A shopping site says, 'Because you bought running shoes, you may like these socks.' Reasonable. Now imagine it says, 'Because you browse after midnight, recently searched knee pain and usually buy discounted items, we selected this for you.' The recommendation may be better. The feeling may be worse.

Personalisation creates a psychological trade-off. Relevance is useful. It reduces search and can make technology feel attentive. But relevance also reveals how much the system knows, or appears to know, about us.

Recent research on AI-mediated personalisation suggests that privacy risk becomes more salient when personalisation feels too intimate or when people do not feel in control. The issue is not only data quantity; it is the inference a person makes about surveillance.

Context matters. We tolerate a map app knowing location because location is necessary. We may be less comfortable when a shopping chatbot infers mood from language and adjusts price or persuasion accordingly.

The line between helpful and creepy often depends on three questions: Did I knowingly provide this information? Is it clearly relevant to what I asked? Can I control or remove it? When the answers are yes, personalisation feels more legitimate.

Businesses sometimes believe more data automatically means better customer experience. Psychologically, 'better' requires trust. A recommendation that increases short-term conversion but creates a sense of being watched may damage the relationship.

Transparency helps, but even transparency is tricky. Telling people that AI was used can sometimes reduce trust in the person or organisation using it, particularly when disclosure makes the technology feel less legitimate for the task.

The best personalisation may therefore be restrained. Know enough to reduce effort, not so much that the customer suddenly wonders whether the system knows where they were last Tuesday.