Govern
AI Presence Disclosure
Am I interacting with an AI?
Makes AI involvement visible at the moment it matters, including labelling synthetic voices in calls.
- Evidence: research-based (Amershi et al. 2019, Mozafari et al. 2022)
- Tested: concept only, mock data
Overview
Why it matters
People trust a machine differently than a person. They need to know which one it is.
People ask different questions, apply different scrutiny and place different trust depending on whether a human or a machine is involved. When AI involvement is hidden, that judgment is distorted. Disclosure is now a legal requirement in many places, but it is also a design decision about how people engage with what they see.
What does it look like?
Applicant receiving an AI-assisted decision
Applicant receiving an AI-assisted decision
- Applicant receives a loan decision.
- Near the decision: an AI badge: "This decision was supported by an automated assessment."
- Applicant clicks the badge.
- Tooltip: "An AI system scored your application based on financial history, employment data, and credit records. A human caseworker reviewed the recommendation before this decision was issued."
- Link: "Learn about your rights regarding this automated assessment."
Conversational interface
Conversational interface
- User opens a customer service chat.
- First message from the system: "Hi: you're chatting with an automated assistant. For complex issues, I can connect you with a human agent."
- The assistant's name/icon is clearly distinct from a human agent's.
- If the conversation is escalated to a human, the transition is explicitly announced.
When does it fit?
Use it when
- AI makes or influences decisions that affect people
- A conversational interface could be mistaken for a person
- AI output could be mistaken for human work
Skip it when
- The AI role is trivial, like a spell-checker
- It's already obvious, and a notice would be patronizing
Design
How do I apply it?
-
Disclose where it matters, not at login.
When people are about to rely on the output, not six screens earlier.
-
Say what the AI is doing.
"An AI system assessed your eligibility" beats "Powered by AI".
-
Say what the human role is.
Will a person review it? Say so. If not, say that too.
-
Don't use disclosure to deflect.
Saying it was AI doesn't move responsibility away from you.
-
Keep it noticeable.
People tune out permanent labels. Repeat the notice at high-stakes moments.
-
Design for the trust cost.
Disclosure can lower trust at first. Pair it with what the AI does well and a path to a person.
How can I show it?
AI involvement badge
A small marker next to the AI output.
A compact, persistent visual indicator showing that AI is involved in the current output or decision: an icon, a label, or a chip. Positioned near the output it relates to, not in a global header where it is quickly ignored.
System identity disclosure
"You are talking to an automated assistant."
For conversational interfaces: a clear statement of what the user is interacting with: "You are talking to an automated assistant. This is not a human." Required when there is any risk of confusion.
How AI is involved explanation
What the AI is doing in this interaction.
A short explanation, such as a tooltip, collapsible section, or link, explaining specifically what the AI is doing: "AI is being used to assess your application and generate a recommendation. A human caseworker will review this recommendation before a final decision is made."
Role-of-human clarification
Who decided: the AI, a person, or both.
When AI and humans are both involved, clarify the division of responsibility: "The AI scored this application. A human made the final decision." Or: "This recommendation was generated by AI and has not been reviewed by a human."
Rights disclosure link
A direct link to people's rights about this decision.
For automated decision-making that triggers legal rights (GDPR Article 22), a visible link to: "Your rights in relation to this automated decision", not buried in a privacy policy.
Further left for background AI, further right for automated decisions that affect people's rights.
What does it cost?
Transparency vs. habituation
Labels everywhere get ignored. Save the prominent ones for high-stakes moments.
Accuracy vs. comprehension
Describing every model and layer is unreadable. Say what people need to judge the output.
Compliance vs. real awareness
A footnote can meet a legal minimum and still inform nobody.
Connections
Which patterns work with it?
AI-Generated Content Marking
If this leaves the system, will it still say AI made it?
Presence disclosure is system-level and lasts only as long as the interaction. Content marking is artifact-level and has to survive the artifact leaving the system entirely. Use both together: disclosure for the conversation, marking for anything the conversation produces.
Model Transparency Cards
How was this system built and tested?
Presence disclosure says AI is involved. Transparency cards explain what that AI is, what it's designed for, and what its limitations are. Disclosure is the entry point; transparency cards are the detail.
Agent Action Transparency
What actions is the AI agent taking?
In agentic systems, AI presence is not just about who you're talking to: it's about what actions are being taken on your behalf. Agent action transparency extends disclosure to cover autonomous action.
Human-in-the-Loop
Who can review or override this decision?
Disclosure of AI involvement should always clarify the human role: especially whether a human is involved at all, and at what stage.
Consent & Data Use Transparency
What data is being used and how?
Disclosure of AI involvement and disclosure of data use are related but distinct. A user who knows AI is involved still needs to know what data is used and how.
Sources (4)
Sources (4)
Amershi et al. (2019) — Guidelines for Human-AI Interaction
Guideline G1 ("make clear what the system can do") and G4 ("show contextually relevant information") directly support the disclosure design approach of this pattern
European Union (2024) — EU AI Act, Article 50
requires transparency and labelling obligations for AI systems interacting with humans. The primary regulatory mandate for this pattern
Mozafari et al. (2022) — Trust Me, I'm a Bot: Repercussions of Chatbot Disclosure in Different Service Frontline Settings
empirical study showing that disclosing a chatbot's identity can reduce user trust and usage intention, particularly for high-criticality services, and identifying mitigation strategies. Important counter-evidence for disclosure design: the trust cost is real and must be designed for, not denied. Published in Journal of Service Management, 33(2)
European Union (2016) — GDPR Article 22
the right not to be subject to solely automated decision-making, with safeguards including human intervention. The duty to disclose the existence of automated decision-making sits in Articles 13(2)(f), 14(2)(g), and 15(1)(h)
Your users didn't know AI was involved.
Or a regulator wants evidence of disclosure at the point of interaction, not in the terms of service. That's where this pattern starts. Let's look at your interface together.
Work with me