Key takeaways
- Earned media is the raw material of AI answers. Muck Rack's analysis of AI-cited links found that 82-89% come from earned media across its editions, and paid content is a rounding error.
- Neither model wins outright. Traditional PR gets prestige placements, digital PR gets links and data pickups, and AI engines cite whichever pages they can read, trust, and extract from.
- News coverage has a short shelf life in ChatGPT. Promptwatch Data shows half of cited news pages stop being cited within 36 days, while listicles, comparisons, and landing pages last longer.
- Brand mentions without links now count as results. Ahrefs found branded web mentions correlate far more strongly with AI visibility than link metrics do.
- Measure pooled citation share per engine, not one screenshot of one answer.
The question behind the question
Most debates about digital PR versus traditional PR are really debates about budgets and org charts. One team wants a feature in a big business title. Another wants 40 referring domains from a data study. Both teams are right about what they are measuring, and both are measuring the wrong thing for AI search.
AI engines do not care about your clip count or your Domain Rating. They assemble answers from pages they can fetch, parse, and trust, then name brands that show up repeatedly across those pages. So the useful question is not which PR model is better. It is which kind of coverage ends up in the sources AI actually pulls from, and how long it stays there.

What each model is built to do
Traditional PR optimizes for human attention. Media training, message houses, embargoed exclusives, and broadcast slots exist to get a journalist to write about you in front of a decision-maker. The success metrics are impressions, clip counts, and sometimes ad value equivalent.
Digital PR grew up around search. Data studies, reactive commentary, and creative campaigns are designed to earn editorial links, referral traffic, and better rankings. The success metrics are referring domains, authority scores, and SERP position.
AI search adds a third audience. It wants earned editorial in publications it indexes, written so that a model can extract a clean claim about your brand. Here is how the three line up.
| Dimension | Traditional PR | Digital PR (link era) | What AI engines reward |
|---|---|---|---|
| Primary output | Prestige placements, broadcast | Backlinks, data campaigns | Third-party coverage that names your brand |
| Success metric | Impressions, clip count | Domain authority, referring domains | Citation share, AI-referred conversions |
| Unlinked mention | Fine, it is still coverage | Counted as a failure | Counted as a result |
| Content format | Press releases, bylines | Studies, infographics, newsjacking | Answer-first, attributable, extractable |
| Lifespan | One news cycle | Link equity builds over months | Depends on the page type that cites you |
The unlinked mention row is the one that changes behavior. Link-era digital PR treated a mention without a hyperlink as a miss and chased the editor to add one. For AI search, a brand named inside a page the model trusts is the whole point.
What the data says about earned media
The headline number comes from Muck Rack. Its May 2026 analysis of more than 25 million links cited by ChatGPT, Claude, and Gemini put earned media at 84% of AI citations and paid or advertorial content at 0.3%. Across the editions of its "What Is AI Reading?" research, the share has sat between 82% and 89%, with journalism alone at roughly 25-27%. An earlier Muck Rack report on more than a million links found the same pattern, with press release visibility growing.
BuzzStream looked at 4 million citations in April 2026 and found news publications supplied about 14% of them, second only to blog content at 53%. Two things follow. Journalism matters, and it is far from the only earned surface. Blogs, reviews, roundups, and community posts that mention you all count.
On what predicts visibility, Ahrefs measured 75,000 brands and found branded web mentions correlating at 0.66 to 0.71 with AI visibility. Domain Rating came in at 0.266 to 0.326, and backlink counts correlated very weakly. These are third-party figures, so treat them as directional, but they all point the same way: being talked about beats being linked to.
That is the strongest argument for putting traditional PR instincts back into digital campaigns. A respected trade title that names your product in a comparison does more for you than 30 low-grade links.
Which coverage actually gets cited, and for how long
This is where I think most PR advice goes soft. It treats a placement as a placement. Promptwatch Data, built from what ChatGPT actually cited, says the format of the page matters a lot.
