Open ChatGPT and ask it which company a buyer in your category should choose. Read the answer closely. Most of what it says about you was written by somebody else.
We wanted to know how far that goes. At Rankfor.AI we looked at 119,043 links that AI tools used while answering questions about 66 brands, in 12 languages across 11 markets.
For the Finnish brands the split was blunt: 81% of the links pointed to websites the company does not own. That is 11,735 links out of 14,487. Their own pages accounted for the rest.
So you control about a fifth of your own story. Wikipedia, market data sites, competitors, national broadcasters and job boards fill in the other four fifths.
Here are five content gaps we found in that data, and the press coverage that closes each one. If you run a blog and a media budget, this is where the money leaks.
What a content gap actually is
A content gap is a question your buyers ask where nothing the AI trusts has your answer.
That sounds obvious until you count them. Take one real buyer question, then check what content exists to answer it. We find about 40 gaps sitting behind that single question. Forty smaller questions where the AI has to use somebody else’s material.
The pattern repeats for almost every brand we checked. Your own website takes the top three spots, and those are safe. Everything below them decides whether anyone else backs up what you say about yourself.
So does publishing more fix it? On its own, no. What works is being easy to find the moment someone asks, on a site the AI already trusts.
Further reading: what enterprise buyers actually ask AI for.
Gap 1: Your buyers ask in a language you never publish in
Brands publish in English. Buyers ask in their own language. Scores move in both directions when they do.
Skoda scores 5.7 points lower in its home language than in English. DNB scores 5.0 lower and Volvo 4.2 lower. Strong brands, losing ground at home, because most of their coverage sits in the wrong language.
It runs the other way for brands with strong local coverage. Vinted scores 9.3 points higher at home, CD Projekt 8.8 higher, ESET 7.7 higher.
Local media does get used. Estonian public broadcaster ERR shows up 558 times in our data, along with Latvia’s LSM and Lithuania’s LRT. A story in a national outlet, in the local language, does something an English press release cannot.
Further reading: where Polish buyers start with AI.
What to do: pick your three biggest markets and get one story into a national outlet in each, in the local language. Putting a translation on your own website does not count. The link has to come from their domain.
Gap 2: Nothing independent backs up your claims
AI can read every claim on your website and still have nothing to check them against.
Wikipedia, Statista and YouTube together make up 11.1% of the outside links for the Finnish brands in our data. These are the places AI goes when it wants a second opinion about you.
So how do you get into an encyclopedia? You do not, and that is the point. Wikipedia editors and market data analysts use what has already been published somewhere else. A number you place in a trade magazine today is what reaches them months later, so you are always working one step behind where the link finally appears.
A number a journalist chose to quote counts for more than the same number on your own homepage. That is the simplest case for press coverage here. You are giving the AI a second source, and a second source is what it is missing.
Further reading: three AI platforms, 24 brands, no agreement and the B2B owned media advantage.
What to do: take the three things you most want AI to say about you and get each one published somewhere you do not own, with a number attached. Numbers travel between sources. Adjectives stay home.
Gap 3: AI describes your category using your competitors
Ask AI about Printful and part of the answer comes from Printful’s rivals.
AI learns a category as a group, so the brands next to you help explain you. Here is what that looked like in our data:
· Printful was described using gelato.com, used 241 times, and printify.com, used 228 times. Both are direct competitors.
· Kone was described using TK Elevator, used 131 times, and Otis, used 116 times.
· Nokia was described using Ericsson, used 112 times.
Across the Finnish brands we looked at, competitors filled four of the top ten sources.
Read that again as a PR person. When a buyer asks AI about your company, a good part of the answer comes from pages your competitors wrote.
The fix is dull, and it works. Skip the category roundups and comparison articles and AI still describes your category. It just uses whoever did show up.
Further reading: why self-promotional listicles win in ChatGPT.
What to do: find the roundups and comparison pages that already cover your category, then ask to be included. A page AI already reads beats a new page of your own that it has never seen.
