TL;DR
- Ask ChatGPT who the best company is in your field. If a competitor shows up and you don’t, the reason is often not your product. It’s what other people have published about you.
- When AI assistants answer questions that need fresh information, they search the web and build the answer from a handful of sources. Those sources decide what the AI says about you.
- In Muck Rack’s May 2026 study of ChatGPT, Claude and Gemini, 84% of cited links came from earned media: coverage nobody paid for. Journalism alone was 27%.
- A mention of your brand in the right context now matters more than a link without one. Ahrefs found brand mentions correlate about three times more strongly with AI Overview visibility than backlinks.
- Digital PR is no longer just about your image with people. It decides whether you appear in the answer at all. Five steps to start are at the end.
Open ChatGPT, Perplexity, Gemini or Google’s AI Overviews and ask: “Which is the best company for [your service] in [your city]?”
You usually get three or four names. If yours isn’t among them and a competitor’s is, most business owners say the same thing: that’s unfair, their product is weaker and their service is worse.
Maybe. But online, quality that nobody has written about doesn’t exist. If your strengths aren’t communicated and spread through real signals from real people, the information never reaches the clients who haven’t yet tried you for themselves.
What I see from years in PR
I’ve worked in PR with brands of very different sizes. Big brands rarely have a problem with PR budgets. Small and mid-sized ones usually hit one of two walls.
The first is that they misunderstand what PR is. They picture a campaign, rumours, a scandal. The second is that they understand it perfectly well, and still never give it a budget, time or anyone’s attention.
Meanwhile, their competitors quietly collect the reviews, the recommendations and the orders, because they know how digital PR works and invest in it consistently.
Recently I worked with a large brand that still had negative reviews online from 2019: products breaking, poor quality. They were part of a targeted smear campaign. In almost seven years, the brand had not taken a single step to answer them, to show that the problem was no longer real, or to explain what had happened.
The reviews were still there. And anyone searching for the brand, person or AI, kept finding them.
From newspapers to AI answers
When I started in PR, monitoring meant going through newspapers and magazines by hand to see what was being said about a person or a business. Then came press releases sent by email, followed by waiting for coverage. Sometimes paying for it.
Today many companies run their own channels, the so-called owned media. They don’t depend only on journalists, and they use many techniques beyond paid placements.
One thing hasn’t changed. One of the strongest signals that a brand is legitimate, in Bulgaria and everywhere else, is that an established outlet or a respected journalist in the sector has interviewed the company, covered its event or praised what it achieved.
What has changed is what that signal does. Until recently it was a matter of image. Today it’s a matter of whether you exist in the answer when a client asks AI for a recommendation. It may sound a little metaphysical, but we now need a good reputation with AI, not only with people.
How AI decides who to cite
An AI assistant has two sources of information. The first is what it learned during training: a huge body of text collected up to a certain date. The second, used more and more, is a live web search.
When you ask for something current, such as a hotel, a restaurant, reviews or a provider, the assistant searches, picks a few sources and builds its answer from them. That’s why answers in ChatGPT increasingly come with links underneath. AI is often criticised for “hallucinating”, so each new model tries harder to ground its recommendations in sources it can point to.
That makes the authority of the source decisive. A journalist’s or a known expert’s words carry far more weight than the same words in an anonymous comment on an unknown site. Media outlets are cited often for exactly this reason: they are seen as credible and relatively objective.
The sources aren’t decoration under the answer. In Muck Rack’s research on how AI reads the web, switching web search on or off noticeably changed what the AI said. If the AI finds nothing about you, it has nothing to say about you.
What the data says
Muck Rack, a PR software company, has published several editions of its “What Is AI Reading?” study. The May 2026 edition, based on more than 25 million links cited by ChatGPT, Claude and Gemini, found that:
- 84% of citations came from earned media: journalism, research, government sites, encyclopedias and third-party content about a brand.
- 27% came from journalism alone.
- For questions about industry trends, AI reaches for news coverage much more often.
Press releases, which many companies still send out in bulk, are cited less and less, mostly because ChatGPT uses them less.
To be fair: Muck Rack sells PR software, so it has an interest in this conclusion. But the results have been consistent across three editions, they match independent research, and they match what I and my colleagues see when we trace why AI recommends a particular company.
That independent research is a 2025 paper by Mahe Chen and colleagues (a preprint, not yet peer-reviewed), which compared AI search with Google. AI search services consistently favoured independent, authoritative sources over brands’ own sites and social media. Google’s mix was more balanced.
