research on earned media vs. owned content citation rates observed a 4.25x citation-rate difference for earned-media URLs versus owned-content URLs in its measured dataset. The measured unit is a bounded observed rate comparison. Boundary to preserve: MR earned-vs-owned does not establish that earned media causes citations for a given brand, is the primary causal input, creates provider recommendation, forecasts visibility, or produces pipeline, revenue, or business outcomes.
Step 3: Prioritize editors whose coverage also matches human editorial fit. Within the high-citation publications from Step 2, identify the reporters whose articles appear in AI answers and whose beat genuinely fits your evidence. Then apply ordinary PR judgment: relevance, newsworthiness, credibility, relationship context, and timing. This is a targeting rebuild around both human editorial value and machine-readable retrieval evidence, not a replacement for editorial fit.
The Muck Rack 2026 State of Journalism data gives context for the human side of that filter: 88% of respondents said they immediately delete pitches not aligned with their beat, 70% prioritize beat alignment, 58% want access to credible sources, and 40% value original data (Muck Rack, 2026). Pitches under 200 words are preferred by 69% of respondents.
| What AI-cited journalists still need | What weak outreach often sends |
|---|---|
| Beat-aligned, data-rich pitches under 200 words | Generic category overviews sent to broad lists |
| Access to credible, named sources with original data | Executive quotes without supporting numbers |
| Specific findings tied to trends they are already covering | Product announcements framed as industry news |
That profile maps to a practical content standard: the pitch should be useful to a human reporter and legible to a machine system. Those are related requirements, not proof of a shared selection mechanism.
Why this changes what earned media measurement should include
The implication is not that PR teams should abandon human-audience targeting. It is that an earned placement may now be measured across two surfaces: the human audience that reads the article and the AI systems that may retrieve or cite it later.
A placement in a publication that AI engines often cite can support both credibility and machine-readable authority. A placement in a publication rarely retrieved by AI systems may still matter for human reach, analyst validation, investor trust, partner confidence, or category reputation. The audit should reveal tradeoffs; it should not collapse every outcome into citation counts.
The Pew Research Center's July 2025 study found lower click rates when AI summaries appeared in Google search results, with clicks dropping from 15% to 8% in its observed browsing data (Pew, 2025). Bain reported that about 80% of surveyed search users rely on AI summaries at least 40% of the time (Bain, 2025). The measured units are user behavior and survey scope. Boundary to preserve: Pew and Bain do not prove that one earned placement feeds both human and AI channels, that earned media is the highest-leverage investment for every growth team, or that AI-mediated discovery produces a specific commercial result.
As I've written previously, the execution challenge is not producing more content but targeting the right publications. The three-step audit above is where that targeting rebuild starts.
This is where Machine Relations provides the operating frame. Machine Relations, coined by Jaxon Parrott in 2024, treats brands as entities that must be understood by machines as well as by humans. In this context, the mechanism to inspect is not a guaranteed loop. It is the relationship among earned media, source retrievability, citation, recommendation, and brand visibility.
Key takeaways
- Muck Rack's 2026 State of Journalism survey reports that 82% of surveyed journalists use AI tools; that is adoption evidence, not proof of AI-only source selection.
- Muck Rack Generative Pulse reported 2% overlap between most-pitched journalists and most-AI-cited journalists in its observed sample; that supports a media-list audit, not a deterministic targeting rule.
- Ahrefs reported a 0.664 correlation between brand web mentions and AI Overview visibility versus 0.218 for backlinks; that is correlation, not causation or provider mechanism.
- Fullintel-UConn reported 47% journalistic-source composition, against 48% corporate, university and health-network sites, in one single-topic sample; that is source-composition evidence, not a universal earned-media mechanism.
- MR earned-vs-owned observed a 4.25x citation-rate comparison; that is a bounded observed rate difference, not causality, recommendation, forecast, or business outcome.
Run the audit
Run the three-step audit this week. If your media list has the 2% problem, the fix is not more pitches to the same targets. It is a targeting rebuild around publications and journalists whose work appears in AI retrieval samples and still matches real editorial fit. AuthorityTech's visibility audit maps your earned media footprint against publications AI engines cite for your category, so you can see the gap before you brief a single pitch.
Frequently Asked Questions
Why does the 2% journalist overlap matter for brand visibility?
It matters because it identifies a possible mismatch between who PR teams pitch and whose work appears in sampled AI citations. Muck Rack's overlap figure does not prove that absent brands are excluded from sourcing or coverage; it supports checking your media list against AI retrieval evidence.
How do I find which publications AI engines cite for my category?
Run repeated category queries across ChatGPT, Google AI Mode, Gemini, and Perplexity. Record cited publications, articles, and journalists. Treat the result as an observed retrieval sample. Do not treat it as a universal source map or a guarantee that those outlets will drive recommendations.
How should I use the earned-vs-owned citation-rate research?
AuthorityTech's MR earned-vs-owned research observed a 4.25x citation-rate comparison in its measured dataset. That does not establish that earned media causes citations for every brand, creates provider recommendation, or produces revenue. Use it as a reason to compare earned and owned citation behavior in your own category.
What does the Ahrefs 0.664 versus 0.218 finding prove?
It proves only an observed correlation within Ahrefs' study: brand web mentions were more closely associated with AI Overview visibility than backlinks. It does not prove three-times causality, an engine mechanism, or that backlinks are irrelevant.
Should PR teams prioritize AI-cited journalists over traditional beat fit?
No. Use AI citation evidence as one layer in media targeting. Beat alignment, credibility, original data, relationship context, and editorial relevance still decide whether a pitch is useful. Separate journalist adoption, AI retrieval, sourcing consideration, earned placement, recommendation, brand visibility, and commercial outcomes.