In Promptwatch Data's report on ChatGPT citation share by content type (Sep 28 to Oct 4, 2026), product pages were the most cited type at 22.4%, and landing pages came next at 16.8%. Editorial formats added up to about 43%:
- Listicles: 13.6%
- How-to guides: 8.7%
- General articles: 6.1%
- Comparisons: 5.9%
- News: 5.7%
So straight news is a minor citation surface. Listicles and comparisons, the "best X for Y" pages that publishers love to write, are bigger. For commercial prompts like "best CRM for small businesses," product pages, landing pages, listicles, and comparisons made up almost 70% of citations. Reviews rose to 5.6% on branded prompts, more than three times their overall share, and press releases did better on branded prompts too (2.1%).
Then there is decay. Promptwatch's content half-life report tracked every page ChatGPT cited on July 6, 2026 for 90 days:
- News pages lost half their number within 36 days, and almost a quarter were never cited again after day 0.
- Landing pages were still cited at 78% after 30 days and 69% after 60.
- Docs held 76% at 30 days and 67% at 60. Comparisons held 72% and 63%.
- Listicles had 60% still cited at day 60.
- Product pages, reviews, and how-tos sat around 55% at day 60, with half-lives of 74-77 days.
The practical read: a news hit gives you a burst. A placement inside a listicle, comparison, or review on a third-party site gives you a long tail. A traditional PR win that lives only as a news item is a short citation window, even if the brand name on the masthead is impressive.
Prestige versus mid-tier publishers
There is a related trap in chasing only top-tier titles. Promptwatch's ChatGPT citation share by domain rank report found sites with domain rank 46-75 earned almost half of ChatGPT citations in August 2026, while DR 91-100 sites fell to about 3%. I would not read that as "avoid big publications." I read it as "a niche trade site that writes the comparison your buyers search for is worth chasing alongside the national feature."
The engines read different press lists
Each AI engine has its own habits. One piece of research from Organik PI says Gemini cites in 82% of responses and averages eight sources, while Claude cites in 55% of responses and averages 13 when it does. Profound's analysis of over 4 billion citations had Reddit as the most cited domain across ChatGPT, AI Overviews, and Perplexity at 3.11%, ahead of YouTube and Wikipedia.
Social and community sources are especially volatile. Promptwatch Data on Reddit in ChatGPT shows Reddit holding a steady 3.83% of ChatGPT citations from July 18 to August 7, 2026, then falling below 1% on August 14. The August 14-17 average was 0.52%. Promptwatch flags the size of the drop as provisional and has not ruled out a data-collection issue, but the lesson stands. Google's surfaces moved more gently: Reddit in AI Overviews eased from about 2.4% to 2.1%, and in AI Mode from 2.22% to 1.54%.
The broader split is in Promptwatch's social media citations report: from Sep 28 to Oct 4, 2026, ChatGPT got 0.46% of its citations from social platforms, while Google AI Overviews got 12.2%. A PR plan built on community seeding looks very different depending on which engine your buyers use.
If you want a conflicting number to reconcile: Profound's Reddit figure and Promptwatch's differ because the periods, prompts, and engines differ. Where they conflict for ChatGPT, use the dated Promptwatch figures. And whichever you use, do not bet the program on one platform.
Where each model earns its keep
What traditional PR does better
- Relationships with editors who write the comparison and roundup pages AI cites.
- Credibility and sentiment. AI answers describe brands using the claims in the coverage, so tone and framing matter.
- Spokesperson and expert placement, which produces named, attributable quotes.
- Entity-building: Wikipedia-style records and knowledge panels lean on reputable coverage.
What digital PR does better
- Data studies that give journalists and bloggers something concrete to cite, which creates repeated, consistent mentions.
- Speed and volume. Reactive commentary scales in a way a rolodex does not.
- Technical hygiene: making sure the pages that mention you are crawlable and the claims are easy to extract.
- Measurement habits. Digital PR teams already track referral traffic and attribution, which transfers to AI-referred visits.
