AI Personalisation UX: How to Adapt Your Product Without Unsettling Users

AI personalisation in product design sits on a knife edge. Done well, it feels like the product understands you. Done poorly, it feels like the product is watching you. The difference is not in how sophisticated the AI is. It is in how transparently and appropriately the adaptation is expressed.

This guide covers what AI personalisation actually does to user experience, which approaches are working in 2026, and the design principles that separate personalisation that builds trust from personalisation that erodes it.

What AI personalisation is really doing to the user experience

AI personalisation changes the relationship between a user and a product. Instead of a static interface that every user experiences the same way, the user sees an interface that has been shaped for them. This sounds unambiguously positive. In practice, it introduces a new kind of uncertainty into the user experience.

Users of personalised products cannot rely on stable mental models the way they can with static interfaces. The feature I used last week might not be in the same place today. The recommendations I see are not the same as what my colleague sees. The product is making decisions about what to show me, and I cannot always see or control those decisions.

This uncertainty is manageable when the personalisation is legible and the user trusts the product. It becomes a real friction when the personalisation produces surprising changes without explanation, or when the product seems to know things about the user that feel private rather than behavioural.

The spectrum from helpful to unsettling

The spectrum from helpful to unsettling personalisation roughly corresponds to the spectrum from behavioural inference to intent inference.

Behavioural inference personalisation adapts based on what the user has demonstrably done. The product surfaces recently used features first. It pre-fills forms with values the user has used before. It hides features the user has never accessed in three months of use. These adaptations feel helpful because the user can explain them to themselves. They know they use those features frequently. The personalisation confirms their own experience.

Intent inference personalisation adapts based on what the product thinks the user is trying to do before they have explicitly done it. The product notices the user spent ten minutes on the pricing page and changes the onboarding flow. It infers that the user is frustrated based on mouse movement patterns and surfaces support options. It predicts the user is about to churn and changes the interface to show retention messaging.

Intent inference is more powerful and more unsettling. Users who notice intent inference personalisation often describe it as the product reading their mind, which is compelling when the inference is right and disturbing when it is wrong or when the user did not realise the product was making inferences at that level.

Making personalisation legible without over-explaining it

The design challenge is making personalisation legible enough that users can understand why their experience looks the way it does, without making the explanation so prominent that it draws more attention to the personalisation machinery than to the product value.

The right level of disclosure is contextual. For minor adaptations like feature surfacing order, a simple setting that says personalise my experience based on usage patterns, visible in settings but not pushed in front of users, is sufficient. For significant adaptations like major interface changes, a brief in-context explanation when the adaptation first occurs gives users the context to understand what happened without making the product feel like it is constantly explaining itself.

The goal is users who can explain the personalisation to themselves when they notice it, and who can adjust or opt out when they want to. Not users who are constantly aware that they are being personalised, and not users who are surprised by adaptations with no path to understanding why they happened.

Personalisation that creates problems in multi-user contexts

Shared accounts, team accounts, and any context where multiple people use the same product instance are a common failure point for AI personalisation. Personalisation systems that adapt based on usage patterns will create different experiences for different users of the same account. This produces confusion, support tickets, and sometimes incorrect assumptions about the product.

The design options are clear and should be chosen deliberately rather than left to chance. Account-level personalisation adapts the experience based on the account's overall usage rather than individual users. This is less targeted but avoids per-user divergence in shared account contexts. User-level personalisation within a shared account requires clear user identity, which most B2B products have through individual logins. Conservative personalisation reduces the range of adaptations in any context where shared access is common, trading targeting precision for consistency across users.

Measuring personalisation correctly

AI personalisation is frequently measured by engagement with personalised elements: clicks on personalised recommendations, completion rates for personalised onboarding flows. These metrics are real but they are not sufficient to evaluate whether personalisation is improving user outcomes.

The right measurement is whether users who experience personalisation achieve the outcomes that matter, activation, retention, feature adoption, task completion time, at better rates than users who do not. High engagement with personalised suggestions alongside flat retention or activation is a signal that the personalisation is effective at capturing attention without improving user value. According to research on personalisation effectiveness from Nielsen Norman Group, the products with the best personalisation outcomes are those that measure the downstream business and user outcomes, not just the engagement with the personalised elements themselves.

How Studio Maydit approaches AI personalisation design

We design personalisation systems around the user's experience of the adaptation, not around the sophistication of the AI doing the adapting. The most important design decisions are how legible the personalisation is, how appropriate the inference level is for the trust relationship the product has with the user, and how easy it is for users to understand and adjust their personalised experience. If you are building or refining AI personalisation in your product, book a free 30-minute call with Studio Maydit.