In the traditional UI the user learns the interface: navigation, taxonomy, position of functions. The learning curve has always been considered inevitable, and many UX heuristics are founded on this assumption. The pattern reverses: the expectation grows that it is the interface that learns the user — who they are, what they want, in which context they are — without a learning phase and in their language.
The expectation, however, conflicts with how chat actually works. Using it effectively requires skills that are not acquired spontaneously: writing good prompts, verifying outputs, managing context. It is the articulation barrier described by Jakob Nielsen: the difficulty of translating a need into an effective request. It is not a niche barrier: according to the OECD PIAAC 2023 survey (published December 2024) about 26% of adults in OECD countries are at the lowest literacy level (Level 1 or below), and 18% are below the basic level in all measured domains — with strong differences between countries. The feeling of not having to learn anything does not correspond to a simpler interaction: the 19.1% figure on users who use prompting strategies confirms it (see Search becomes conversation).
A second element: the mental model is also formed by how the system is presented. A MIT Media Lab study (Pataranutaporn et al., Nature Machine Intelligence, 2023) shows that the exact same system is perceived as more reliable, empathetic, and performant if presented as “caring” rather than neutral, and that subsequent interaction reinforces the initial mental model. Onboarding, naming, and framing are not accessory communication, but a lever that determines the experience.
The practical consequence concerns the products we already manage, not the chatbots to build: our application, our portal, our management system are now judged by people who are no longer willing to invest time to learn them. The learning budget the user grants has shrunk, and this puts pressure on everything in the product that presupposes training: tutorials, first-access wizards, manuals, taxonomies to memorize, functions hidden in deep menus.
Effect on the professions
- Information Architect — This is where an architecture of intent must be added alongside the information architecture. The distinction, in short: the information architecture organizes content according to the logic of the domain — categories, hierarchies, taxonomies — and answers the question “where is this thing?”; the architecture of intent starts from the goals the user wants to achieve and maps them into direct, actionable access points, answering “how do I get X?”. In practice it is built in four steps. (1) Collect the real intents from sources that already exist in the company: internal search engine queries, support tickets, call center transcripts, assistant logs if any. Not from business categories. (2) Normalize them into a finite list of goals expressed in the user’s words — “cancel the subscription”, “change the IBAN”, “understand why I was charged this amount” — not in internal ones (“Registry management”, “Dispositive area”). (3) Give each intent a direct access point: a highlighted CTA, a home shortcut, a deep link, an answer at the top of the page — reachable without crossing the hierarchy. (4) Make the two levels coexist: the structure remains for those who explore and for those who do not yet know what they are looking for; intent access serves those who already know what they want and are the majority of returns. Example: a banking site that archives direct debits under “Products › Accounts › Ancillary services › Direct debits” is not wrong — it is just unusable for someone arriving thinking “I need to set up the direct debit for my electricity bill”. The architecture of intent adds that item as a first-level entry point, without dismantling the taxonomy. Governance note: intents change faster than taxonomies and must be periodically reviewed on internal search and support data, otherwise direct access ages worse than the structure. → Useful resource: Rethinking Jobs-to-be-Done for the AI Era — Momentum Design Lab
- Product Manager — The assumption that the user will train themselves falls. Long onboarding, guided tours, and first-access wizards must be considered a cost the user is no longer willing to pay, not an investment: the function must be understandable on the first attempt, or it will not be used. To review in the roadmap: functions with a high abandonment rate in onboarding and those that require training to be adopted — in many cases the problem is not the missing explanation but the poorly exposed function.
- UX Writer / Copywriter — Labels must be self-explanatory without context: those arriving have read nothing and will read nothing. Internal vocabulary (“reserved area”, “position”, “practice”) must be replaced with the verb of the action. Microcopy becomes the substitute for documentation: explain where it is needed, not in a help page.
- Content Designer — Online guidance moves from manual to contextual support: short content, situated in the flow, exposed at the moment of doubt instead of collected in a “Support” section. Content must be composable based on the user’s context, not a fixed block identical for everyone. → Useful resource: Designing for Generative AI Experiences — NN/g
- Interaction Designer — Reduce the entry cost to functions: less hierarchical depth, frequent actions on the surface, sensible default states instead of mandatory configurations. Every step that asks you to decide before understanding is an abandonment point.
- UX Researcher — The reference metrics change. Learnability measured on repeated sessions loses weight compared to first-attempt success and time to first useful result. Tests must be conducted without preliminary instructions and with users who have never seen the product — if you need to explain the task, the data is already compromised. First impression must also be measured, not just steady-state use: how the product is presented conditions the mental model and, consequently, satisfaction and trust. For conversational scenarios there are first validated tools, including the Chatbot Usability Scale in the Italian adaptation Ciao AI (Springer), but classic methods — task-based testing, card sorting, tree testing — remain the right tool for traditional products. → Useful resources: Mental Model Shifts in Human-LLM Interactions; Influencing Human-AI Interaction by Priming Beliefs about AI (Nature Machine Intelligence)