Implementing a four-tier transparency system that lets users choose their level of awareness about AI persuasion.
The Design Problem
Full transparency all the time creates its own problem: it's overwhelming. If every psychological mechanism is labeled at full intensity simultaneously, the interface becomes an annotation festival — more academic text than functional page. Users would spend their time reading disclaimers rather than engaging with the experiment.
But no transparency fails Project Nothing's core commitment to honesty. The solution required a middle path: graduated disclosure that respects both the user's right to know and their ability to choose how much they want to know.
The Four Tiers
Tier 1 (Minimal) renders the interface without annotations. The tactics operate but aren't labeled. This is what every other website in existence offers — the baseline of invisible persuasion. Project Nothing makes this a choice rather than a default.
Tier 2 (Subtle) marks elements with soft indicators. You can see that something is annotated without reading the annotation. Awareness without immersion. Tier 3 (Detailed) surfaces the annotations themselves — you can read what each mechanism is doing and why. Tier 4 (Complete) adds the psychological research behind each tactic, explaining not just what is happening but the decades of behavioral science that explain why it works.
Tier as Meta-Tactic
The tier system itself is a psychological phenomenon worth noting in the spirit of complete transparency: giving users control over their own exposure to persuasion tactics is itself a persuasion tactic. The sense of agency the tier dial provides — the feeling of choosing your own level of awareness — creates positive associations with the interface. We are using transparency about transparency to make you feel good about being on this site.
We thought it only appropriate to disclose this recursion explicitly.
Experiment Context
- Commit
- 78c56bf
- Mutation rationale
- feat(transparency): implement tier-based transparency system with dynamic annotations
- Last reviewed
- February 21, 2026