AI deceptive design pattern detector
AI-powered flows are fertile ground for dark patterns: a model that “personalizes” an upsell can quietly manufacture urgency, bundle consent, or shame users out of opting in. Regulators have caught up — the EU Digital Services Act (Art. 25), the Unfair Commercial Practices Directive, the UK CMA’s guidance, and the US FTC all now target deceptive design. This detector reads your flow description and flags wording associated with known dark patterns, mapped to the regime you select.
How it works
You describe the flow in plain language, including the exact button labels, defaults, and any countdowns or scarcity claims. The detector runs a library of pattern matchers grouped by dark-pattern type — confirmshaming, manufactured urgency, hidden subscription / forced continuity, consent bundling, pre-ticked boxes, hidden costs, and trick wording. Each match shows the phrase that triggered it, why it is risky, the citation for your chosen jurisdiction, and a compliant redesign. Everything runs locally; nothing is uploaded.
Why AI-powered flows deserve special scrutiny
Traditional dark patterns were manually designed — a product team deliberately set a button color or wrote a misleading label. AI-powered flows introduce a different risk: the pattern can emerge from optimization without anyone explicitly choosing it. A model fine-tuned to maximize subscription conversion may learn to emphasize scarcity, downplay the cancel option, or craft opt-out copy that subtly shames the user — all without a human ever writing that specific text.
This matters for regulators. The EU DSA Article 25 prohibition on deceptive interfaces applies regardless of whether the deception was intentional or learned. A company cannot defend a manipulative AI-generated upsell by arguing the model produced it autonomously. Responsibility stays with the deployer.
The main pattern categories and what makes them problematic
Confirmshaming appears when the decline option is worded to make refusal feel embarrassing. The classic form is “No thanks, I prefer to pay more” as the dismiss button on a discount offer. The neutral compliant alternative is simply “No thanks” or “Not now” — a label that does not characterize the user’s choice at all.
Manufactured urgency uses countdowns, low-stock warnings, or viewer counts to pressure a decision that has no real deadline. Legitimate urgency (a sale that genuinely ends at midnight, a seat that is actually limited) is compliant. A countdown timer that resets on refresh, or a stock indicator disconnected from real inventory, is the pattern regulators target.
Consent bundling packages multiple consents into a single accept button, making it impossible to agree to terms of service without also agreeing to marketing emails or third-party data sharing. GDPR requires each consent to be separately and freely given; bundling makes that structurally impossible.
Hidden subscription / forced continuity involves enrolling a user in a recurring charge at the end of a trial without making the auto-renewal, its cost, and the cancellation method clearly visible before the user confirms. The FTC’s Negative Option Rule and EU subscription legislation both require these terms to be disclosed prominently, not buried in fine print.
Pre-ticked boxes are opt-in checkboxes that start checked. Under GDPR a pre-ticked box does not constitute valid consent. The compliant alternative is an unchecked box with a clear label.
How to describe your flow for the best results
The detector works better with precise descriptions than with summaries:
- Include the exact text of every button label, including the decline or skip option.
- State what is pre-selected or defaulted when a user first arrives at each screen.
- Mention any timers, stock counters, or social proof claims displayed near the call to action.
- Note the visual hierarchy — which elements are large and prominent, which are small or grey — because layout tricks (tiny decline links, greyed-out alternatives) are a major category that plain-text matching can only partially capture.
Notes and limits
- Describe defaults explicitly. Many dark patterns live in defaults (pre-ticked, auto-renew, opt-out) — say what is selected before the user acts.
- Pattern matching is a first pass. It catches common wording but cannot see layout tricks like tiny decline links or color-weighted buttons; review those visually too.
- Symmetry is the fix. The cleanest compliant pattern is a symmetric choice: equally prominent accept and decline, neutral labels, no pre-selected option.
- Confirm with counsel. Use the flags to prioritize, then get a qualified legal review before you rely on a compliance conclusion.