Gap 4: AI is still describing the company you were two years ago
You relaunched in March. AI still talks about the old company. Why?
Answers about your brand come from two places, and they run on different clocks. About a third comes from pages the AI looks up while it writes the answer. That part updates in days or weeks, and normal SEO work moves it.
The other two thirds come from what the model already learned during training. That only changes when the model is retrained, every three to six months. The first part is paint. The second is concrete.
To change the concrete, the same message has to appear in the same words across a lot of different sites. One launch will not do it. Neither will one very good article.
Further reading: what Google’s grounding leap means for CMOs.
What to do: report on two clocks. Judge the fast half in weeks, give the slow half two quarters, and keep your key wording the same in everything you publish in between.
Gap 5: A small group of sites carries most of the answer
The sources are not spread evenly, and the gap is wider than most teams expect.
We counted 131,667 links across 21,077 different websites. Fifty sites carried a quarter of everything, 24.5%. A thousand sites carried nearly two thirds, 62.3%. The remaining 20,000 sites shared what was left.
Where you get published matters far more than how often. Most media lists are built without anyone ever checking this.
Then look at which sites sit at the top, and a second problem appears. One in every nine sources AI uses to describe Polish brands is a job board. Nobody in marketing owns those pages, so recruitment ads end up answering questions about what a company sells.
Further reading: the job boards describing Polish brands.
What to do: before your next buy, list the sources AI actually uses in your category and compare it with your media list. Add the sites that keep showing up. Drop the ones that never do.
Five things to put in your next brief
When you commission an article or write a release:
1. Say what category you are in. AI needs the words that connect you to the question. “Payment software for Baltic online shops” works. “Innovative fintech solutions” does not.
2. Give one number worth repeating. Journalists and AI pass along the same thing: a specific figure someone can point to.
3. Write your company name the same way every time. Three spellings turn one brand into three weak ones.
4. Publish in the buyer’s language, next to the English version.
5. Check search engines can read the page. If crawlers cannot reach it, AI will not see it either. Ask whether the outlet republishes to partner sites, because that multiplies the effect.
Further reading: a PR guide for AI validation in Northern Europe.
Try this before you buy the next placement
The gap that matters is the distance between what you say about yourself and what anyone else confirms. All five findings describe that same distance from a different angle: wrong language, no outside proof, missing from the category, too new to have landed, or published where AI does not look.
You can check your own position this week. Ask ChatGPT, Gemini and Perplexity the question your buyers really ask, in their words: “Which company should I choose for [your category] in [your market], and why?” Then ask each one to list the sources behind its answer.
Run it five times on each. The answers change from run to run, and the sources that keep coming back are the ones holding your reputation.
Now count how many of those sources you own. If the number lands near a fifth, the rest of the work belongs to your media budget.
That check covers one question. Your buyers ask hundreds. Handling that scale is what our platform does. It maps the questions your target customers really ask, shows which sources feed each answer, and turns the gaps into a content plan a writer or an agency can work from.
The part that matters most here is the Answer Trail. It records the searches an AI runs before it names any company. In one run this July, a single question about accounting firms set off 28 more searches before the answer appeared.
Every source in that trail that leaves you out is one clear opportunity, with a name on it. You can see your own Answer Trail in about two minutes.
Method: Rankfor.AI Index 2026, Nordic-Baltic and CEE Edition. 35,640 AI answers collected across 66 brands, 12 languages, 11 markets and 3 AI models, plus an analysis of the 119,043 links behind them. Google’s internal grounding URLs, 12.9% of the raw links, were removed before every count above. Full method and results at open.rankfor.ai/index-2026.
Dmitrij Zatuchin, PhD is founder and CEO of Rankfor.AI and a researcher and lecturer at Estonian Entrepreneurship University of Applied Sciences. He holds a PhD in Computer Science and publishes peer-reviewed research on how AI models perceive and recommend brands.
