Which Publications Does AI Actually Cite? A Media Targeting Guide for PR Pitching in 2026

Key takeaways

• More than 84% of links cited by AI come from earned or third-party sources, and paid or advertorial content is just 0.3% (Muck Rack Generative Pulse, September 2026). PR still feeds AI answers, but the mix is moving.
• Journalism's share of citations fell from 27% in May to 22% in September 2026. Government, encyclopedic, academic and third-party corporate sources gained.
• ChatGPT, Claude, Gemini and Perplexity cite very different sources. Only 212 domains were cited by all four, so one placement does not travel across models.
• News is only 5.7% of ChatGPT citations. Listicles, comparisons, product pages and landing pages matter more, and news hits decay fast.
• Build your target list from citation data in your own category, then check crawler access and prompt-level results before you pitch.

The old target list is out of date

Most PR media lists were built around reach: monthly uniques, domain authority, whether a client's CEO would brag about the logo. That list still has value for humans reading the press. It is a weak guide for what AI systems pull into answers.

AI engines do not cite by prestige. They cite pages that answer a specific query, that their crawlers can read, and that their retrieval system happens to favor that month. The result is a source mix that looks odd next to a traditional media ranking. Mid-authority sites in particular punch above their weight. Promptwatch Data's domain rank report found that sites with domain rank 46-75 earned almost half of ChatGPT citations in August 2026, while DR 91-100 sites fell to about 3%.

So the question for pitching in 2026 is not "which outlets are biggest?" It is "which outlets does the model I care about cite for the prompts my buyers type?"

What the 2026 data says about earned media

The most useful third-party dataset is Muck Rack's "What Is AI Reading?" series from its Generative Pulse team. The October 2026 edition draws on more than 40 million AI responses and 207 million citations from 2026, plus a September analysis of 1,020 prompts across ChatGPT, Claude, Gemini and Perplexity. Some agency write-ups quote a looser "82-95% from earned media" figure. Muck Rack's 84%+ is more precise, and it is current through September.

The headline numbers:

• Nearly 99% of cited links are non-paid.
• Paid and advertorial content is 0.3%, unchanged since May.
• Journalism was 22% of citations in September, down from 27% in May. It had held at 25-27% since July 2025.
• Government/NGO sources rose from 8.6% to 9.6%, aggregators and encyclopedic sources from 17.4% to 18.3%, academic from 4.0% to 4.4%, third-party corporate from 24.0% to 25.1%, and owned media from 13.7% to 15.2%.

Two takeaways. First, earned coverage is still the main lever, and buying your way in does not work. Second, journalism is not the only earned channel. If you are only pitching reporters, you are going after roughly a fifth of the citation pool.

One detail I find telling: Forbes editorial content outperformed Forbes Councils pay-to-play content by about 1.7 to 1 in ChatGPT citations. Same domain, very different treatment. The models, or the retrieval systems feeding them, appear to separate real editorial from sponsored contributor material.

Each AI model is its own media environment

This is the part that breaks the "get into one big outlet" plan. Muck Rack's May 2026 edition, covering more than 25 million links across 17 industries, showed how differently the models behave:

ChatGPT cites almost every time but sparingly, so each slot is competitive. Claude cites less often but pulls many sources when it does. Journalism shares have also diverged over time: ChatGPT's fell from 28.1% in January to 22% in September, Claude's from about 28% to 24%, and Gemini's from 27.6% to 23.6%.

In the May data, Axios appeared in ChatGPT's top three cited domains in 13 of 17 industries, the only journalism outlet in any provider's top three. If you want one cross-industry name to include on a ChatGPT-focused target list, that is the data-backed one. Treat it as a May snapshot, though, because ChatGPT's journalism mix has shifted since and several major publishers dropped to near zero in ChatGPT citations despite allowing OpenAI's search crawler.

Only 212 domains were cited by all four major providers in the September analysis. That number should change how you report results. A feature that shows up in Gemini answers tells you little about ChatGPT.

Social and Reddit are not a PR shortcut anymore

A year ago, plenty of PR and SEO teams treated Reddit as a back door into ChatGPT. That door closed. According to Promptwatch's data on Reddit citations, reddit.com's share of ChatGPT citations fell from roughly 4% to 0.5% on August 14, 2026, with Google AI Overviews and AI Mode declining more gradually.

The October 2026 social citations report (Sep 28 to Oct 4) shows how lopsided it has become:

In ChatGPT, Reddit dropped to 0.09% and LinkedIn is now the top social platform at about 0.3%. In AI Overviews, YouTube is 4.28%, Facebook 2.38%, Reddit 2.15%, Instagram 1.61% and LinkedIn 1.55%. TikTok shows up only in AI Overviews, at 0.22%.

The PR implication is plain. If ChatGPT is your priority surface, earned coverage on publisher and third-party sites matters far more than social presence. If Google AI Overviews matters more to your category, video and social deserve a real budget. Decide per model rather than treating "AI" as one thing.

Pitch for the formats AI cites, not just news

Here is the finding that changed how I would brief a PR team. In Promptwatch's content type report for Sep 28 to Oct 4, 2026, ChatGPT's citations broke down like this:

News is under 6%. Listicles are more than twice that. For commercial prompts, product pages, landing pages, listicles and comparisons add up to almost 70% of citations. For branded prompts, reviews, FAQ, social posts and press releases all take a larger share than they do overall.

