Why is AI recommending our competitors?

“Why does ChatGPT recommend our competitor and not us?”

Avisible hears this question every week. Usually someone in the company has tried it: they typed in the kind of question a customer would ask, and the answer listed three companies. Theirs was not one of them.

The first explanation people reach for is that the competitor is bigger, spends more on marketing or has somehow “gamed” AI. In Avisible’s data, that is rarely the reason.

Since October 2025, Avisible has run about 90,000 AI queries: hotels and hotel chains, gym chains, consumer brands and online stores, industrial companies, professional services, congress organisers and non-profits, in the Nordics, the DACH region, the UK and the US. AI seldom recommends a competitor because it thinks the competitor is better. It recommends the competitor because the competitor was there first: on the shortlist, in the sources, and in the words of the question.

In Avisible’s data, the competitor makes the shortlist for these reasons:

1. The competitor answered the question. You didn’t.
2. AI has given the competitor a lane. Your buyers ask from someone else’s lane.
3. The competitor is in the sources AI reads. You are not.
4. The competitor speaks the buyer’s language. A translated website is not enough.
5. The competitor is there when you are not. Opening hours, resellers, prices.

The reasoning follows – but first, a surprise.

When Avisible asks AI to compare a brand with a named competitor, the brand often wins.

A marine services firm was compared with its five main Nordic competitors in 119 answers. In 76 answers AI named no winner. When it did choose, it picked the firm 41 times and the competitor twice. An industrial sensor manufacturer did even better: AI named a winner in 93 comparisons and chose the manufacturer’s product in 77 of them.

Yet the same marine firm appeared in only 7% of open questions in Sweden and 9% in Denmark (base: 90 answers each), meaning questions where the buyer describes a need without naming any company. The same sensor appeared in 7% to 41% of open questions, depending on the AI system. In those questions, AI named local competitors instead.

So the competitor usually doesn’t beat you in the comparison. It beats you one step earlier, when AI builds the shortlist. Avisible calls that difference the discoverability gap: the company is named when it is asked about by name, but missing from the list when a buyer describes a need without naming anyone. If you are not on the list, the buyer never gets to ask for the comparison.

This does not always hold. A tourism operator lost almost every head-to-head against a smaller boutique rival: 0–7, 1–8 and 0–9 in three languages (base: 30 answers per language; the rest named no winner). There, the reason was the lane AI had given each company. More on that below.

AI can only recommend what someone has written down. When a buyer asks about a specific situation, AI looks for a page that answers it and names the company behind that page.

An industrial equipment maker was named in only 3 of 9 answers about asphalt recycling. The answers named two competitors that have separate application pages for recycling. The marine firm lost in Denmark to a local surveyor whose website lists, plainly, which surveys it does and which ports it covers. The firm’s own Danish office page said neither, and AI did not cite it once.

The competitor didn’t do better work. It wrote down the answer to the question buyers ask.

AI divides every market into roles. In one Finnish city, AI describes one hotel as “the energetic city hotel”, a second as the calm and atmospheric one, and a third as the one with the best sleep. When a traveller asks for a quiet hotel, the energetic hotel is not named, 0 times out of 9.

The same happens everywhere. A premium gym chain is “the gold standard for premium fitness”, so questions about affordable gyms go to budget chains. The marine firm is “the one with the large international network”. Its competitors are “the local specialists”, and AI gave that reason 68 times. Local questions go to local specialists. The tourism operator’s own message, “the largest in the Nordics”, became “large, commercial, at times touristy” in 46% of answers about guest reviews. So small-group experiences went to the boutique rival.

A lane is not a problem in itself. It becomes one when your buyers ask questions from someone else’s lane.

When the question names you, AI reads your website. When it doesn’t, AI reads everyone else’s.

In Sweden, the marine firm’s main competitor’s own site was cited in 41 of 89 answers. The firm’s site was cited in 3. A hotel’s competitors’ websites together collected more than twice as many citations as the hotel’s own site. For a tourism operator, the most-cited source in every market was an activity marketplace, cited in about a quarter of answers. The operator is not listed there, so AI recommended the operators that are.

