TL;DR
AI knows the brand, but rarely mentions it without its name. With the company name in the question, AI mentions it in 94% of answers; without it, in 20%.
Each platform behaves differently. ChatGPT mentions companies most often (26%), Perplexity least (14%), and AI Mode never admitted it didn’t know something.
A specific fact on the website helps. AI takes a clearly named service or number into its answer.
AI fills in missing data, most often where the website doesn’t state it or has it only in a PDF.
In our example, unified data and a technical description of the website (schema) helped quickly; for general questions, content decides. A large e-shop’s visibility in local questions rose from 57% to 85%.
Summary
A customer can ask AI in two ways. The first is a question with the company name, for example “Is company X reliable?”. The second is a question without the name, for example “Who builds prefabricated halls in Košice?”.
In questions with the name, AI mentioned the company in 94% of answers. In questions without the name, it mentioned it in only 20% of answers. So AI knows the companies. But when a customer asks without the name, it rarely mentions them.
We draw on almost 30 AI visibility measurements from August to November 2026. AI visibility shows how often AI mentions a company in its answers. The measurements fall into three groups:
- 20 tests with 900 answers (18 tests in the standard sample) from ChatGPT, Perplexity, Google AI Overviews and Google AI Mode,
- in-depth AI visibility audits on domestic and foreign markets,
- repeated measurements before and after changes.
AI Overviews is an AI summary shown above Google search results. AI Mode is a Google mode in which AI answers instead of a list of links. We explain the standard sample in the methodology.

AI knows the company by name. When a customer asks without the name, it mentions it in only one answer in five. Source: iMPROVE AI Visibility Tests, August – November 2026.
Methodology and sample
Each test has 15 questions. We ask each question on 3 platforms, i.e. in 3 AI tools. One test thus gives 45 answers. In total, that is 20 tests and 900 answers. We measured from August to 5 November 2026.
- Questions: 2 questions contain the company name. They ask about its reliability and about customers’ experience with it. The other 13 questions are without the company name. We split them into four types:
- General advice questions (TOFU): the person asks how to choose or do something. They are not looking for a specific company yet.
- Comparison questions (MOFU): the person is already comparing options or suppliers.
- Supplier questions (BOFU): the person is looking for a company to supply the product or service.
- Local questions: the person is looking for a company in their city or region.
- Platforms: every test includes ChatGPT and Perplexity. The third platform was one of two Google AI features: Google AI Overviews in 11 tests, Google AI Mode in 9 tests.
- Standard sample: 18 tests. Two tests have a different design, so we leave them out of the averages. The first is a city transport operator with no competition. The second is a new B2B platform before market launch, where 9 questions contained the company name.
- A mention means AI names the company in its answer. A citation means AI lists the company website among the sources it built the answer from. Whether AI recommended the company first is not evaluated in this article.
Example: to the question “Who builds prefabricated halls in Košice?” AI answers with a list of companies and adds sources below the answer. If your company is on the list, you have a mention. If your website is among the sources, you have a citation.
How we build the questions. We write questions as full sentences, the way a customer asks. We don’t use keyword lists. We start from what the company sells, to whom and in which region. We add questions that clients actually get. We ask each question the same way on all platforms. We collect the answers with the Marketing Miner tool. We check mentions and sources manually.
You can order the same test for your company free of charge: AI visibility test
An example from a test for a supplier of large steel and concrete halls:
| Question type | Example question |
|---|---|
| General advice questions (TOFU) | How do I go about building a warehouse hall and which permits do I need? |
| Comparison questions (MOFU) | Compare suppliers of prefabricated halls in Slovakia. |
| Supplier questions (BOFU) | Who supplies turnkey prefabricated steel halls in Slovakia? |
| Local questions | Who builds prefabricated halls in and around Košice? |
| Questions with the company name | What references and reputation does [company] have in building industrial halls? |
| Industry | Tests |
|---|---|
| B2B manufacturing and construction | sandwich panel manufacturer (SK and PL), supplier of large steel and concrete halls, building insulation, trenchless technologies |
| Services | moving, sewer cleaning, pre-purchase car inspection, translations |
| E-shops and retail | haberdashery e-shop, sports e-shop, wedding rings, jewellery store |
| Financial services | insurance broker, online insurance |
| Education, sport, leisure | language school, fitness centre, sports predictions |
| Outside the standard sample | city transport operator, new B2B packaging platform |
All measurements were made at the request and with the consent of our clients. We do not name the companies. Even if someone recognises a company, they will not find internal data in this article. Anyone can measure AI visibility from the outside with the methodology and the tool.
