Cognitive Sovereignty in the Age of Personalisation Algorithms

On Regressive Disclosure And The Right To Browse Like A Human

6 February 2026Series: Agency, Personalisation and Sovereignty

We’ve become remarkably comfortable with machines anticipating our needs.

Your streaming service lines up the next thing you’ll probably watch. Your feed serves exactly the post that keeps your super evolved thumb moving. Your shopping app hints at a purchase before you’ve even admitted you want it.

It feels like magic, and according to Arthur C. Clarke’s Third Law it might as well be.

But a quiet trade is happening underneath the convenience. We’re not just outsourcing tasks. We’re outsourcing the next thought.

What to explore. What to consider. What to think about next.

Those choices are increasingly mediated by systems optimised for engagement, not understanding. And once your “next thought” becomes a product surface, you start to lose something that’s hard to measure and very easy to miss.

This is where cognitive sovereignty comes in.

Cognitive sovereignty is the principle that people have a right to think, explore, and decide independently - without algorithmic steering nudging them towards predetermined outcomes.

It isn’t anti-personalisation. It’s pro-agency.

And right now, agency is getting quietly eroded.

This isn’t actually new. It’s been a challenge for about a century. The only thing that’s changed is the fidelity.

In the 1920s, Bernays was writing openly about mass persuasion and how public opinion can be shaped at scale. In the 1950s, Packard popularised the fear that advertising was happy to work below the level of conscious reasoning, nudging desire rather than informing choice. In the 1970s, Herbert Simon gave us the cleanest framing for what was coming next: in an information-rich world, attention becomes the scarce resource.

Then the web turned “mass persuasion” into “personal persuasion”.

Cookies (1994) made tracking practical. AdWords (2000) industrialised intent as a marketplace. News Feed (2006) normalised algorithmic ranking as your default reality. Real-time bidding (introduced in 2009) made buying slices of human attention something you could do per impression, in milliseconds.

So cognitive sovereignty hasn’t suddenly become a problem. It’s the same old “who gets to steer your mind?”, just running on better data, tighter feedback loops, and a business model that treats your next thought like inventory.

The Personalisation Paradox

Modern interface design is full of good intentions.

Reduce friction. Reduce cognitive load. Help people get to what they want.

The paradox is that the same systems that reduce cognitive load often reduce cognitive agency in the process.

When Netflix suggests your next show, it’s solving a real problem. Decision fatigue is real. Choice paralysis is real. Most people don’t want to browse an infinite warehouse after a long day.

But in solving that problem, something else happens.

You stop browsing.
You stop bumping into the unexpected.
You stop choosing in any active sense.
You let the system decide what is worth your attention.

This isn’t inherently malicious. The system is doing what it was designed to do. The trouble is the cumulative effect: your digital world narrows. You see variations of what you already liked. You’re served familiar angles on familiar topics. The path becomes smooth, frictionless, and quietly pre-decided.

Jakob Nielsen wrote about progressive disclosure as a way to manage complexity. What we’re dealing with now is something different.

Regressive disclosure.

Systems that progressively hide the full breadth of possibility, presenting only what they’ve decided is “relevant to you”.

The filter bubble isn’t just about information anymore.

It’s about imagination.

When Convenience Becomes Constraint

Personalisation doesn’t just predict preferences. It encodes assumptions about what a “good” preference even is.

Here’s where this gets concrete: culture.

Spotify’s research into socially-motivated music recommendation is a useful example. In more collectivist contexts, communal listening and social bonding can be central - meaning “good recommendations” are not only about individual novelty and personal taste graphs, but about shared familiarity and group context.

If your engine is built on “you”, it can miss the point when the user is trying to build “us”.

That’s not a niche edge case. It’s a reminder that personalisation is never neutral. It carries a theory of the person.

And even where the training data is representative, there’s a deeper issue.

Personalisation optimises for what you already chose.

It does not optimise for what might change you.

It privileges the familiar over the challenging, the comfortable over the necessary. And because the system gets rewarded when you stay, it gets very good at keeping you in the warm bath of “more like this”.

Which brings us to serendipity.

Serendipity Isn’t A Feature

Most platforms now offer some kind of “Discovery” mode.

It’s well-meaning. It’s also not the same thing as serendipity.

Real serendipity is not “the algorithm widened the radius”. Real serendipity is what happens when the system doesn’t fully predict you. When you take a wrong turn. When you follow a human trail instead of a ranked list. When the interface gives you space to wander without immediately snapping you back onto the engagement rail.

The moment we try to design “the unexpected” as a reliable output, we domesticate it. We turn the accident into an appointment.

So no, we can’t manufacture serendipity.

But we can create conditions where it becomes possible.

Interfaces that sometimes decline to be helpful
Recommendation surfaces that occasionally include “doesn’t match your pattern”
Moments where auto-play waits, rather than pouncing
Browsing paths that feel like a library, not a conveyor belt

Not serendipity as a button.

