Uhlík vs Křemík · Czech podcast about artificial intelligence

Only 8% of Czechs pay for AI. Ipsos's Jakub Malý on where AI helps in research and where it does not

Episode 17 · 3 September 2026 · 59:51 · Guest: Jakub Malý

According to Ipsos research, only 8% of Czechs pay for any artificial intelligence tool and 61% do not even plan to. Jakub Malý, CEO of Ipsos for Czechia and Slovakia, discussed this with hosts David Salomon and Adam Hanka on the Uhlík vs Křemík podcast. They also talked about persona bots, synthetic respondents and who decides today what we end up buying.

Free means with ads

Malý recalls the well-worn saying that "when someone tells you it is free, you are part of their product." In his view, paid news works much the same way, with the share of paying users at around 10%. That is the tenth of people who choose their content themselves and believe this gives them more control. Everyone else takes what is served to them for free.

That is why Malý expects ads in cheaper and free AI plans. Adam Hanka uses only paid models and says people without a subscription have no idea what they are missing. For 20 dollars a month, anyone gets the same artificial intelligence as the head of Google. It took him three quarters of a year to build an intuition for working with the models.

Malý adds another problem. People often use AI for isolated tasks and only replace individual steps of a process they used to do by hand. It would be better to rethink the activity from scratch, because some of the steps may no longer make sense in today's context.

AI breaks down the brief, the client conversation stays with people

At Ipsos, AI already helps at the start of a project. Pre-prompted models break down the client brief, so even a less experienced team member knows what to ask the client. According to Malý, clients often come straight with a technicality, such as a request for a thousand respondents or four focus groups. But "nobody actually needs to do a thousand interviews, nobody needs to do four focus groups. Those people need to solve some problem, some question."

AI prepares a first draft of the question areas. Ipsos does not hand two things over to it, though: talking the client out of a brief that makes no sense, and discussing with the client at the end of the project what the results mean for their organisation. Internally, the company also uses cheaper models to help formulate a good prompt for a more expensive model.

Persona bots and synthetic respondents

Persona bots, as Ipsos understands them, always work with data from real consumers. Clients can converse with them about research results instead of reading a presentation, and the bot says openly when a question goes beyond the data that was studied. Synthetic respondents, by contrast, are artificial from the start and are built from aggregated sources, such as Eurostat or data from a telecoms operator.

Experimenting with them is fine, Malý says. The problem comes when the results are published as a prediction with some kind of guarantee. Research and development that Ipsos carried out with Stanford shows that the context of the specific brief matters. Synthetic data is better suited to screening ideas, while a product close to market launch has to be validated on real people. Hanka adds that a fully synthetic panel makes no sense and a control sample of people is always needed.

The risk of abuse is illustrated by an August case involving the American firm Median Strategies. It published polls on mayoral elections and Democratic primaries, the media took them over uncritically, and in the end it came out that the aim was to profit from betting odds.

Who decides what you buy

According to the Ipsos survey, 39% of Czechs use AI regularly and another 27% have at least tried it. Malý watches search without clicking with concern. People used to fight for the top three positions on Google, and today often only one option wins. 11% of Czechs use AI as their main source when choosing products, and 23% among young people. According to Malý, the algorithm prioritises authoritative sources, in Czechia for example ČTK. When they are missing, it reaches into its training data, and Reddit can become the main source about a brand.

Trust is lagging so far. Only 32% of Czechs believe that companies using AI will protect their personal data. According to Malý, people want regulation, but not too much of it. They often start with AI by planning trips, where it is hard to go wrong, and the share of users who use AI this way is rising from roughly a quarter last year to a third.

The full interview with Jakub Malý is in episode 17 of the Uhlík vs Křemík podcast.

The episode is in Czech. Czech page with chapters, quotes and the full transcript