There’s a local nuance worth adding. In Bulgaria there simply aren’t many independent sources or expert sites that publish regularly. So I often see the opposite: a company’s own website becomes the main source the AI uses about it. In less crowded markets, both matter.
Why AI trusts others more than you
The logic is simple. When you write on your own Instagram, Facebook or website that you’re the leading company in your field, that’s a claim. When someone else leaves a verified review on your Google Business Profile, or a media outlet writes about you, or you win an award, the AI treats it as evidence.
AI assistants are built to give answers that can be checked. So they trust what others say about you more than what you say about yourself. People work the same way: a recommendation from a friend beats self-praise.
Mentions are the new backlinks
For years, SEO taught us to count backlinks. How many sites point to ours? Raise the domain rating, buy the tools, build the links, sometimes spam Web 2.0 and satellite sites.
Then Ahrefs analysed 75,000 brands and found that branded web mentions correlate with visibility in Google’s AI Overviews at 0.664. The number of backlinks: 0.218. About three times weaker. It’s a correlation, not proof of cause, but the practical lesson is clear.
One article that mentions your name in the right context can be worth more than several links without context. The AI needs to understand what you’re known for. As a colleague puts it, it needs to recognise you as an entity.
It learns that from the text around your name. If ten articles mention you in sentences about accounting services for small businesses, the AI will connect you with that topic, and you won’t be surprised when small businesses start contacting you. If you have a hundred directory links with no meaningful sentence around them, the AI has nothing to build that connection from.
Freshness matters more than ever
In Muck Rack’s first study, among cited news articles the most common publication date was the day before the query. Not every answer works that way, but the direction is clear.
An article from two years ago is not enough. If you want AI to mention you, you need to be present regularly, with new data, new comments and new reasons to be written about. Digital PR works as an ongoing process. One campaign won’t keep you at the top of the recommendations.
And you need balance. AI cites media mostly for questions about trends and news in an industry. For “how do I…” questions it relies more on reference content and brands’ own sites. So you need both: useful, clear content on your website that answers the “how” questions, and a presence in the media that shows you’re current and that others consider you an expert.
Five steps to start
- Check who AI cites in your niche. Ask ChatGPT and Gemini ten questions your clients would really ask. Not “who is company X”, but “which is a good company for…”, “how much does … cost”, “what should I choose between … and …”. Write down the sources under the answers. That’s your list of outlets and sites to be present on.
- Create a reason the media will want. Journalists don’t look for advertising. Advertising finds them, through the ad department. They look for stories and data. A small survey of your clients, an analysis of data only you have, or a clear position on a current topic works far better than a press release about a new product. Ask yourself: if I were a journalist, would I write about this for free?
- Become the expert people call. Offer comments on topics where you’re strong. Reach out to journalists who already cover your industry. One expert quote in the right article ties your name to the topic, and it snowballs: after the first piece, other outlets often come to you.
- Keep your information consistent everywhere. Your company name, what you do and who you work for should read the same way in the media, in your profiles and on your site. If you’re a marketing agency in one place, an ad company in another and an IT firm in a third, the AI genuinely won’t know what to say about you.
- Measure. Once a month, ask the same questions and check whether you appear and in what context. Keep a simple spreadsheet. Expect at least three months before you see real movement. After a few months, you’ll see which publications made a difference.
What doesn’t work
- Mass-sending press releases with no real information in them.
- Paid placements that are obviously commercial but dressed up as editorial content.
- Low-quality links bought by the kilo from strangers, on sites nobody reads.
I don’t recommend any of these. We can’t control what AI says. We can make sure that when it looks for information about us, it finds accurate, current and independent sources.
The bottom line
AI assistants are increasingly the first place clients go for a recommendation. When they answer, they search the web and trust what others write about us far more than what we write about ourselves. They prefer fresh sources, and they understand what you’re known for from the context you’re mentioned in.
That’s why digital PR is no longer only about image. It decides whether you’re in the answer at all, and in some sectors, whether you sell at all.
Try the test today: ask ChatGPT five questions your clients would ask, and see who it recommends.
This article is based on my video (in Bulgarian) on digital PR and AI recommendations. For a neutral explainer with the data, read Digital PR and AI Citations, Explained on Egorithms. For the mechanics of how AI search picks its sources, see How AI Search Decides Who to Cite.
About The Author
