Where does AI look to decide what to think about you?
Let's start with the first thing I ran into while fact-checking. It's the number every conference is quoting right now. Analytics firm Muck Rack went through more than a million links that AI tools cite in their answers. The result: 95% pointed to non-paid sources. Journalism made up 27% of all citations, and for questions about current events, as much as 49%. AI also loves freshness. It often cites content published the day before the question.
To be fair, while checking I also found the other side: Muck Rack's "non-paid" bucket also includes Wikipedia, Reddit, academic research and government sites. Critics in the industry warn that the headline number may push companies to overinvest in media relations. But for my argument, that's even better news. AI doesn't just listen to journalists. It listens to everyone except you and your ads.
And there's fresher data: the May 2026 edition of the same study analyzed more than 25 million links from ChatGPT, Claude and Gemini. Earned media made up 84% of citations, journalism 27% and paid content just 0.3%. The platforms differ, though: ChatGPT cites sources in 96% of answers and most often Wikipedia, Claude in 55% and most often PubMed Central, Gemini in 82% and most often Reddit. If you want to show up in AI answers, you can't bet on a single source.
And here's the uncomfortable truth. ChatGPT can't read a billboard by the highway. It doesn't watch your TV spot either. It skips the banner on a news site just like you do. If you've spent the last ten years putting most of your budget into advertising and treating PR as "what someone in marketing does when there's time left," you now have an almost empty profile in the AI world.
Why do more robots than people read your website?
The second thing shifting under our feet. According to the Thales (Imperva) Bad Bot Report 2026, automated traffic made up 53% of all web traffic in 2025. Machines, not people, now make up most of the internet's visitors. To be precise, though: 40 percentage points of that are malicious bots, like the ones trying to crack a login form a thousand times a minute. On its own, the number says nothing about how AI assistants read websites. It just says more machines than people read your site today. And some of those machines are search engines and AI tools reading it so they can answer questions about you.
People, meanwhile, click less. In 2025, Pew Research tracked 900 American adults and found that when Google showed them an AI summary, they clicked a link in only 8% of searches. Without the summary, 15%. Google disputes the methodology, but anyone looking at their own analytics sees the trend.
So your website now has two audiences. People, of whom there are fewer and fewer. And robots, of whom there are more and more, and who then talk to people about you.
On your website, AI acts like an ad blocker
Now the thing I promised at the start. In my experience, AI treats two kinds of sources very differently.
In the media, reviews and discussions, it looks for story, reputation and opinions. There, the "marketing fluff" matters, because that's how it figures out what others think of you.
On your own website, it acts like an ad blocker. "We're market leaders with a passion for innovation" gets skipped. It looks for clean, concrete information: what you do, what it costs, who it's for, where to find you and how fast you respond.
The research backs this up. In 2023, researchers from Princeton and other universities published the paper that coined the term GEO (generative engine optimization). They tested 10,000 queries to see what makes AI more likely to use your text. Citing sources, quotations and concrete numbers helped most. Visibility in AI answers rose by up to 40%, and on the live Perplexity engine by up to 37%.
What that means for your website:
| Instead of fluff | Put this there |
| "A personal approach to every client" | Who exactly it's for, and who it isn't |
| "Fair prices" | A price list, or at least a range |
| "Years of experience" | How many years, how many clients, what results (with numbers) |
| "We're here for you" | Opening hours, address, how fast you reply |
| A poetic intro about your vision | Answers to the questions people actually ask (FAQ) |
By the way, that's why every Don Marketer article has a quick answer at the top, tables, sources and FAQs at the end. Not for the people who read only headlines. For the robots that read everything.
How to do PR in every direction (even without an agency)
Fine, AI listens to others. How do you get them talking about you? Here's my "PR in every direction" framework:
- Expert commentary for the media. Journalists look for people who can say one smart sentence with a number. Offer yourself as a source to trade publications and local media.
- Your own data. Nothing gets cited as eagerly as a number nobody else has. A small customer survey, a summary of your orders, a local price comparison.
- Podcasts and webinars. Their transcripts and descriptions end up in the sources AI draws on.
- Reviews and ratings. Google reviews or listings on comparison sites are PR written by your customers.
- Discussions (Reddit, Quora, LinkedIn). A useful answer that discloses who you are is PR. An ad dressed up as advice is spam.
- Consistency. AI loves fresh sources. One press release a year is one ball a year. It won't reach the target.
And one warning: Wikipedia is not the place to write an entry about yourself. The community will delete it, and the damage will outweigh the benefit.
For a hair salon: her PR is a mention in the local newsletter, a piece about her new salon on the town website, thirty honest Google reviews and helpful answers in the local Facebook group. And on her website: prices, opening hours, address, parking and online booking. No "passion for beauty." When someone asks AI for "a hair salon nearby with an opening on Saturday," she has a shot at being in the answer.
For a corporation in Miami: PR is no longer "crisis comms and press conferences." It's infrastructure for AI. Company experts in the media and on podcasts, original research, up-to-date facts on the website and clear structured data. Treat it as a separate department that "sends a press release now and then," and you'll lose to the competitor people simply write about more.
Expert perspective: "Advertising buys you people's attention. PR buys you the robots' trust. And those robots now decide who gets recommended to people. If you spent years betting only on ads, your résumé in the AI world is almost blank," says Michal Krčmář, founder of Don Marketer and a CMO with nearly two decades of experience in global marketing.
The takeaway
Go back to the doodle from the start. Stick figure, hill, target. If you throw only one kind of ball, say just advertising, most of them roll off to the side. If you throw all three, a fact-driven website, third-party mentions and active social media, and you throw them consistently, you start hitting the target.
And a little test for tonight: ask ChatGPT or Gemini "who's the best [your industry] in [your city]." If you're not there, you know where to start. If you are, look at which sources the AI knows you from. It usually won't be your website.
And if you're figuring out how much of your budget should go to advertising and how much to PR and content, the Don Marketer calculator works out the whole mix in 2 minutes, including a recommended share for content and PR.
Sources: Mediaboard and StoryMatters webinar "Ví AI o vaší značce? A doporučuje vás?" (November 5, 2026, Sergej Pavljuk – the "hill and target" doodle); ContentGrip: What Muck Rack's "What is AI Reading?" report reveals about AI brand visibility (Muck Rack, 2025); PR Daily: The 84% myth: Why the PR industry's AI numbers are wrong (methodology critique); Thales (Imperva) Bad Bot Report 2026 (automated traffic 53% in 2025, malicious bots 40%); Muck Rack: What Is AI Reading?, May 2026 (25M+ links, earned media 84%, paid content 0.3%, platform differences); Pew Research Center (July 2025, 900 US adults, 68,879 searches); Everything PR: We found the paper that launched GEO (Aggarwal et al., GEO: Generative Engine Optimization, arXiv 2311.09735, 2023); the author's own experience.