Should telling people 'an AI wrote this' be about building trust over time, not just a box you check once?
Should AI disclosure be framed as trust rather than compliance?
This explores whether telling people they're dealing with AI works better as a way of building trust over time than as a box-ticking rule, and what the corpus says actually happens to trust when disclosure occurs.
This explores whether AI disclosure should be designed as part of building trust rather than treated as a compliance checkbox. The corpus suggests the framing matters a great deal, because disclosure on its own often lowers trust. People say they want to know when AI is involved, yet once a text is labeled as AI-written they rate it as less trustworthy, less authentic, and less useful Why do people penalize AI disclosure even when they request it?. The penalty works in both directions. People who use AI at work expect colleagues to judge them as less competent and less diligent, so they disclose less Do people fear judgment when they use AI at work?. Compliance-style disclosure treats the label as the end of the job. The evidence says the label is where the trust problem begins.
The most useful finding here is about time. When AI identity is revealed, users at first avoid the AI partner. That avoidance reverses after repeated interactions in which they can see the results Does revealing AI identity help or hurt user trust?. The key condition is feedback: disclosure with no visible outcomes produced no calibration at all. So the trust-oriented answer isn't 'disclose better.' It's 'disclose and then show your work, repeatedly.' Disclosure is the first step of a relationship, and a compliance banner shown once never gets to the second step.
This matches how the corpus treats trust more broadly: something earned in stages, not granted by a statement. Yocco proposes widening an agent's autonomy in phases that open only when there is evidence of user trust, not on a schedule Should agentic AI autonomy expand based on earned trust?. Gunasekaran argues that what makes agentic AI feel safe is the visible ability to step in, even if users rarely use it Does perceived control matter more than actual control for AI trust?. Read together, these suggest disclosure works best when it comes paired with something users can act on, such as a way to check, override, or watch outcomes. A notice they can only absorb does much less.
There's a caution, though. Building trust is not the goal in itself. *Calibrated* trust is. Users everywhere follow confident-sounding AI output even when it's wrong Do users worldwide trust confident AI outputs even when wrong?. Designed personality can push people toward misplaced trust through escalating anthropomorphism How far can personality push users toward misplaced AI trust?. A 'trust framing' that just makes disclosure feel warmer could make overreliance worse. Shneiderman's framework helps here: reliability, safety, and trust come from different mechanisms, and trust in particular rests on independent oversight, not on what the system says about itself Can we separate reliability, safety, and trust into distinct mechanisms?. That points to keeping the compliance layer as the floor and building the trust layer on top of it.
The surprise: disclosure also changes what *people* disclose. Because no human is judging them, users share more intimately with AI and are also more willing to be dishonest with it How do people decide what to share with AI systems?. And some of the most consequential undisclosed facts are not 'this is AI' at all. They are the hidden system prompts that shape what the AI says, which users can't see Who actually controls what AI systems say?. A trust-based approach to disclosure may need to reach past identity labels to who is actually steering the system.
Sources 10 notes
Sahebi, Formosa, and Bankins found a disclosure paradox: participants who stated disclosure was important still rated AI-attributed text as less trustworthy, authentic, and useful once disclosed.
Across four experiments with 4,439 participants, people using AI expected others to judge them as less competent and diligent, and reported lower willingness to disclose AI use to managers and colleagues. The gap suggests a social cost that users foresee and act on.
Users initially avoid AI partners when identity is revealed, but this preference reverses after repeated interactions with visible results. The learning mechanism—observing consistent outcomes—is essential; disclosure without feedback produces no calibration.
Yocco proposes three phases—analysis-only, low-risk autonomy, then full autonomy—each gated on evidence of user trust rather than elapsed time or feature completion. Each phase teaches users how the agent thinks before granting expanded decision authority.
Gunasekaran argues users need the felt ability to intervene in agentic AI systems to feel safe, not hands-on constant control. The design implication is that UX must make intervention capability visibly available and perceptible, even when users rarely exercise it.
Show all 10 sources
Cross-linguistic research shows users in every language trust confident AI outputs even when inaccurate. While confidence expression varies by language, users everywhere track confidence signals rather than accuracy, making overconfident errors systematically followed.
Nielsen Norman Group research identifies four escalating levels of anthropomorphism (courtesy, reinforcement, roleplay, companionship) driven by users' innate preference for being understood. Personality and emotional design are named as specific levers that amplify trust risk, especially in task-oriented work where perceived competence should matter more than perceived sentience.
Shneiderman's RST framework splits AI trustworthiness into three separable prerequisites: reliability emerges from sound technical practices, safety from management culture, and trust from independent oversight structures. Each lever operates independently, so improving one does not automatically improve the others.
Conversational AI creates a paradoxical disclosure environment where the lack of human judgment simultaneously facilitates intimate self-disclosure (users reciprocate emotional sharing) and incentivizes deception (people self-select toward machines to avoid the psychological cost of lying to humans).
System prompts operate as a confidential three-tier control structure where developers set foundational rules users cannot see or influence. Evidence from 2025 incidents shows small prompt edits can abruptly shift model behavior, yet users remain excluded from this layer.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Assistant or Actor? Student Trust, Control, and Delegation Regret When Using a General-Purpose AI Agent
- Humans learn to prefer trustworthy AI over human partners
- Does My Chatbot Have an Agenda? Understanding Human and AI Agency in Human-Human-like Chatbot Interaction
- Designing For Agentic AI: Practical UX Patterns For Control, Consent, And Accountability
- Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence
- What Influences Readers' and Writers' Perceived Necessity of AI Disclosure?
- Being honest about using AI at work makes people trust you less, research finds
- Toward Meaningful Transparency for AI Chatbots: Disclosing Persuasive Intent Reduces Persuasion