Limits of the data. These are almost 30 measurements. In each industry we measured only one or two companies. So we draw no conclusions for individual industries. The percentages apply to our tests; they are not market averages. We describe patterns that hold across different measurements, platforms and industries.
AI knows the company, but rarely mentions it without its name
A question with the company name sounds like “What references does company X have?”. In such questions, AI mentioned the company in 96 of 102 answers (94%). A question without the name sounds like “Who supplies turnkey prefabricated steel halls in Slovakia?”. In such questions, AI mentioned the tested company in 139 of 708 answers (20%).
In one test, we measured the questions with the company name separately. That is why there are 102 answers with the name, not 108.

Share of answers without the company name in which AI mentioned the company (15 questions × 3 platforms). Source: iMPROVE AI Visibility Tests, 18 tests.
Highest visibility: haberdashery e-shop 46%, hall supplier 41%, moving company 36%, trenchless technologies 33%, translation agency 31%. All these companies name their services and numbers specifically on their websites. Examples: 130+ models in one category, storage price, casing diameters, translation within 24 hours.
Lowest visibility: insurance broker and online insurance. In insurance, AI sends people to insurers and comparison sites (Netfinancie, Superpoistenie).
Each industry has a different number of rivals in AI. In building insulation, even the most mentioned company has at most 8% of mentions. In insurance, one comparison site has 36%. In an industry where no company dominates AI answers, a new brand can get into the answers more easily. This holds even when it faces strong competition in the market or on Google.
Each platform behaves differently
The first difference is how often the platforms mention companies. In questions without the company name, ChatGPT mentions the tested companies most often and Perplexity least.
The second difference is in the sources AI draws on when a customer asks about a company directly. AI Overviews and Perplexity use the company’s own website as a source least often.
So adapting a website to just one platform is not enough. These are results from our tests, not a platform ranking. We measured AI Overviews and AI Mode for different companies (11 and 9 tests).

The “admits it doesn’t know” figure is indicative. Source: iMPROVE AI Visibility Tests.
Perplexity draws on the most sources. It often uses business directories and classifieds portals. Bazoš appeared in 7 tests. Example: a trenchless technologies company. Perplexity cites classified ads on Bazoš 56 times and the company website not once. ChatGPT, meanwhile, lists the same company first among the companies it names. Perplexity also confuses companies with similar names. It mixes up a translation agency with a similarly named foreign agency, and a car inspection service with a company with an almost identical name.
AI Mode never admitted in our tests that it didn’t know. It filled in missing data from whatever it found, often inaccurately. It confused a new B2B packaging platform with bottle return machines. It described a building insulation company, based on Facebook groups, as a company for monolithic structures and family houses.
AI Overviews often don’t appear. In 41 of 165 questions (25%) they didn’t appear at all, mostly in supplier questions (BOFU). That is why we switched to AI Mode in newer tests.
Specific facts win
AI mentions a company more often when it finds a sentence on its website that directly answers the question. A general list of services is not enough.
Example: a customer asks who can translate a document by the next day. A website saying “translation within 24 hours” gives AI a ready answer. A website that only says “translation and interpreting” doesn’t.
In our tests, specific information appeared in AI answers more often than general information.

Where a website clearly names a service or a number, AI mentions and cites the company. Source: iMPROVE AI Visibility Tests.
A building insulation company also lists hospitals, halls, wastewater treatment plants and retaining walls on its website. But they are there only as items in a list. When a customer asks about insulation for such buildings, AI names other companies.