Serendipity as the result of intentional incompleteness.

A Small, Very Danish Example

I’ve got a very personal, and possibly arguable, view on this.

I live in Denmark. Danish supermarkets often have what I experience as a fairly limited selection. To be honest I experience this as something more acute., but that’s not an accident. It’s shaped by regulation, supply chains, consumer expectations, and the economics of not carrying twenty variants of the same thing.

In other words: it’s curated. It anticipates “the average need”. It trades breadth for predictability.

And yet there’s a loophole.

Spotvarer - spot deals - are these limited-run, slightly chaotic items that sit outside the permanent inventory. One week it’s a strange seasonal snack. Another week it’s an Italian thing I’ve never heard of. Occasionally it’s something genuinely weird and wonderful.

For me, that’s the saving grace of the Danish supermarket system. The shop is optimised, but it still has a little side-door where discovery can happen.

The point isn’t that Danish supermarkets are “good”. It’s that even a curated system can leave a lane for surprise.

That’s roughly what I want from digital systems too.

Curate the basics. Reduce the chaos. Help me get what I came for. But leave a real, unpolished lane for the unexpected - not an “engineered surprise”, but a space where the system isn’t trying so hard to be right.

One UN-Flavoured Vignette

I work in international development, designing systems in contexts where the margin for error is not “annoying”, it’s material.

Imagine a public information portal used during a crisis. It contains guidance on health, shelter, legal rights, and how to access support. Someone adds a personalised “For you” panel because it reduces drop-off.

Here’s what happens in practice.

A user reads two pieces about emergency cash assistance
The system learns: “This person engages with cash assistance content.”
Next visit, the panel prioritises more of the same: cash, cash, cash
Meanwhile, updates about legal requirements, protection services, or changes in eligibility slide down the page because they look “less relevant” to this user’s inferred interests

The portal is still “usable”. It’s even “engaging”.

But it has quietly swapped “help me navigate the situation” for “keep me inside my own behavioural shadow”.

That’s cognitive sovereignty failing in a context where people can least afford it.

If you want the everyday version: think about searching a health symptom at 01:00. You click one anxiety-shaped result. The system “helpfully” serves ten more. You didn’t become informed. You became more certain, and less capable of stepping away.

Designing For Agency, Not Just Efficiency

If you’re building personalised systems, transparency and control can’t be decorative. They’re not a nice-to-have.

They’re dignity.

Here are four practical moves that raise cognitive sovereignty without pretending you can “solve” it.

Make The Curation Visible

Users deserve to know why they’re seeing what they’re seeing - in context, at the moment it matters, in plain language.

Not a 50-page terms doc. A simple, legible “why”.

And here’s the bit we rarely do.

Show what they are not seeing.

Because “personalisation” is not just what you show. It’s what you quietly bury.

Let People Set The Personalisation Intensity

Not everyone wants the same amount of algorithmic help.

So treat personalisation like a setting, not a fate.

Low: chronological, broad, lightly filtered
Medium: ranked, with visible diversity constraints
High: heavily personalised, fast, narrow

If this control is real, it changes the relationship.

Users are opting into curation, not being quietly funnelled by it.

Introduce Small, Deliberate Friction

Not all friction is bad. Some friction is a seatbelt.

Moments that interrupt drift are moments that return agency.

A pause before auto-play
A meaningful “keep watching?” prompt (not theatre)
Clear off-ramps: “save for later”, “mute this topic”, “reset interests”

For vulnerable users, these are not obstacles. They’re safeguards.

Design Defaults Like They Matter - Because They Do

Most users never change settings. Defaults aren’t neutral. Defaults are policy.

If “reduce personalisation” exists but is buried, you haven’t built sovereignty. You’ve built plausible deniability.

A simple test: if a journalist screenshotted your default settings, would you feel proud or twitchy?

The Designer’s Dilemma, The Business Model Problem, And The Clickless Web

Let’s be honest. Most designers aren’t operating in a moral vacuum. We’re operating inside businesses.

And many businesses still run on the same engine: attention in, revenue out.

In ad-funded models, cognitive sovereignty often costs money.

Less scroll time means fewer impressions
Less precision means less valuable targeting
More friction means fewer conversions
More agency means less predictable behaviour

That’s not a design problem. It’s an incentive problem.

Now add AI summaries, and the incentives get sharper.

Pew’s analysis of Google browsing behaviour found that when an AI summary appeared, people clicked traditional search results far less often than when no AI summary appeared. Links inside the summary were clicked even less.

That matters because a lot of the open web still runs on a simple exchange: you publish something, search sends you traffic, you monetise the visit.

If the answer is delivered before the click, the model flips.