Where both fall short
Both tend to stop at the placement. Neither asks whether the page that mentions you is one an AI engine reads, cites, and keeps citing. A press release on your own site, or an advertorial, sits in the 0.3% bucket and does almost nothing for you in AI answers. Meanwhile your own product, landing, and docs pages are the biggest citation surface in the Promptwatch data, so third-party coverage should link and describe consistently with what those pages say.
A combined playbook
Here is how I would run a hybrid program in 2026.
- Start from the prompts. List the 30-50 questions buyers ask AI about your category, and pull the sources those answers cite today. Your target list is those domains, not a generic top-100 media list.
- Sort targets by page type. Prioritize publishers who run listicles, comparisons, and reviews in your category, because those have longer citation lives than news.
- Build one citable asset per quarter. A proprietary data study with clear, quotable findings gives every outlet the same fact to repeat, and consistent repetition is what builds brand-level association.
- Pitch for naming, not just linking. Brief journalists with the exact product name, category, and one-line claim you want attributed. Accept unlinked mentions.
- Treat news hits as bursts. Plan to follow them with placements in evergreen formats, and update your own comparison and landing pages so there is a stable page to cite after the news cycle ends.
- Cover the community layer lightly. Reddit, YouTube, and LinkedIn mentions matter on some engines and not others, so diversify rather than depend on one.
- Keep the technical basics clean. Make sure AI crawlers can reach your own pages, since third-party coverage often leads the model to verify claims on your site.
How to measure whether it worked
Impressions and ad value equivalent tell you nothing about AI search. The metrics that do:
- Share of citation: how often your domain, or pages naming you, appear in sources for a tracked prompt set.
- Share of voice and position inside answers: whether you are named, and how early.
- Sentiment and the specific claims AI makes about you.
- Which third-party pages are cited, and whether your new coverage shows up in that list.
- AI-referred visits and conversions on your own site.
One caution from experience: single answers are noisy. Responses vary by run, model, and location, so pool results across many prompts and days before you report a change to a client or a CFO.
Tools for tracking coverage in AI answers
Promptwatch is the one I would look at first if the goal is to connect PR coverage to AI outcomes. It tracks which pages and third-party domains AI cites, surfaces offsite mentions where your brand is named without a link, and measures AI-referred traffic, so you can see whether a placement turned into a citation and then a visit.

Other tools worth knowing, depending on your stack:


| Tool | Strength for PR measurement | Best for |
|---|---|---|
| Promptwatch | Citation tracing, offsite mentions, crawler logs, AI traffic attribution, content creation | Teams that want to act on the data, not just watch it |
| Profound | Enterprise monitoring and analytics | Large brands with big budgets |
| Ahrefs Brand Radar | Brand mentions inside an SEO suite | Teams already in Ahrefs |
| Peec AI | Straightforward prompt tracking | Smaller teams starting out |
| Otterly.AI | Affordable brand monitoring | Budget-conscious monitoring |
For wider lists, the GEO software directory at bestgeosoftware.com covers the category.
If you would rather hand the work to people, an agency that combines digital PR with generative engine optimization can run this end to end. 1001 SEO Media is the agency behind this site and offers digital PR, link building, and AI search work with senior specialists and no long-term contracts.
Which model should you invest in?
| Your situation | Lean toward | Why |
|---|---|---|
| B2B SaaS with a crowded category | Digital PR with traditional-style editor relationships | Listicles and comparisons drive commercial citations |
| Established brand with weak AI sentiment | Traditional PR | Framing and credibility change what AI says |
| New brand with little coverage | Digital PR data studies | Gives outlets a reason to name you repeatedly |
| Local or multi-location business | Mixed, plus reviews and listings | Reviews gain share on branded prompts |
| Team reporting on link metrics only | Switch the scorecard first | Links alone predict little about AI visibility |
The bottom line
Earned media wins in AI search. Whether it comes through a rolodex or a data campaign matters much less than whether it lands on pages that name you, in formats that keep getting cited. Chase comparisons, listicles, and reviews in addition to news. Measure citation share across engines rather than one lucky answer. And run the work as one program, because the people who write the roundup AI cites do not care which department emailed them.

