Personalize onboarding without a long questionnaire
Use saved progress and explicit preferences before inferring interests. Give people clear controls and compare recommendations with a useful general experience.
In this article
Begin onboarding personalization with information the user has made clear: saved work, a chosen display setting, or an explicitly followed subject. Treat early browsing as an uncertain signal. Keep recommendations explainable and easy to correct, preserve a general browsing route, and compare personalization against an experience that already handles basic continuity well.
On Monday, someone opens an essay about insomnia to help a friend. By Tuesday, the reading app has decided that sleep is their defining interest. Its home screen offers more insomnia articles, its notifications mention bedtime, and the book they were actually reading has slipped out of view.
This hypothetical app has learned too much from too little. Early behavior can help a product become more useful, but a first session provides a thin and ambiguous record. The design needs to distinguish what happened from what the system thinks it means.
Remembering is a useful place to start
Consider the first few sessions of a reading app. The reader opens a long article, increases the type size, saves another piece, and closes the app halfway through a chapter. On the next visit, several improvements are possible without assigning that person an identity or an interest category.
Their reading position can be preserved. Their chosen display settings can remain in place. The saved article can appear in their library. A “Continue reading” area can offer the unfinished chapter while leaving the rest of the collection accessible.
These changes use direct evidence of an action or preference. They still need careful behavior: opening an article briefly should not necessarily displace a book from the main continuation slot, and a finished item should not linger there indefinitely. But the question is concrete. Which piece of work or explicit setting should the product retain?
Nielsen Norman Group distinguishes user-controlled customization from system-controlled personalization. Keeping a reader’s chosen display settings respects customization; choosing which unfinished item to emphasize introduces system judgment. That distinction helps teams decide where an explanation or correction belongs.
Make both visible enough to understand. “Continue chapter 4” states what the product knows. “Your perfect evening read” claims an understanding of the person and the moment that a saved position cannot support.
Give different signals different authority
The temptation is to pour every event into one interest profile. Opening, saving, finishing, and explicitly following a subject then become interchangeable signs of enthusiasm. They are not interchangeable from the reader’s point of view.
A saved item may be a future chore. A completed article may have been disappointing. A short visit may mean the answer appeared in the first paragraph. Duration alone cannot resolve those meanings.
Use a narrow interpretation unless you have a reason to go further:
| Observed signal | Defensible immediate use | What remains uncertain |
|---|---|---|
| Reader changes text size | Retain that explicit display preference | Whether it should apply on every device |
| Reader leaves a chapter unfinished | Offer the saved position | Whether they want to continue it today |
| Reader follows a named subject | Include that subject in recommendations | How much of the collection should concern it |
| Reader opens one article | Keep it in recent history, if that feature is enabled | Whether the subject is a lasting interest |
| Reader hides a recommendation | Remove that item and acknowledge the action | Whether they dislike the item, author, or whole subject |
NN/G’s personalization guidance cautions against narrow profiles built from isolated behavior and recommends preserving user control. Apply that caution during the first few sessions, when each observation can otherwise have an outsized effect.
Do not infer a health condition, religion, sexuality, or other sensitive personal trait from reading history in order to “improve” onboarding. Someone may be researching for work, supporting another person, or reading critically. Even accurate-seeming guesses can make an ordinary discovery screen feel invasive. A useful content recommendation rarely needs a claim about who the reader is.
Put the controls beside the recommendations
A general preferences page is useful for deliberate adjustment. It is a poor sole remedy for a bad recommendation sitting on the home screen.
Give the reader a local action with a specific meaning. “Hide this article” should hide the article. “Less on this subject” should change the subject’s influence. Avoid one unexplained thumbs-down control if its effect could range from removing an item to reshaping the entire feed.
For the insomnia example, the reader might choose “Don’t use this article for recommendations.” That lets them retain the piece in their library without making it a lasting signal of taste. The distinction matters: saving material and endorsing it are different actions.
Show a brief acknowledgment and an undo route. If the action affects future suggestions rather than the current list, explain that instead of pretending an immediate refresh proves the preference was understood.
Provide a way to browse without recommendations as well. Search, the full library, and editorial collections should remain stable destinations. Someone should not have to retrain a model to find a different kind of book. Keep personalized ordering within a recognizable part of the interface so that the navigation itself does not move after every interaction.
Leave room for a new interest
A reader who followed history in the first session may want fiction in the fifth. Explicit preferences should be editable, and a personalized shelf should not consume every discovery surface.
Mixing in broader material is an editorial and product decision, not merely a technical adjustment to recommendation weights. Decide which spaces promise relevance to prior choices and which invite discovery. Label them accordingly. “Because you follow history” and “New in the library” set different expectations and can sit comfortably on the same screen.
When there is too little history to support a useful recommendation, show a good general collection. Avoid manufacturing a personal explanation for an ordinary popularity list. A small app with a well-chosen catalog may serve a new reader better through clear categories than through an elaborate inference system.
Compare against a baseline worth using
To evaluate the adaptation, give both groups a competent experience. Both should retain reading progress and explicit display preferences. One can receive a clear general collection; the other can receive the personalized shelf. Otherwise, the experiment may credit recommendation logic for basic continuity that everyone ought to have.
Keep catalog availability and major layout differences consistent where possible. Compare whether people find reading they consider worthwhile, using behavior alongside direct feedback. Opening more items can indicate discovery, but it can also mean that none of them delivers what the headline promised. Finishing every item is not a sensible universal target either.
Inspect hiding, corrections, and switching to general browsing. Those actions can be healthy uses of control; repeated corrections to the same subject suggest that the system is not honoring the reader’s instruction. Review the recommendation explanations too. A technically valid association may still be an unhelpful thing to say aloud.
A useful closing question for a research session is, “What do you think this app has learned about you?” Compare the answer with the behavior you actually implemented. If the reader believes one exploratory click has permanently defined their interests, the problem is already present, whatever the recommendation score says.
A tailored demonstration still needs a route to the customer’s own task. Plan the handoff from sample data to real work alongside the personalized starting point.
Sources
- Amy Schade, Nielsen Norman Group: Customization vs. Personalization in the User Experience.
- Amy Schade, Nielsen Norman Group: 6 Tips for Successful Personalization.