If fewer people reach your site, they don’t load your pages. If they don’t load your pages, they don’t see your ads. They don’t trigger video inventory. They don’t click affiliate links. The whole “publisher monetises the visit” loop starts to collapse.

Publishers can see where this is heading. Reuters Institute’s Trends and Predictions 2026 report describes an expectation that search traffic will fall substantially over the next few years as search becomes more answer-led.

So the tension isn’t only “sovereignty costs money”.

It’s also: the click economy is shrinking, and the remaining incentives reward anything that keeps people inside the summarising surface. That tends to mean less wandering, fewer off-ramps, and more “here’s the answer, stay here”.

There’s a second-order effect too. When the click becomes optional, the safest commercial bet is to produce content and interfaces that are maximally “answerable”. Smooth. Extractable. Summarisable. Less nuance, fewer edges, less context. The kind of information that fits neatly into a paragraph, and rarely forces anyone to think twice.

Google’s public line is that AI in Search drives more queries and “higher quality clicks”, and that Search still sends billions of clicks to the web every day. That may be true and it still doesn’t change the design reality: the web is being reshaped around on-platform answers, and that shifts power towards whoever controls the answering surface.

This is why regulation is starting to bite at the system level. The EU Digital Services Act explicitly pushes recommender transparency and options to influence parameters, and it places systemic risk obligations on very large platforms and search engines.

And it’s why business models matter more than ever. Subscription and utility-driven products have a more natural alignment between user trust and product success, because they’re not trying to monetise drift.

None of this magically fixes the problem. But it changes the negotiation.

And it makes one thing clearer: when answers become ambient and frictionless, agency doesn’t just erode inside feeds. It erodes at the level of the web’s plumbing.

A Practical “Sovereignty Audit” You Can Run This Week

If you’re a UX team building personalisation, here’s a lightweight audit you can run without launching a new “initiative”.

The Four Checks

Legibility
Can a user understand why this was shown - right now, in one sentence?
Agency
Can a user choose a meaningfully different mode (less personalised, more breadth) without the experience becoming punishment?
Escape
Can a user break a loop easily - mute, reset, browse broadly, switch to unranked?
Respectful Defaults
If the user never touches settings, are we still behaving like we respect their mind?

Three Signals To measure (Not Just Engagement)

Mode switching: do people use different modes at different times?
Reset behaviour: do people feel the need to wipe and restart their inferred profile?
Regret markers: “I didn’t mean to spend that long” or “I don’t know why I’m seeing this” (survey, diary studies, intercepts)

If your only dashboard is “time on platform”, you’re blind to sovereignty by design.

Intelligence Shouldn’t Eclipse Empathy

As AI systems get smarter, they get better at predicting what you’ll do next.

That doesn’t automatically mean they’re serving you.

They might be optimising based on your past behaviour without accounting for who you’re trying to become. They might be making you comfortable when you need challenge. They might be keeping you in the warm bath because warm baths monetise nicely.

The goal isn’t to build systems that replace judgement.

It’s to build systems that support judgement - more options, clearer context, better off-ramps, visible trade-offs.

Cognitive sovereignty isn’t anti-personalisation.

It’s the idea that personalisation should answer to the person, not the business model.

And as designers, we don’t get to pretend our interfaces are neutral.

They shape attention. They shape discovery. They shape belief. They shape who people become.

So here’s the line I keep coming back to:

If your product is choosing people’s next thoughts, you’ve taken on the job of a teacher. Act like it.

Sources

Spotify Research — Socially-Motivated Music Recommendation

https://research.atspotify.com/2024/6/socially-motivated-music-recommendation

Pew Research Center — Google users are less likely to click on links when an AI summary appears in the results

https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/

Reuters Institute — Journalism, media, and technology trends and predictions 2026

https://reutersinstitute.politics.ox.ac.uk/journalism-media-and-technology-trends-and-predictions-2026

Search Engine Land — How AI answers are disrupting publisher revenue and advertising

https://searchengineland.com/ai-answers-disrupting-publisher-revenue-advertising-465185

Google blog — AI in Search: driving more queries and higher quality clicks

https://blog.google/products-and-platforms/products/search/ai-search-driving-more-queries-higher-quality-clicks/

EU Digital Services Act (Regulation (EU) 2022/2065) consolidated text

https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32022R2065

Juniper Research — Subscription economy to reach $1.2 trillion by 2030

https://www.juniperresearch.com/press/subscription-economy-to-reach-1tn-by-2030/

If You Enjoyed This Piece

Follow me for more articles about design-driven coding, usability, and the messy, wonderful middle ground between aesthetics and logic.

Author’s Note

I’m currently deeply interested in using AI to generate both visual and text-based content. I’m actively collaborating with AI on multiple platforms to explore my thoughts on what creativity is and is not. My current approach is to collaborate with AI by using the output as a foundation upon which to build and modify.