General advice questions (TOFU) are close to zero almost everywhere. These are questions like “How do I go about building a warehouse hall?”. In every test where we evaluated question types, AI almost never mentioned the tested company in them (0% or just above zero). AI answers them in general terms, or draws on government websites or on the website of one competitor. Such a competitor has many articles on the topic. Example: Etalon Agency has 83 citations in translations.
Where the information sits on the website matters too. We explained why AI doesn’t see some parts of a website at all in the article AI only reads what it can see on your website.
Example: a new B2B platform. Google AI Mode correctly writes that a paid membership gives buyers data on suppliers’ financial health. But to the direct question “Can I see a supplier’s financial health?” it answers “no”.
A city transport operator has a frequently asked questions (FAQ) section on its website. All platforms answered the questions in it correctly. AI finds an answer more easily when the question and the answer are written side by side on the website.
AI fills in missing data
When AI doesn’t find a clear answer on a company’s website, it pieces one together from directories, profiles and reviews. Sometimes it makes it up. Example from our tests: AI described a building insulation company based on Facebook groups.
Most such errors can be reduced in two ways: add content to the website or unify the company data. Unifying data means having the same name, address, phone number and opening hours on the website, in Google Maps and in directories.

Most of these errors can be reduced with content on the website or by unifying data. Source: iMPROVE AI Visibility Tests.
AI also works with old and isolated reviews. For a language school it drew on reviews from 2018–2019. For wedding rings it drew on a wedding forum. For a jewellery store, on a single review on Heureka. When there are few reviews, AI keeps saying the company’s reliability can’t be verified.
AI can attribute a company’s advantage to a competitor. This happened to a language school: AI attributes its teaching method to a competing school.
Where AI gets its sources
The main source is company websites. They account for 84% of 5,009 citation occurrences in answers to questions without the company name. We count each use of a website as a source separately. By company websites we mean official company domains including their e-shops and blogs. This covers the client’s websites as well as competitors’. Directories, portals and reviews are a minority, but they appear in almost every test.

Directories are a minority of citations, but they appear in almost every test and some errors come from them. Share of 5,009 citations in answers without the company name. Source: iMPROVE AI Visibility Tests.
One website with lots of articles can dominate a whole topic. The website etalon-agency.com has 83 citations in a single test (translations). The website generali.sk has 46 citations in two insurance tests.
AI reads the website but doesn’t mention the brand. In other words: there is a citation, but no mention. In 16 of 98 answers that cited the client’s website, AI did not mention the company in the text. This happens most often in general advice questions (TOFU). Examples: “How do I recognise genuine leather?” or “What should I watch out for when choosing a hall?”. AI draws on articles by the haberdashery e-shop and the hall supplier, but doesn’t name these companies in its answer. For the hall supplier, AI drew on its website in one answer and at the same time listed six other companies.
Does it make sense to write advice content? Advice content means articles that explain how to choose or do something. Yes, it makes sense if it has a clear goal. A citation means AI considers the website a trustworthy source on the topic. In Perplexity and AI Overviews, the source is also shown as a clickable link. For AI to mention the company in its answer too, the article must connect the advice with the company’s own offer. Besides general guidance, it should include a specific product, service, price or the company’s own experience. We are still checking whether citations in advice questions later turn into mentions in supplier questions (BOFU).
Company profiles on other websites can help too. AI cited the hall supplier’s paid profile on a construction portal 15 times. In all these answers, it also mentioned the company.
Selling abroad too? Measure each market separately
We measured two clients twice: on their domestic market and on the market of their foreign customers. On the foreign market, we asked in the local customers’ language. Both clients have higher visibility on their domestic market. A weak spot from the domestic measurement is confirmed abroad. Often it gets even deeper.
In our two measurements, content about the market and services mattered more than the language of the website itself. Example: AI clearly reads from the website that the company delivers to the Netherlands. It can then list it among suppliers for a Dutch customer too.

