Dark Patterns in the Age of AI: Manipulation Just Got a Lot More Personalised

Not long ago, I was booking train tickets to London with my colleague. We were using the same app and looking at the same journey at the same time. The only difference was that I travel by train regularly, while my colleague rarely does.

As we compared screens, something odd stood out to me. My colleague’s app showed a red "Few tickets left" warning, encouraging a quick decision, whereas mine did not. Same app, same route, same moment but a different message.

The urgency was not about the train journey itself. It was about who the app thought it was talking to. This is a dark pattern: a design that pushes users towards decisions they might not otherwise make.

Dark patterns have existed for as long as digital products themselves. For years, the industry largely ignored them because they improved conversions, reduced drop-out, and grew email lists. Companies accepted them, and often deliberately designed them in, but that equation is starting to shift and Artificial Intelligence (AI) is a major reason why.

Generative AI models are trained on the web and the web is full of dark patterns. So, when AI designs sign-up flows or onboarding copy, it often repeats the manipulative patterns it has seen most frequently because it cannot tell good design from deceptive design. Many teams now use AI to speed up design and development work, which means a manipulative interface can deploy without anyone ever actually choosing to build one.

But that is only part of the story.

Beyond generative AI sits a second, older and in some ways more deliberate problem: AI systems that learn from user behaviour to apply dark patterns in ways that are targeted, adaptive and personalised to each individual. Here, the manipulation is no longer an accident; it is the goal.

These two problems are not the same and they do not have the same fix. One can be a deliberate choice, which means it can also be reconsidered; the other can slip into a product unnoticed, which means it has to be actively looked for.

Telling them apart is the first step, because in an era of AI-generated interfaces both can quietly produce frustrated users, regulatory scrutiny and even lasting reputational damage.

The most common Dark Patterns

That train ticket warning is a textbook example of one type of dark pattern: the urgency illusion - fake scarcity or time pressure, such as a "Few tickets left" message or a countdown timer - designed to push users into deciding quickly. Dark patterns are not bugs. They are decisions, and the urgency illusion is only one of the forms they take.

Some of the other most common forms are:

• The roach motel: Signing up is simple but cancelling is a nightmare. For example, a subscription can be started in a few clicks but requires navigating multiple menus or contacting customer support to cancel
• Confirmshaming: Opt-out options are phrased to make users feel guilty. For example: "No thanks, I don't want to save money"
• Hidden costs: Extra fees appear late in the checkout process. For example, booking fees, service charges, or delivery costs only become visible on the final payment screen. This is often called “drip pricing” and it is now one of the practices UK regulators are actively fining businesses for
• Trick questions: Confusing wording, pre-ticked boxes, and double negatives obscuring what users are agreeing to. For example: "Uncheck this box if you do not wish to not receive marketing communications"
• Misdirection: Design choices such as colour, size, and placement draw attention to the company's preferred option. For example, the "Accept All" button is bright and prominent while the privacy-friendly alternative is small and greyed out

These patterns are everywhere, in e-commerce, subscription services, social media platforms, travel booking sites and increasingly in apps you'd expect to know better. They work because they exploit predictable quirks of human behaviour: our urge to avoid losses, the "decision fatigue" that sets in after too many choices and our tendency to stick with whatever option is pre-selected.

Because these biases are so consistent, they can be deliberately targeted and that last point is where everything changes.

Enter AI: When manipulation learns your name

The real shift comes here; traditional dark patterns are applied uniformly to all users and they exploit common psychological biases, but they are ultimately one-size-fits-all tactics rather than personalised forms of persuasion.

AI changes the architecture of manipulation entirely.

The first is personalised friction. AI systems can learn from a user's browsing history, purchases and in-session behaviour to build a profile of how they make decisions, recording whether they are price sensitive, responsive to scarcity, influenced by social proof or more likely to abandon at checkout. That profile is then used to serve each person a version of the checkout tailored to where their resistance is weakest, so different users may see different persuasive designs in the same flow. It is most likely why my colleague, whose travel habits marked them as a less frequent buyer, saw a "Few tickets left" nudge that I never did.

The second is dynamic pricing. Repeated searches for the same flight or hotel can signal buying intent. Platforms can use those behavioural signals to adjust prices or offers over time - a practice already common in travel systems and made more granular by AI-driven prediction models.

The third is conversational manipulation. A chatbot handling cancellations or refunds is rarely neutral. If its goal is retention, it can adapt tone, framing and offers in real time based on the user's responses, making the interaction feel personal while still working to reduce churn.

The change is not that design has suddenly turned unethical. It is that manipulation, which once treated everyone the same, can now be aimed at each person individually, which makes it both more effective and much harder to detect.

