The customer who is used to asking AI.

A teenager does not understand an explanation, so asks AI to explain it differently. Wants to learn a skill, so finds a video. Notices an interesting topic in recommendations and starts exploring. Doubts an answer and checks another source.

What interests me in these actions is the habit of getting help at the moment and in the form it is needed. If that experience becomes routine, it may change expectations of other digital services: an online store, a banking app, a learning platform or a workplace tool.

That is the perspective from which I read The Future Report 2026, a study of European teenagers delivered by Livity and commissioned by Google. The report does not predict future purchasing behaviour. But it offers useful starting points for product and marketing hypotheses that businesses can test today.

What the researchers studied

The authors report an online survey of more than 16,000 teenagers aged 13–18 in 11 European countries, alongside qualitative research involving over 100 teenagers.

One important detail: the report combines data collected at different times. Seven countries were surveyed in spring 2025 and four more in spring and early summer 2026. The UK sample of 6,000 was weighted down to the equivalent of 1,000. This is therefore neither a simultaneous snapshot of all European youth nor a study of Russian teenagers. [9–10]

Participants were also reporting on themselves. Their answers describe perceptions of technology and stated habits; they do not demonstrate improved knowledge, greater creativity or an ability to identify misinformation.

For me, this is material from which to draw questions about our own products.

1. Users get used to explanations they can adapt

Page 13 reports that, among those who used AI in the past year, 64% use it for educational or creative purposes daily or almost daily. That denominator matters: the figure concerns AI users. [13]

The answers about learning benefits are particularly interesting. Explaining difficult topics in different ways was selected by 44% of respondents, instant answers and feedback by 41%, and learning at one's own pace by 36%. [16]

Interviews provide concrete examples: asking for a simpler explanation, creating revision flashcards, generating exercises and checking understanding. [17–19]

My interpretation is that some users are developing an expectation that they can keep refining an explanation until it becomes useful.

For businesses, this is a good reason to review instructions, product descriptions and help sections. A laptop shopper can be given a specification table. They can also be helped to understand which specifications matter for their tasks and where paying more would offer little practical benefit. An analytics user can be shown a chart and given a way to ask what changed and which data to investigate.

In both cases, the quality of the explanation creates the value. A chat window by itself guarantees nothing. I would evaluate such a feature through task completion, repeat support requests and time to a comprehensible result.

2. Video helps people move from interest to action

In the qualitative research, teenagers describe video as a way to learn a concrete action: understand a school topic, improve filming skills, learn a craft or acquire another skill. YouTube stands out as an important venue for independent learning. [44–47]

There is a useful idea here for content marketing: an instructional video should help someone do something.

A product video can show how to choose the right option, complete initial setup, avoid a common mistake and recognise a successful result.

The same principle applies to complex services. Instead of a general presentation of analytics capabilities, explore one situation: why revenue is rising while profit falls; why advertising-platform reports disagree with the CRM; or how to investigate an unusually high conversion rate.

The report did not measure the sales effectiveness of these videos. This is a hypothesis to test. But it gives the creator a substantive brief: what action will the viewer be able to perform afterwards?

Content can then be evaluated beyond views, through visits to instructions, use of the demonstrated feature and fewer recurring questions.

3. Discovery starts before a search query exists

The question about how teenagers most often discover new topics and interests produced the following distribution. [38]

This is not a distribution of purchases by channel or a guide to allocating an advertising budget. The question was about interests, not product selection.

The distinction still matters for marketing. Someone may first encounter a use for a product, become interested in the task, and only then start looking for a solution.

It therefore helps to prepare content for different situations: show a clear example when someone is unaware of a possibility; explain how it works once they are interested; and provide comparisons, limitations and evidence when they are choosing.

A company present only at the last stage is relying on someone else to shape interest and selection criteria.

4. Using AI and trusting it are different things

Among participants who use the internet, 62% say they always or often consider the trustworthiness of information when using the internet, AI or other technologies for learning. [26]

“Consider” is the key word. The study did not test how successfully they distinguish correct answers from incorrect ones.

The interviews illustrate that distinction. Teenagers describe checking sources, but also occasions when AI's confident wording led them to believe an incorrect answer. [29–31]

For a business, the practical question is: how easy is it to verify what your service says?

Useful features in an AI tool include links to source documents, the period covered by the data, explanations of calculations and a route to a human. In a case study, useful information includes the starting situation, comparison conditions and limits of the result.

This is particularly important when AI explains numbers. A claim that sales fell because of advertising should allow users to check which data were used and why that explanation was proposed.

I would assess trust through concrete actions: could the user locate the source, notice a limitation and make an informed decision?

5. A long session does not always indicate a good product

Two figures appear in the report. Of participants who use the internet, 86% feel they manage their online habits, including screen time and safety, quite or very well. Meanwhile, 42% feel they spend too much time online. [51, 55]

These are answers to different questions. Without cross-tabulated data, we cannot say how many teenagers selected both responses. But the combination shows why self-reported control should not be treated as a sufficient measure of user wellbeing.

For a product team, this is a reason to examine engagement more carefully. A user spent twenty minutes in the app. Were they absorbed in a useful task? Unable to find a setting? Repeatedly checking an unclear answer? Still scrolling after they wanted to stop?

Session duration alone does not tell us.

For services that help people complete tasks, I would supplement activity metrics with successful completion, time to outcome and perceived clarity. For products used by teenagers, I would separately examine notification controls, ways to stop and default settings.

6. AI users do not necessarily want to hand over all the work

Among those who have used AI, 73% believe it improved at least one aspect of their creativity. Specifically, 53% report help with new ideas or solutions and 47% with creative writing. [20]

These are subjective assessments, not the results of independent creativity tests.

In interviews, teenagers also describe wanting to retain their own involvement: get a hint, understand code, refine an idea or write something themselves. Some explicitly worry about relying too heavily on ready-made answers. [21, 23, 31]

This is a useful distinction for teams developing AI features. Different users and tasks may benefit from different levels of help: a hint, an explanation, a check or a completed draft.

In education, for example, it matters whether someone can repeat an action independently. In a workplace tool, it matters whether they understand the result and can correct it. “Do everything automatically” will therefore not always be the best scenario.

That should be investigated through user research, rather than decided solely by what the model can do.

What I would put into practice

After reading the report, I would ask a team to walk through the user journey and answer five questions. Can someone get an explanation suited to their task? Is there a clear example that helps them start? Can they see where recommendations and numbers come from? Can they choose the level of AI assistance? Can we distinguish useful engagement from difficulty?

Testing this does not require immediately rebuilding the product. Pick one frequent scenario: comparing plans, choosing a product, creating a first report or setting up an integration. Make it clearer and observe what changes in successful task completion.

Reading the report without overgeneralising

The study was commissioned by a company that develops AI products and owns YouTube. That context matters when reading its emphasis and recommendations. It does not, by itself, invalidate the survey findings.

The report does not establish that AI improves learning, that teenagers reliably recognise false information, or that these digital habits will necessarily persist into adulthood.

For Russian businesses, testing the conclusions with their own audiences is particularly important: platforms, educational environments and conditions of technology use differ.

My working hypothesis after reading is this: experience with AI may raise expectations of clarity and responsiveness in any digital service. Users are becoming more accustomed to refining questions, asking for another example and receiving help without a lengthy search through menus.

The best place to test that hypothesis is a specific customer task: how quickly did they understand what to do, could they verify the answer, and did they get the outcome they needed?

← All posts · Talks and publications