Personalisation Without Being Creepy: What to Show Different Visitors
Personalisation is treated as mandatory. The reasoning goes: show everyone something relevant and sales increase. In practice the line between “helpful” and “unsettling” runs far closer than it appears, and is easy to cross.
The bigger misunderstanding lies elsewhere: most of what gets sold as personalisation loses to well-chosen defaults.
Levels of personalisation
| Level | Based on | Consent needed | Risk of irritation |
|---|---|---|---|
| Page context | What the person is viewing now | No | None |
| Referral source | Where they arrived from | No | Low |
| City or language | Technical request data | Usually no | Low |
| Behaviour within the session | What they viewed a minute ago | Depends | Medium |
| History over months | A user profile | Yes | High |
| Data from other sites | External sources | Yes | Very high |
The pattern is obvious: the deeper the personalisation, the more it demands and the less it adds. The first three rows deliver most of the benefit and require nothing.
Where the unsettling line runs
The practical test is simple: personalisation stops being helpful when a person cannot explain to themselves how you know that.
“You viewed this item” is clear, they viewed it a minute ago. “Products for your dog’s breed” on a site where they filled in no questionnaire is unclear and alarming, even if the guess is right.
A second test: does the personalisation reveal something the person would rather not display? The classic example is recommendations visible to others on a shared device. Medical, financial and personal topics require separate caution here.
What works without any data about the person
The most useful part of personalisation needs no profile at all:
- Page relevance. Someone on a service page sees examples and pricing for that service, not a generic block.
- Session memory. Selected filters, the basket and partially completed form fields survive a return visit.
- Arrival context. They came for a specific query, so the first screen answers exactly that.
- Language and currency. Offer a switch via a banner rather than switching for them.
- Sensible defaults. The most popular option preselected, the most frequent questions at the top.
That last point usually delivers more than any recommendation engine. A well-chosen default suits the majority and requires knowing nothing about the individual.
What requires consent and caution
Anything relying on history beyond the current visit is processing of personal data with all that follows: a lawful basis, consent, retention periods, the ability to delete.
Practical rules that reduce the risk:
- Explain the source. A caption reading “based on what you viewed” removes most of the discomfort.
- Allow it to be switched off. A visible setting, not one buried deep in an account area.
- Do not use sensitive categories. Health, finances, religion, orientation are not material for recommendations.
- Do not personalise price. Different prices for different people based on behaviour gets discovered quickly and costs reputation.
The technical cost
This is rarely discussed: personalisation conflicts with caching. A page unique to each visitor cannot be served from cache in full, which means it loads more slowly.
For a site whose server response is already approaching a second, a personalised block on the first screen means degrading things for everyone to benefit some. The sensible arrangement: cache the main page and load the personalised block separately, without it blocking the rest.
How to tell whether you need it
Three questions:
- How many items or services do you have? Under twenty and a clear catalogue works better than recommendations.
- Do people come back? If they buy once a year, no history accumulates.
- Are the basics solved? Personalisation on a site with a ten-field form and a slow server is optimising the wrong thing.
The most common situation looks like this: a company plans a recommendation engine when the gain would come from shortening the form and fixing catalogue search. That sounds less exciting, but it can be measured.
Checking your own site
- Open the site in a private window. How much worse is it than the personalised version? If the difference is small, personalisation is not working.
- Find every personalised block and explain aloud where the data came from. If the explanation sounds awkward, it will read that way too.
- Measure speed with and without personalisation.
Good personalisation is invisible: the person simply sees something relevant and does not wonder why. Bad personalisation is noticed immediately, and the thought is not “how convenient” but “how do they know that”. The difference is not technological, it is whether you can explain the source of the data in one sentence.