Reddit, local city guides, booking sites, resellers, industry associations and “best of” lists are where the shortlist is written. If the competitor is there and you are not, AI takes the competitor’s word for it.

The tourism operator beat its main rival in English (11–6) and German (12–7). In French it lost, 4–9 (base: 45 answers per language). When French buyers asked for a reliable operator in the region, AI named a small French-speaking company. That company was not on anyone’s competitor list. The tourism operator was named in 7% of those answers.

Every language is its own market, with its own sources: a French travel guide, a German tour catalogue, a Swedish trade directory. A translated website is not enough if the sources AI reads in that language don’t mention you.

Some of the “competitors” AI recommends are not really competitors. They are simply available.

A helpline that is open on weekday evenings loses the night and weekend questions: AI sends those people to services that are open around the clock. An industrial manufacturer’s product was named in buying answers, but 67 of 71 of those answers sent the buyer to a reseller, and only 22 named the manufacturer’s own web store. None of 448 answers gave the store price, while a reseller’s lower price was quoted in 55. A pet gear brand shipping from Finland watched AI suggest large local pet retailers as the “easier” option in Germany.

The buyer was ready to buy. AI just sent them somewhere else.

The strangest “competitor” in Avisible’s data is the brand itself.

A gym chain renamed its clubs. Under the new name it appeared in 6.6% of answers. Under the old, retired name it appeared in 72.5%. When people asked about the new name, AI took the facts from the chain’s Norwegian and Swedish websites. An industrial manufacturer’s own sister product was named more often than the product it most wanted to sell (67 vs 61 answers in ChatGPT).

And the competitor AI names is often not the one in your strategy deck. One industrial company’s main tracked competitor did not appear in any of 443 answers. Three brands nobody was tracking appeared in about half of them.

Almost none of these reasons is about being better. The competitor didn’t win on quality, price or reputation. It won because it was written down, in the right place, in the right language, at the right moment.

That is fixable. Writing an application page, listing the ports you serve, getting into the local guide, publishing your own price, telling AI where to go when you are closed: none of these needs a new brand or a bigger budget than your competitor’s.

Ask the same AI three questions.

First, compare yourself with your main competitor by name: “Which is better for [need], [you] or [competitor]?” If AI picks you or calls it even, your reputation is not the problem.

Then describe your customer’s need in their own words, without naming any company: “I need a marine surveyor for a cargo claim in Gothenburg.” “Where can I find a quiet hotel in the city centre?” See who is named.

Finally, look at the sources under the answer. That is where your competitor lives, and that is where you need to be.

1. List the companies AI actually names, not the ones in your plan. Your real competitor in AI may be a local specialist, a marketplace, a reseller or your own old name.
2. For every question a competitor wins, open the source AI cites. Find what it says that your pages don’t: the application, the location, the price, the fit. Then write it.
3. Choose your lane on purpose. If you want the questions from the neighbouring lane, write about them. AI will not give you the quiet-hotel questions if you have only written about energy.
4. Be where AI reads in each market: marketplaces, city and travel guides, industry associations, directories and review sites, in each language you sell in.
5. Close the leaks at the moment of purchase: your own price, where to buy, opening hours, and what to do when you are closed.

The order matters. A comparison page won’t help if you never make the shortlist.

The findings are based on Avisible audits and Snapshots run between October 2025 and September 2026, about 90,000 AI queries in total, in hospitality, retail, consumer brands and e-commerce, industrial B2B, professional services, congresses and events, and non-profits, in the Nordics, the DACH region, the UK and the US, in English, German, French and the Nordic languages. Each audit asked 100–200 questions, repeated 3–5 times, in ChatGPT, Gemini and Google AI Overviews. Some questions named the company, some named the company and a competitor, and some named no company at all.

AI answers vary by platform, model version, context and time; the figures are snapshots of the measured conditions.

Read also: Why doesn’t AI search find our company? and How to improve your AI visibility

Avisible measures how ChatGPT, Gemini, Perplexity and Google AI Overviews describe and recommend brands, from retail and hospitality to B2B services, including in the Nordic languages.

The Visibility Snapshot uses 900 AI queries to show who AI recommends instead of you, which sources it reads and why. Delivered in three working days, €499.