Infinite Personalization: Why “One More Thing for You” Can Become the New Infinite Scroll
AI recommendations can remove choice overload, but endless personalization can eliminate stopping cues and turn “one more thing for you” into an adaptive form of infinite scroll.
Infinite personalization removes the natural moment when a user asks whether to continue
The UK government’s 2026 response on children’s online experiences identified infinite scroll, content recommendation algorithms, and autoplay among the persuasive features respondents were most concerned about. [1] The underlying design pattern is the absence of an endpoint. AI makes that pattern stronger because the feed no longer needs a finite catalog or editorial queue; it can continuously rank, remix, or generate another item tailored to the user. When the interface always has a plausible next recommendation, stopping requires an active act of resistance rather than arriving naturally at the end of a page, lesson, chapter, or session.
Stopping cues are product features
A boundary such as “you are caught up,” “session complete,” or “continue tomorrow” gives deliberation a place to occur. Removing every boundary is itself a design choice.
Recommender systems can support autonomy by filtering overload, but can also narrow it
Research on AI-powered recommender systems argues that they can help people navigate overwhelming option spaces while also posing risks to personal autonomy through manipulation, identity effects, and constrained exposure. [2] Infinite personalization should therefore be evaluated by whether it expands useful choice or traps the user inside the system’s prediction of what will keep them engaged. A learning product that selects the next exercise from a curriculum serves a bounded goal. A feed that keeps generating “one more thing for you” after the stated goal is complete may be optimizing continuation rather than value.
The Digital Services Act treats interface architecture as part of free decision-making
The DSA prohibits interfaces that deceive, manipulate, or materially impair free and informed decisions. [3] An endless feed can become problematic when stopping, switching recommendation modes, or reaching chronological content is made unnecessarily difficult.
Make the non-personalized path real
Users should be able to choose chronological, finite, topic-limited, or non-profiled modes without navigating an obstacle course.
User research suggests meaningful demand for contestability
A cross-country study of personalized recommender systems found that a notable minority of users would choose a non-personalized option and linked contestation preferences to attitudes, awareness, and privacy concerns. [4] This supports a broader product principle: personalization should be contestable. Let users reset a recommendation profile, remove an inferred interest, ask why an item appears, or switch to a mode that does not learn from every interaction. Infinite personalization is less coercive when the user can see and alter the machinery generating “next.”
Dark-pattern research warns that attention extraction can be a consumer harm
The OECD includes practices that spur increased attention among the harms associated with dark commercial patterns and notes that such patterns can exploit biases and become more targeted through machine learning. [5] Endless recommendation should therefore have an attention budget just as notifications do.
Measure session quality, not just session length
A shorter session that completes the user’s goal can be a better outcome than a longer session filled with low-value recommendations.
Finite sessions can still feel personalized and generous
A recommendation system can offer a deliberately bounded set: five articles for this goal, three exercises for this skill gap, one optional extension, then a clear end. The AI can explain why the set was chosen and invite the user to request more. This reverses the control direction. Instead of the product continually asking the user to stop, the user asks the product to continue. That small architectural change creates a meaningful consent point without removing personalization.
Generated content makes stopping policy more important because supply is no longer the constraint
With generative AI, the system may never run out of plausible content. It can invent another quiz, story, challenge, shopping suggestion, or conversational thread on demand. That abundance shifts responsibility from content inventory to session governance.
Generation should have a stopping function
Define completion conditions, diminishing-value thresholds, cooldown rules, and explicit “done for now” states before deploying infinite generation loops.
The best personalization helps the user leave with something completed
A trustworthy system should be able to say, “You achieved what you came for; here are your next options when you return.” That is the opposite of an engagement machine that treats departure as failure. Personalized feeds, tutors, coaches, and companions can still be delightful, but they need boundaries that preserve reflection and outside activity. The product should optimize the value of the session, not the difficulty of ending it. In an era where AI can always produce one more relevant thing, deliberate endings become a core element of ethical experience design.
Designers can also make stopping visible in the data model. Track the user’s stated session purpose, completion condition, and whether recommended items still contribute to that purpose. When marginal value falls, shift from feed generation to summary: what was completed, what can wait, and what the user may want next time. Give users a visible queue they can save rather than forcing consumption now. For children or vulnerable users, conservative defaults should reduce autoplay, overnight recommendations, and endlessly generated content. For adults, offer explicit modes such as focused, exploratory, chronological, or finite-session. Personalization then becomes a tool the user directs rather than an invisible engine that decides the session should continue.
Works Cited
- 01
- 02
- 03European Union — Digital Services Act, Regulation (EU) 2022/2065 eur-lex.europa.eu
- 04
- 05OECD — Dark Commercial Patterns oecd.org
CodeHistory is a living archive. Citations document the evidence used for this edition; later evidence may refine the account.