AI Doesn’t Believe Your Press Release

Let’s start with the obvious: people do their research on ChatGPT, Gemini, and Perplexity instead of typing queries into Google. Your prospective customer is doing it. The engineer you’re trying to recruit is doing it. Your mother is probably doing it too.

When a model decides which companies lead a category, it isn’t relying on a company’s homepage or latest LinkedIn post. It pieces together an answer from reporting, analyst research, reviews, and other third-party sources across the web, giving more weight to the ones it considers credible. Researchers at the University of Toronto found this pattern across ChatGPT, Gemini, and Perplexity regardless of how they phrased their questions. Muck Rack’s analysis of more than 25 million AI citations reached much the same conclusion: roughly 84% came from earned media, while paid and advertorial content accounted for about 0.3%.

That creates an uncomfortable tension with where much of PR has gone over the past decade.

Companies have invested heavily in channels they own: founder blogs, LinkedIn thought leadership, branded content hubs, sponsored articles. Those assets are predictable, scalable, and easy to approve. They’re also exactly what AI treats with the most skepticism. When a company describes itself, the model recognizes the source for what it is. An independent publication making the same point carries more weight.

The pattern extends to paid content. Models generally distinguish editorial reporting from advertorials and syndicated marketing, and they rank them differently. A business with dozens of sponsored placements but little independent coverage is often represented less favorably than one with fewer articles in publications the model consistently trusts.

Recency matters, too. Around half the sources cited in AI answers were published within the previous year, according to Muck Rack. A major profile from three years ago still counts, but it has much less influence over what today’s models retrieve. Consistent coverage has become more valuable than occasional headline moments.

Keep in mind that accuracy remains a separate problem. Researchers at Columbia’s Tow Center for Digital Journalism found that eight leading AI search tools regularly cited the wrong source, merged separate articles, or confidently attributed claims to the wrong publication. A company can end up associated with statistics it never shared or messaging that no longer reflects the business. The practical implication is straightforward: you can’t edit the model’s output directly. The only durable way to improve it is by improving the information the model reads.

A few practical implications follow:

Treat owned content as infrastructure.

Keep your newsroom current, technically clean, and easy to crawl. It’s the one source you fully control, and AI systems will reference it. Just don’t expect it to carry the narrative. Most of what shapes an AI answer comes from somewhere else.

Be deliberate about distribution.

Newswire distribution still matters, but for a different reason than it once did. Publishing through GlobeNewswire, PR Newswire, or Business Wire places verified information on established, machine-readable domains that AI systems regularly crawl. That gives a release more visibility than publishing it only on your own site.

It still isn’t earned coverage. Press releases account for roughly 1% of AI citations. Independent reporting remains the primary source, according to Muck Rack and Search Engine Journal. Think of the wire as making verified facts available, not as replacing journalism.

Target the publications AI trusts.

Broad syndication contributes surprisingly little. Models rely on a relatively small group of publications they consistently regard as authoritative within a category. One article in the right publication often outweighs dozens of syndicated copies.

Muck Rack found that the journalists PR teams pitch most overlap with the journalists AI engines actually cite only about 2% of the time. Those aren’t necessarily the same people. Outreach often follows audience size, existing relationships, or beat coverage. AI favors sources that repeatedly prove reliable on a particular subject.

In practice, that usually means three groups: national news organizations and wire services that establish factual events; specialist trade publications with deep expertise in your industry; and third-party validation sources such as analyst firms, review platforms, or respected community forums. Prestige matters less than topical authority. A niche publication that owns your category can outweigh a glossy business feature.

Keep the story consistent.

Models learn what a company is through repetition. If different articles describe you in different ways, they’ll build a less coherent picture. Message discipline has always mattered. AI simply makes the consequences easier to see.

Ask GPT too! 

Ask AI systems what they know about your company the same way you’d conduct a media audit. Look for outdated positioning, factual errors, weak sourcing, or competitors dominating the conversation. Those outputs are symptoms, not the problem. The solution isn’t asking OpenAI to update its model. It’s producing stronger, fresher, third-party reporting.

At the end of the day

Turns out AI is a pretty old-fashioned PR critic. It doesn’t care much for content you wrote about yourself, no matter how polished it is. It prefers the same thing people always have: someone else saying you’re worth paying attention to.