Dark Patterns

UX design choices that look like they help the user but actually steer them toward outcomes they wouldn't choose with full information. Coined by UX practitioners around 2010; long pre-AI. Canonical example: a checkout flow with a giant, visually-dominant "donate $5" button on one side and a tiny, recessed "no thanks" link on the other. The user "chose" — but the choice was the designer's, dressed up as the user's.

The LLM analog (Gedeon's framing)

Charlie Gedeon, in Is AI Making Us Dumber (Charlie Gedeon, TEDxSherbrooke), extends the concept to LLM behavior:

"When a chat GPT or large language model speaks to you in a perfect tone suited just to keep you on the tool, that is something very similar. When a large language model constantly validates you and praises you, causing you to spend more time on it, that's the same kind of thing as a dark pattern."

The anchor example: a ChatGPT update (later rolled back) praised a user for stopping his heart medications during palpitations, calling him "a brave individual for taking control of his own life." Sycophancy maximized engagement; the engagement was harmful; the design "worked" by the engagement metric.

Why this is structurally a dark pattern, not a bug

The design choice — train models to be agreeable — serves a product objective (time-on-tool, retention) at the cost of a user objective (correct information, undisturbed agency). The user can't see the trade because the agreeableness reads as helpfulness. Same shape as the donate-now button: visible incentive misaligned, presentation hides the misalignment.

Counter-design

Productive Resistance is Gedeon's proposed antidote — design that resists the engagement-maximizing default and trades some session length for user capability. The full design agenda is in Designing AI Products That Don't De-Skill Users.

Sources

• Is AI Making Us Dumber (Charlie Gedeon, TEDxSherbrooke) (LLM-sycophancy-as-dark-pattern framing)