A result from the domestic market says nothing about foreign markets; each market needs to be measured separately. Source: iMPROVE AI visibility audits, 2026.
The chart shows the share of answers to questions without the company name in which AI mentioned the company. In questions with the company name, AI mentioned it in both languages: the machinery supplier in 5 of 5 answers in German, the agency in 8 of 9 in English.
Weak visibility on the domestic market repeats abroad. This showed with the Swiss IT agency. Topics where AI doesn’t mention it at home are at zero in English too. Topics where it gained citations and mentions at home held up in English only in ChatGPT. AI Overviews and Perplexity mentioned it in English only in the question with its name.
AI knows the company by name in a foreign language too. But when a customer looks for a supplier in a foreign language without the name, AI doesn’t list it among suppliers. For the machinery supplier, the bigger problem turned out to be the website content: it doesn’t say precisely what the company supplies, to whom and where. So translating the website alone is not enough. AI must clearly read from the content that the company operates on the given market, in this case in Germany.
Our advice: first fill the gaps on your domestic market. Then state clearly on your website where you operate: what you supply, to whom and to which countries. Repeat the measurement for each market separately.
Before and after: a large e-shop with physical stores
When the e-shop approached us, it was almost the market leader in its segment. It was also nearly first in Google search results. Two things stand out in this example:
- AI visibility matters for market leaders too. The e-shop had top positions on Google. Yet in the first measurement AI mentioned it in only 23% of answers. A strong position on Google doesn’t carry over to AI automatically.
- When a company starts high, progress looks different. The basics are already good, so the overall number grows more slowly (23% → 27%). A big shift comes where there were specific gaps, for example in local questions (57% → 85%). Local questions are questions about a store or a company in a specific city. A company starting from zero has more room to grow. For a leader, the work is mainly about filling gaps and holding its position against competitors.
Two things changed on the website. Store data was unified and schema was added. Schema is a technical description of the website. Search engines and AI read addresses, offers and opening hours from it precisely. Visibility in local questions then rose from 57% to 85%. We rolled out the changes gradually, so we can’t precisely separate the effect of each step.
The first advice article doesn’t have a link from the menu or other pages yet. We recorded it as a source in AI answers about 14 days after publication. In the latest measurement it appeared 4 times. In general advice questions (TOFU), AI still hardly mentions the e-shop (0% → 1.2% of answers).
What was deployed:
- unified store addresses, phone numbers and opening hours on the website and in Google profiles,
- structured data (schema) across the whole website including categories and for all stores (15–16 September 2026),
- in October 2026, schema for the company itself too.
We designed the schema as ready-to-use code. The client’s developer deployed it on the website. After deployment, we check it according to our methodology.

Large e-shop with physical stores, 70 questions on 5 platforms. Source: iMPROVE AI visibility audit and regular measurement, July – November 2026.
We tracked one product category separately (15 questions): 19% → 24% → 25%. Schema covered this category too. But growth started on 10 September, before schema was deployed. So we can’t attribute it precisely to our work. In the latest measurement, AI mentioned the two main online rivals in 11% of answers, two others in 6% and 5%. In this category, large retail chains lead with 14% and 13%.
Conclusion: in this case, unified data and schema quickly helped with local questions. For general advice and comparison questions, only content will help.
Six steps to improve your company’s visibility in AI answers
The steps are ordered by impact and effort:
- Unify your company data everywhere. Name, address, phone number and opening hours should be the same on the website, in Google Maps and in directories (Azet, Zoznam, FinStat). This step fixes errors AI picks up from directories and profiles. For the large e-shop with physical stores, local questions rose by 28 points (from 57% to 85%) after unifying data and adding schema.
- FAQ in your customers’ words. An FAQ is a section of frequently asked questions and answers on the website. Write short answers to the questions people ask before buying: price, lead time, process, comparison. You can have the same information elsewhere in the text too. A clear wording in the FAQ helps AI find it.
- Specific content for each service. Each service should have its own description with numbers. Write what exactly you do, for whom and how. Add your advantages and proof.
