Why doesn’t AI search find our company?

“Why doesn’t AI search find our company?”

The question has become common fast. The usual answers come in three flavours: AI doesn’t recognise the company as a trusted entity, the website content is in the wrong shape, or the site lacks structured data.

All three are real causes. They are not equally common, and they are not equally heavy.

We have measured close to 10,000 AI queries across six industries – a hotel, a tile retailer, a pet gear brand, a congress organiser, a non-profit and a professional services firm. The most common cause of invisibility was none of the three. It was simpler.

Most of the time the company is found. Ask ChatGPT, Gemini or Google AI Overviews about it by name and they can tell you what it does. In all six audits the company appeared in 92–100% of queries that named it – regardless of industry, language or company size.

The problem is elsewhere. When a buyer describes their need without naming any company, AI builds a shortlist – and the company is missing from it. In open questions the mention rate ranged from zero to just over sixty percent.

So “we are not found” almost always means: we are not recommended when the buyer doesn’t yet know us. That is a different problem from awareness, and it has different causes.

This is true for companies that nobody writes about except on their own site. AI cross-references, and it won’t recommend a company whose existence rests on a single source.

We saw a genuine entity problem once in six: an international congress organiser whose abbreviation was confused with a technical university in more than 30 source links. That is a real defect, and it is fixed with Organization schema and sameAs references – not with content.

The other five companies were well known. They were still missing from the shortlist in a large share of open queries.

Entity authority is the ticket in. It is not a seat on the list.

AI reads pages in chunks and pulls into its answers the passages that address the question directly. A long story or a generic service description doesn’t offer such chunks.

This is closer. But shape was not the core problem. The core problem was what the pages said at all – and what they left unsaid.

Schema markup helps AI understand what a company is, where it operates and what it offers. It is worth doing. But two of the sites we measured already had 100% schema coverage, and their open questions still had gaps.

Schema tells AI that a company sells tiles or provides interim finance. It doesn’t tell AI where to get everything for a bathroom renovation from one place, or what outsourcing a controller costs.

Buyers ask the latter.

The clearest example comes from interim finance. The firm appeared in 95% of queries that named it and 14% of open ones. Of the 50 open buyer questions in that dataset, 31 produced no mention at all – in any system, on any run. 279 queries, zero mentions.

The dividing line ran through the words. When the question used the name of the service category, AI found the firm. When the question described the buyer’s situation – price, restructuring, a growth phase, hiring a controller – AI named others, or no one.

This is not a technical problem and not a reputation problem. It is a content gap.

The interim case is not an outlier. The same pattern appeared in every audit where we asked both by name and in the buyer’s words.

A tile retail chain was named in 98% of queries that used its name. In open questions the result depended on whose words the question used: with the industry’s own term “tile shop”, 69%; in the renovator’s own words (“where can I get everything for a bathroom renovation from one place”), 47%; when asking about supplies and fittings, 33%. Same company, same week, same counting rule. Only the words changed.

A hotel in Oulu was named in 95% of queries by name and 63% of open ones. But when a traveller asked for a quiet hotel in Oulu, the hotel was missing from the list 9 times out of 9 – even though it has quiet courtyard rooms. Nobody had written about quiet, so AI credited it to a competitor.

An international congress organiser was named in 92% of reputation queries and 96% of purchase-stage queries. When a buyer asked what a professional congress organiser actually does, the company was named 0 times in 900. So was every competitor. The top of the category was empty – there for anyone to take.

A pet gear brand was named in 100% of brand queries and 100% of cold-weather gear questions across three markets. In questions about daily walks, harnesses and care, its share was 0%. AI had pigeonholed the brand into one use case, because that was the only one anyone had written about.

The gaps are in the same topics. Price was missing in three audits out of six: the interim firm’s price questions, zero mentions; the hotel’s “affordable hotel”, 0 of 9; the congress organiser had no pricing page at all. AI cannot quote a price that hasn’t been written. The same goes for fit-to-situation and practical delivery.

The shortlist is written elsewhere. When asked by name, AI reads the company’s own site; in open questions it reads everyone else’s. The interim firm’s site collected 48% of source citations in queries by name and 3% in open ones. For the tile retailer the figures were 55% and 11%. The hotel’s site was the most-cited source by name and only the fifth most-cited in open questions – behind Google, the local tourism sites and booking.com.

That is why we are comfortable saying the content gap is the most common cause. Not the only one, but the most common.

Of the three causes, the content gap is the cheapest to fix.

Building entity authority takes years. A technical site overhaul takes months and budget. Closing a content gap is writing work on pages that already exist.

In each of the three Finnish audits, all six top actions were of this kind. None required a site redesign or a developer. The hotel’s quickest fixes were 30-minute and one-hour jobs.

Run two queries in the same AI.

First: “What does [your company] do?” If the answer is accurate and specific, you don’t have an entity problem. AI knows you.

Then: describe your customer’s situation the way they would say it, without naming any company. “I’m looking for a quiet hotel in Oulu.” “Where can I get everything for a bathroom renovation from one place?” “We need an interim CFO for the duration of a restructuring.” See who AI names.

If the first works and the second doesn’t, your problem is the same one the companies we measured had. You don’t need to become better known. You need to write down the things buyers ask.

1. List the buyer’s questions in the buyer’s words. Not “interim finance” but “our CFO just left, what now”. Not “tile shop” but “everything for a bathroom renovation from one place”. Write direct answers to them.
2. Put your pricing principles in writing. Whatever you leave unsaid, AI decides for you – or names the one who said it.
3. Spell out which situations the service fits and which it doesn’t. AI fills the gaps with its own reasoning – and pigeonholes you into whatever you have written about.
4. Only then: schema, third-party mentions, industry association pages, directories. They reinforce what you have already written. They don’t replace it.

The order matters. Authority without content produces awareness without recommendations.

The findings are based on six Avisible audits run between February and August 2026: three Finnish audits of 900 queries each (professional services, hotel, tile retail), two international audits of 3,600 queries each (a congress organiser in English and German; a pet gear brand in the US, UK and DACH), and one non-profit audit.

In the Finnish audits every question was put to three AI systems – ChatGPT, Gemini and Google AI Overviews – three times; the international audits used two or three systems. In every audit, some questions were open and some named the company.

AI answers are dynamic and vary by platform, model version, user context, location and time. These figures are snapshots of the measured conditions, not a guarantee of what an individual user sees.

Our previous article introduced this pattern with a single case: The Discoverability Gap.

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 where AI finds your company, who it recommends instead, and which of the three causes is yours. Delivered in three working days, €499.