Why this is also a business problem

It is easy to see dark patterns as just an ethical problem, something for designers or regulators to worry about, but they are a business problem too:

• Trust is expensive to rebuild: When users feel manipulated, even if they cannot quite put their finger on how, that feeling sticks. They are less likely to recommend the product, more likely to leave a bad review, and quicker to leave the moment something better comes along. Winning a new customer usually costs a lot more than keeping one.
• The legal risk is growing: Consumer protection regulators in the UK, EU, and US have all started taking real action against deceptive design. The EU's Digital Services Act, the UK's Digital Markets, Competition and Consumers Act (DMCCA) and the Federal Trade Commission's (FTC) stance on dark patterns all point the same way. Recent cases highlight this risk. In 2026 the Competition and Markets Authority (CMA) in the UK issued its first fine under the new Act, fining the AA £4.2 million and ordering around £760,000 in refunds after its driving-school websites added a mandatory £3 booking fee that was hidden until the final checkout screen. In 2025 the FTC reached a $2.5 billion settlement with Amazon over the way Prime sign-up and cancellation were designed, and a 2024 FTC and Department of Justice (DOJ) case targeted Adobe over a hidden early-termination fee. The EU's Digital Services Act now bans dark patterns for large platforms outright.
• Personalised manipulation at scale: Now, more than ever, this will almost certainly draw far more attention than static dark patterns ever did. Collecting data, guessing at behaviour, and targeting people based on it sits very close to several areas of consumer and data protection law that already exist.
• Frustrated customers are loud customers: Social media and review sites have completely changed what a bad experience costs a business. Someone stuck in a subscription they cannot cancel or quietly charged more at just the wrong moment, now has a public platform to say so. A happy customer tends to stay quiet, but an angry one rarely does, so the damage from manipulative design usually spreads far beyond the people it actually happened to.

The alternative is to design things that genuinely help people do what they came to do, make consent easy to understand, and use AI to remove friction rather than create it.

How to design better: AI as an ally

The irony is that the same capabilities that make AI dangerous in the hands of manipulative design can make it genuinely useful in the hands of ethical design. The tools are not the problem.

Used well, AI can make a product clearer instead of more confusing. The same data that shows where people hesitate at checkout can be used to fix the real problem rather than paper over it. If users keep dropping out at the payment step because delivery costs only appear at the end, a manipulative design adds a timer and a pre-ticked box to rush them through whilst a better one simply shows the cost up front.

AI can also be pointed at the design itself, to flag dark patterns rather than just chase clicks and to choose defaults that genuinely help users. AI can also turn dense privacy policies and terms into plain language, because a user who understands what they are agreeing to is far less likely to feel tricked later.

Closing thoughts

I only noticed the train app because my colleague and I happened to compare screenshots. On my own, I would never have known there was another version of the same screen, showing someone else a different message. That is what worries me as a designer. Dark patterns used to be one trick shown to everyone, something you could learn to spot. Now the interface quietly learns each person and shows them whichever version is most likely to work, which makes it much harder to notice at all.

What stays with me is that none of this is the fault of the tools. The same data that decided which version to show me could just as easily be used to make that booking clearer and more honest. That choice sits with whoever is designing the screen.

So the question I try to keep asking of my own work is a simple one: would I be comfortable if the person using this could see exactly why I designed it this way? When the honest answer is no, I know that is where I need to start again.

References

1. AI Critique (2025) Generative AI and Dark Patterns in UX Design. AI Critique. Available at: https://www.aicritique.org/us/2025/03/27/generative-ai-and-dark-patterns-in-ux-design
2. CMA (2026) CMA issues multi-million drip-pricing fine in enforcement first. Pinsent Masons. Available at: https://www.pinsentmasons.com/out-law/news/cma-issues-multi-million-drip-pricing-fine-enforcement-first
3. European Commission (n.d.) The Digital Services Act. European Commission. Available at: https://digital-strategy.ec.europa.eu/en/policies/digital-services-act
4. FTC (2024) FTC takes action against Adobe and executives for hiding fees and preventing consumers from easily cancelling. Federal Trade Commission. Available at: https://www.ftc.gov/news-events/news/press-releases/2024/06/ftc-takes-action-against-adobe-executives-hiding-fees-preventing-consumers-easily-cancelling
5. FTC (2025) FTC v. Amazon.com, Inc. (ROSCA). Federal Trade Commission. Available at: https://www.ftc.gov/legal-library/browse/cases-proceedings/2123050-amazoncom-inc-rosca-ftc-v
6. UK Government (2024) Digital Markets, Competition and Consumers Act 2024. legislation.gov.uk. Available at: https://www.legislation.gov.uk/ukpga/2024/13/contents
7. UX Design Institute (2025) What are dark patterns in UX? All you need to know. UX Design Institute. Available at: https://www.uxdesigninstitute.com/blog/what-are-dark-patterns-in-ux