Tell them where they would want to know, or where the law requires it, and say what a person did as well as what the tool did. Disclosure has a measured price. Across thirteen experiments, Schilke and Reimann found that people trust actors who disclose AI use less than those who do not. The same paper found the worst case was being exposed by someone else, which cost more trust than saying so first. The law asks for less than many firms assume: on the European Commission's summary, Article 50 of the AI Act requires telling people they are dealing with an AI system, labelling deepfakes, and labelling AI-generated text on matters of public interest published without human review, from 2 August 2026. Nobody has measured what disclosure does to a real firm's customers.
The answer, in one line
Tell them where a reasonable customer would want to know, or where the law requires it, and say what a person checked. Thirteen experiments found disclosure lowers trust, and that being exposed by a third party lowers it further.
Definition#
AI disclosure: telling the people who receive a piece of work, or who are dealt with by a system, that AI produced or shaped it. A firm chooses whether to do it. The transparency duties in the AI Act are a different matter, attaching only to chatbots, deepfakes and certain published text. The term is descriptive, and neither Schilke and Reimann nor this page treats it as a coinage.
Disclosure lowers trust, and being found out lowers it further#
Schilke and Reimann ran thirteen experiments, published in Organizational Behavior and Human Decision Processes in 2025, across communication, analytic and creative tasks. Their central result is that actors who disclose their AI use are trusted less than those who do not. In the final experiment a tax advisor exposed by a news leak was trusted less than one who disclosed voluntarily, though voluntary disclosure still cost trust against saying nothing. The penalty was smaller among people with positive attitudes to technology and among those who rated the AI as accurate, and it did not disappear for them.
The workplace version of the same pattern is in should I tell my boss I used AI, where Reif, Larrick and Soll found a penalty for described AI use that shrank when the tool's fit for the task was stated. Schilke and Reimann did not test that wording, so the two results are neighbours and not a pair that confirms each other.
The AI Act names three duties, and ordinary business documents are not among them#
The Commission's factpage on Article 50, dated 29 July 2026, says providers must design systems so people are told they are interacting with an AI system, and must apply machine-readable marks to synthetic content. Deployers must label deepfakes and AI-generated text published on matters of public interest without human review or editorial control. The duties apply from 2 August 2026, with a grace period until December 2026 for generative AI systems placed on the market before that date, and fines reach 15 million euros or 3 per cent of worldwide annual turnover. On that summary, a proposal, a report or a marketing email drafted with AI and checked by a person does not fall under a labelling duty. A customer-facing chatbot does. The factpage is a summary and not the regulation, and anyone deciding on compliance should read Article 50 and the Commission's guidelines directly.
Three gaps between the experiments and your customers#
- Described actors, not your account. Participants rated an advisor or a writer they had never met. A client with a history of your work may react differently, and no study here observes real customers.
- Wording was not the variable. The experiments compare disclosure with silence. They do not test a sentence that says what a named person checked or changed.
- Rules outside the EU differ. The Article 50 summary covers the EU Act. Sector rules, contracts and other countries' laws may add duties this page does not cover.
A named person makes the disclosure worth reading#
This section is interpretation, kept apart from the evidence above.
The evidence prices two outcomes, silence and disclosure, and finds a third worse than both: discovery. That leaves the sentence itself as the thing a firm controls. A bare "we used AI" invites the reader to fill in the blank. A sentence that names what the tool produced, what a person checked or changed, and who stands behind the result gives the reader something else to weigh. That is an inference from the pattern of the results and not a tested wording, so treat it as a hypothesis to try on a small scale.
A policy that decides in advance which work is disclosed, and says so in the same words each time, also removes the risk of a customer discovering AI use in a document nobody flagged. The accountability question underneath is covered in can a company blame its AI agent.
Deciding what a customer would reasonably expect to be told#
- Start from the customer's expectation. Ask what a reasonable client would assume about who wrote, analysed or decided the thing they are paying for.
- Separate chatbots from documents. If a customer may believe they are talking to a person, the AI Act duty applies in the EU. For documents the choice is yours unless a contract or sector rule says otherwise.
- Say what a person did. Name the check, and name the person or team accountable.
- Choose the channel. Put the statement in your own terms and conditions or covering note, so nobody learns of it from a leak.
- Keep a record. If you are asked how a piece of work was produced, you should be able to answer without reconstructing it.
Key sources
- Schilke, O. and Reimann, M. (2025). The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes 188, 104405. Graded entry.
- European Commission (2026). Quick facts: transparency rules for AI systems. Digital Strategy factpage, 29 July 2026. Graded entry.
- Reif, J. A., Larrick, R. P. and Soll, J. B. (2025). Evidence of a social evaluation penalty for using AI. PNAS 122(19). Graded entry.
Related SuperSkills research#
On telling colleagues, should I tell my boss I used AI. On honest description of your own use, how to be honest about using AI. On responsibility for what a system does, what is a moral crumple zone. On oversight, can a company blame its AI agent.
About this research#
Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. Both studies and the Commission factpage were read at source on 30 September 2026 and are graded in the evidence base. This page is not legal advice. AI disclosure is a descriptive term and not a SuperSkills coinage.
Evidence review · SS-2026-374 · Graded against the published rubric
Hirji, R. (2026). Should we tell customers we used AI?. The SuperSkills evidence base, SS-2026-374. https://thesuperskills.com/research/should-we-tell-customers-we-used-ai. Last reviewed 30 September 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work