What this means for pitching:

• Aim for inclusion in "best of" roundups and comparison articles in your category. A single line in the right listicle can outperform a standalone news story.
• Pitch review outlets and comparison publishers, not only newsrooms.
• Remember decay. Promptwatch Data on content half-life tracked pages ChatGPT cited on July 6, 2026 and found news lost half its pages within about 36 days, while landing pages and docs stayed cited longest. A news hit is a short-lived asset. A durable evergreen mention keeps working.

Press releases are a mixed bag. They are a small slice of citations overall, and a bigger one on branded prompts. Distribution vendors say structured information, original data and named experts with full titles help. That is plausible and matches how citations work, but treat it as vendor advice rather than measured fact.

Vertical matters: the list changes by industry

Citation mixes differ sharply by sector. A few examples from Promptwatch's October and August 2026 reports:

• Healthcare: NIH, the FDA and the NHS lead ChatGPT citations, and government and public health sites earn almost half. See the healthcare citation report.
• Travel: Skyscanner, Going and Tripadvisor lead, and booking and search sites earn more than half. See the travel citation report.
• Software: GitHub, Microsoft's docs and G2 lead, and documentation and help centers earn the largest share. See the software citation report.
• Automotive: marketplaces like Autotrader and Cars.com outrank every car brand, and Chevrolet is the top manufacturer site. See the automotive citation report.

Notice how few of those are classic news brands. A healthcare PR team may get more from a placement with a public health body or a medical association than from a national daily. A software team may get more from G2 and docs ecosystems than from TechCrunch-style coverage. Check the monthly top-domain reports, such as ChatGPT citation share for June 2026, for specific names, and remember the numbers shift month to month.

How to build an AI-informed media target list

Step 1: collect the prompts your buyers actually use

Start with 30 to 100 prompts across three types: category questions ("best X for Y"), comparison questions ("A vs B"), and branded questions. Add the specific regions and languages you sell in.

Step 2: see which domains get cited

Run those prompts across ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews, and record every cited domain and page. Do not trust a single run. In Muck Rack's September test, running the same prompt twice in one day, ChatGPT repeated only 41% of cited domains, Gemini 45% and Perplexity 94%. You need repeated runs and averages, which is why one-off screenshots are bad evidence.

This is where tooling helps. Promptwatch tracks the real interfaces of the major AI engines daily and shows which domains and pages get cited per prompt, plus Reddit, YouTube and LinkedIn citations separately.

Promptwatch

Track and optimize your brand's visibility in AI search engines

View more

Other options in the same category include [tool:profound], [tool:peec-ai], [tool:otterly-ai] and [tool:ahrefs-brand-radar]. For a wider list, browse the GEO software directory at bestgeosoftware.com.

On the PR side, Prowly's Semrush AI PR Toolkit has an AI-Cited Media Database that surfaces the outlets AI platforms reference most in your industry, and Cision bundles AI analytics with PR Newswire distribution. I could not verify pricing for either, so check directly. Media monitoring tools such as [tool:meltwater] and [tool:mention] cover the coverage-tracking side.

Step 3: sort the domains into tiers

Take the cited domains and group them:

Step 4: check crawler access before you pitch

This step saves a lot of wasted effort. In Muck Rack's data, among major publishers blocking Claude's search crawler, 81% got almost no Claude citations, versus 12% of publishers that allowed it. A placement on a site that blocks the relevant AI crawler is much less likely to help you in that model. You can check a publisher's robots.txt in about a minute, and Promptwatch's free AI Bots and Web Crawlers Directory at promptwatch.com/bots helps you identify which user agents matter.

Step 5: pitch the right person with the right asset

Journalists get 50+ pitches a day, so the basics still apply: a named writer, a reason tied to their recent coverage, and something concrete. For AI citation purposes, add things models can lift: original data, full-name expert quotes with titles, and clear structured claims. Fractl's pitch checklist and OnePitch's journalist-matching guide are both reasonable references on the craft itself.

Step 6: measure citations, not just clips

After coverage lands, track whether the page gets cited for your target prompts, in which models, and how long it lasts. Given the 36-day news decay, plan refreshes and follow-up placements. Your PR report should include citation share for tracked prompts next to impressions and links.

Common mistakes

• Treating Reddit as a ChatGPT lever. The data says it no longer is.
• Paying for advertorial placements and expecting citations. They are 0.3% of the pool.
• Judging success from one prompt test. Single runs are unreliable, especially in ChatGPT and Gemini.
• Ignoring Google AI Overviews. Its source mix, including 12.2% social, is quite different from ChatGPT's.
• Pitching only news. Listicles and comparisons earn more ChatGPT citations than news does.
• Using stale numbers. This field moves monthly. Journalism's share dropped five points in four months.

A quick quarterly routine

1. Refresh your prompt set and re-run it across all models.
2. Export the top cited domains per model and per prompt type.
3. Compare with last quarter and flag publishers rising or falling.
4. Re-check crawler access on your top 20 target publishers.
5. Update your pitch calendar with a split between evergreen roundup targets and news targets.
6. Review which earned placements are being cited and which were never picked up.

If you want a team to run this for you, an agency that combines digital PR with AI search work is the right fit. 1001 SEO Media does exactly that, pairing digital PR and link building with generative engine optimization.

For more first-party numbers on how AI engines pick sources, the Promptwatch Data hub is updated regularly.

The short version

AI engines still lean on earned media, but not on the same outlets or formats as your old media list. Build the list from citation data per model and per industry, put roundups, comparisons and trusted institutional sources next to news, and verify crawler access first. Then measure citations on a schedule, because the mix shifts faster than any annual PR plan.