Sooner or later, a client will ask a version of the same question: did AI make this?
The question may arrive after a proposal feels generic, when an image looks too polished, or when a client discovers that a team has used a transcription service in a meeting. It is easy to hear it as an accusation. It is usually a question about trust: what did you do with my information, who checked the result, and who remains accountable for it?
Small firms do not need to respond by placing a disclaimer under every email. They do need a clear internal standard. The standard should distinguish between AI as an assistive tool and AI as an unexamined substitute for professional judgement.
This is becoming more relevant for businesses that work with European clients. The European Commission says certain AI transparency obligations under the AI Act began to apply on 2 August 2026. Their application depends on the system and context, so this article is not legal advice. The broader commercial lesson is simpler: clients are beginning to expect clarity about when AI shapes the work they receive. The Commission’s overview is the appropriate starting point for a legal assessment.
Three questions before disclosure
Before deciding whether to tell a client about AI use, ask three questions.
First, did the tool receive client information? A grammar assistant used on a generic public announcement is different from a system that processes a client brief, interview transcript, customer record or confidential design material. The more sensitive the information, the more the client deserves to understand the arrangement before the work starts.
Second, did the tool materially shape the output? Using AI to generate a first outline, then conducting research, making the argument and editing the final piece is not the same as delivering largely unreviewed output. The distinction is not whether a tool was ever opened. It is whether professional judgement was actually applied.
Third, could a reasonable client be surprised? This is a useful test because it forces the business to look beyond technical definitions. If the client would reasonably expect a human expert to have performed the work personally, or would care that their material was processed in a particular way, silence is a poor long-term strategy.
A proportionate standard
For everyday work, a small firm can adopt a simple three-level standard.
Level one: internal assistance. AI helps with internal notes, formatting, brainstorming or a rough first draft. No client data is entered, and a staff member checks the result. This normally does not require a client-facing disclosure, although it still needs sensible internal rules.
Level two: client-work assistance. AI is used to help prepare work for a client, but the professional remains responsible for the analysis, facts, decisions and final quality. Where relevant, the client should be told that AI-assisted tools may be used under human review, particularly if their information is processed.
Level three: material AI involvement. The system produces, analyses, recommends or interacts in a way that significantly shapes a client outcome. This requires explicit conversation about the tool, the data involved, the review process and the person accountable for the final decision.
The point is not to create perfect labels. It is to prevent staff from making a different judgement every time a client asks.
Write disclosure in ordinary language
Poor disclosure sounds like a legal defence. Good disclosure tells the client what they need to know.
For example: “We use AI-assisted tools for transcription and initial drafting where appropriate. Your confidential material is not entered into public AI services without approval, and a member of our team reviews and remains responsible for all work delivered to you.”
That statement makes four things clear. It names a use, draws a boundary, confirms human review and keeps accountability with the business. If a service uses a particular provider, processes sensitive information or produces recommendations that affect people, the explanation needs to be more specific.
The real work is upstream
Transparency cannot repair a careless process. A business cannot credibly say that people review outputs if review consists of glancing at an answer and sending it on. Nor can it promise responsible data handling while staff are uploading whatever they need to meet a deadline.
The underlying controls are ordinary management controls: approved tools, a boundary for sensitive information, a named person who owns the use case, and a route for raising problems. A minimum viable AI governance framework can be small and still be real.
The client’s question is therefore useful. It asks whether the firm has retained the judgement, confidentiality and accountability that justified the client relationship in the first place. If the answer is yes, transparency is not a burden. It is evidence of how the work is being managed.