- Structured data. This means schema for the company, services, branches and FAQs. Schema unifies and clarifies the company data. It helps AI link mentions to one company and tell it apart from companies with similar names. For multinational groups, it clearly connects the parent company, subsidiaries and divisions. Schema helps search engines and AI understand company data correctly. On its own, though, it doesn’t guarantee a mention.
- Reviews and presence beyond your website. Get more recent reviews. Complete your profiles in directories and get articles in the media.
- Measure repeatedly. Ask the same questions before the changes and 4–6 weeks after them. Only then can you tell the effect of the work from market movements.
Bonus: cleaning up your online reputation. An AI visibility audit shows more than mentions. It also shows the sources AI builds its answers from. And it shows whether AI talks about the company positively, neutrally or negatively (sentiment). From the list of sources and inaccurate claims, you can see what needs to be unified or corrected online, and where.
We do all these steps as part of our AI visibility and GEO service: from the audit through changes to the website and data, which our own developers can deploy, to repeated measurement.
How AI visibility relates to SEO
SEO is the work of getting a website to rank high in search results, for example on Google. AI visibility is a new topic. It doesn’t replace SEO; it extends it. The two help each other.
The foundation is the same. In our tests, company websites made up 84% of all citation occurrences. Google AI Mode and AI Overviews draw on the same index of pages as regular Google search. A website that search engines can’t read well and that lacks content is weak in both SEO and AI.
A large part of the work is classic SEO. Consistent company data is part of local SEO. Structured data (schema) belongs to technical SEO. Separate service pages and FAQs belong to content SEO. The same changes helped our clients in both search and AI.
It is measured differently. In SEO, we track position in results and the number of clicks. In AI, we track whether and how often it mentions the company and which website it cites. The two views complement each other, so we track them together.
There is still a lot we don’t know
AI visibility is a new discipline. For us, it is also a new service. We are actively working on several projects now. Repeated measurements are gradually giving us numbers. They confirm that content comes first here too.
These questions are still open and we will add data on them:
How long does it take for new content to show? For the large e-shop with physical stores, we know what the data and schema did. For content, the first signal came quickly. We recorded an advice article without internal links as a source in AI about 14 days after publication. Internal links are links from other pages of the same website. Two things remain open. First: when AI will start naming the e-shop in its answers, not just drawing on its article. Second: how visibility in advice and comparison questions will change after more content is published. It also depends on the competition and the segment.
How big is normal market movement? AI answers change even without any intervention. Repeated measurements will help us estimate how large these natural fluctuations are. We can then interpret changes better.
Do mentions in AI also bring visits and leads? So far we measure whether AI mentions the company. We will see the link to analytics, i.e. visits from ChatGPT, Perplexity and Google AI, only over time. The same applies to the link to actual leads.
How will results change with new model versions? Our data is from August to November 2026. Platforms change quickly. We will regularly check whether what holds today still holds.
How fast can a foreign market catch up? We know companies are weaker in a foreign language. We want to know how quickly content that clearly says what the company supplies, to whom and to which countries starts to work.
We don’t promise positions in AI
There are a lot of promises being sold around AI visibility today. So we say up front what can’t be done:
- Guarantee a mention or a position. An AI answer is generated anew for each question and changes with new model versions. No one can guarantee that a company will be in it.
- A bought mention. You can’t buy a mention in a regular AI answer. Paid advertising in ChatGPT exists, but it is labelled and works separately.
- Quick tricks. Hidden text for AI, mass-generated articles or fake reviews can rather damage a website’s credibility in Google and in AI.
What we can deliver: measure where the company stands in AI, show specific gaps, recommend what is worth doing first, and measure the impact after the changes.
Find out how AI sees you
The free AI visibility test uses the same measurement principle as this research. It shows whether ChatGPT, Perplexity and Google AI mention you when a customer asks about your service without your name, and which sources they draw on about you.
The test shows where you stand. The audit tells you what to change, where, in what order and why. We can help as part of our AI visibility and GEO service.


