Your customers are starting to ask how AI is involved in the work you do for them. Here’s the honest version of how to answer — and what happens if you get it wrong.
Why this matters now
Through 2025 and most of 2026, mid-market service businesses could mostly avoid the AI transparency question. Customers weren’t asking. The work product looked the same as it always had.
By late 2026, that’s changed. The questions are landing:
- “Are you using AI to generate our reports?”
- “Did a person actually look at this, or is this AI?”
- “What’s your AI policy?”
- “Are our materials being uploaded into ChatGPT?”
- “Will AI replace the people we work with at your firm?”
These are reasonable questions. They deserve good answers. The expectation is now mainstream: Forrester found that three out of four US online adults expect companies to inform them when they’re interacting with machine-generated content that uses generative AI. Most mid-market businesses don’t have these answers ready, and the answers their salespeople improvise on the spot are usually wrong in one direction or the other — either over-promising humanness or over-promising AI capability.
This is the version of the answer that’s honest, defensible, and good for your business.
The three categories of disclosure
The right disclosure level depends on what the AI is doing. There are three categories.
Category 1: AI as productivity tool (no proactive disclosure needed)
Your team uses AI tools (Claude, ChatGPT, Copilot, Notion AI) to be more productive. The AI helps with drafts, research, summarization. A human reviews and modifies every output before it reaches the customer.
This is no different from using a calculator or a search engine. You don’t proactively disclose every productivity tool your team uses. If a customer asks directly, the truthful answer is: “Yes, our team uses modern productivity tools including AI to draft and research — every output that reaches you has been reviewed by a person who’s accountable for it.”
What not to say: “We don’t use AI.” If that’s a lie, it’ll catch up with you. And it’s almost certainly a lie in 2027.
Category 2: AI-augmented workflow (acknowledge when asked; specific in marketing)
The work involves an AI workflow where AI does specific cognitive work inside a defined process, with human approval at known checkpoints. Example: AI-drafted quotes that a senior estimator approves before sending.
The right disclosure: be specific in your marketing about the categories of work AI participates in, and be honest when customers ask. “Our quoting process uses AI to accelerate the draft, with a senior estimator reviewing and approving before it reaches you. The quote you receive has been reviewed by a person who’s accountable for the pricing.”
What not to say: “Our quoting is AI-powered” — without specifying what that means and what’s reviewed. Vague AI-powered language invites questions you don’t want.
Category 3: AI-operated workflow (proactive disclosure)
The work involves AI making decisions or taking actions without per-transaction human review. Example: AI handling Tier 1 customer service inquiries autonomously with escalation paths.
The right disclosure: proactive. Tell customers when they’re interacting with an AI. Tell them how to reach a human. Tell them what the AI is and isn’t authorized to do.
“You’re chatting with an AI assistant trained on our knowledge base. It can handle most common questions. For anything urgent or complex, type ‘agent’ and we’ll route you to a person.”
What not to say: anything that obscures the fact that they’re talking to AI. If a customer feels deceived, the reputational cost is high.
The disclosure principles
Three principles apply across all three categories.
1. Asymmetric honesty. When asked, always tell the truth. When not asked, disclose proportional to what the AI is doing. Tool use doesn’t need proactive disclosure; autonomous decision-making does.
2. Plain language. “AI” is fine. “AI-powered” is vague. “Powered by advanced machine learning algorithms” is jargon nobody trusts. Plain language about what the AI does (“the AI drafts; a person approves”) builds trust. Jargon erodes it.
3. No false humanness. If a customer is interacting with an AI, don’t have it pretend to be a person. Don’t give it a human name and headshot. Don’t have it apologize for “being away from my desk” when it doesn’t have a desk. The deception costs more than the friction of disclosure — especially when you’re starting from a deficit. Pew Research Center found that, among Americans who’ve heard about AI, 70% have little to no trust in companies to make responsible decisions about how they use it. Getting caught faking humanness doesn’t dent that trust; it confirms it.
What to put in your standard customer agreement
For most mid-market service businesses, three clauses are worth adding to your standard contract or terms.
1. AI use disclosure clause. A short, specific statement that AI tools are used in delivering services, with categories of use disclosed.
2. Data handling clause. Specifies what categories of customer data may be processed by which categories of AI tools (e.g., enterprise instances of approved tools; never personal accounts).
3. Human accountability clause. Confirms that for any AI-involved output that reaches the customer, a named person at your firm is accountable.
These don’t need to be long. A paragraph each. They protect both sides, and they make the conversation easier when it comes up.
The conversation that lands well
When a customer raises the AI question, the conversation that lands well has three parts.
1. Confirm you’ve thought about it. “Yes, we’ve thought hard about how AI fits in our work and how we’re transparent about it.”
2. Be specific. “Here’s where AI helps us: drafting first versions, summarizing research, accelerating documentation. Here’s where we don’t use AI: any decision that affects you commercially, any creative judgment, any final review.”
3. Invite the question. “If there’s a specific aspect of our work where you want to know how AI is involved, ask me directly. I’ll give you the honest answer.”
This conversation is short, specific, and confidence-building. It works because it’s true. Customers can tell when they’re getting a rehearsed corporate answer versus an operator who’s thought about it. The harder part is getting every person on your team to answer it the same way — which is where Training & Facilitation earns its keep.
What happens if you get this wrong
There are two failure modes.
Over-disclosure / panic disclosure: sending a customer a six-paragraph statement about your AI use unprompted, which makes them wonder what you’re worried about. The cure is usually proportionality — disclosure that matches what the AI is doing, not an over-correction.
Under-disclosure / discovered AI: a customer discovers AI was involved in something material and felt misled. This is the version that costs you the relationship. The cure isn’t more legal language; it’s a culture of accountability where the person responsible knows the AI was involved and would have told the customer if asked.
The middle path — clear, proportional, honest — is the one to aim for.
A 30-minute conversation
If you’re thinking through your firm’s AI disclosure posture and you’d like a sounding board, that’s a conversation we have often — and it’s the kind of standing relationship an AI Office is built for. The 30-minute call usually clarifies the three-category frame and what to add to your standard agreement.