When using AI customer support, customers should be told four things clearly: that AI is involved, what the assistant can help with, what information guides its answers, and when a human can become involved.
This disclosure does not need to sound technical or alarming. For a small business or lean team, the goal is to set realistic expectations while showing that the business remains responsible for the support experience.
A useful customer-facing explanation might be:
You’re chatting with Coffey, our AI customer support assistant. Coffey can help with questions using information we have approved for customer support. It can work through capabilities we control, and your conversation can be escalated to our team when human help is needed.
The exact wording should reflect how the assistant is configured. Treat examples in this article as suggested language, not universal legal or compliance requirements.
Should a business tell customers that AI is involved?
In most customer-support experiences, clear disclosure is the more trustworthy approach.
Customers should not have to guess whether they are speaking with a person. Telling them that an AI assistant is involved helps them understand why a response may be immediate, how the conversation works and what to expect if the question needs human attention.
Disclosure does not mean leading with a warning or making the experience feel impersonal. It can be brief, direct and part of a helpful introduction:
- “I’m Coffey, an AI customer support assistant.”
- “I can help with common questions using this business’s approved support information.”
- “If your question needs human judgement, I can help connect you with the team.”
The important distinction is between transparent disclosure and overemphasis. Customers need enough information to make sense of the interaction, not a long technical explanation about models, prompts or system architecture.
A business should also avoid implying that disclosure alone solves every trust issue. Customers judge the experience through the accuracy of the answers, the clarity of the assistant’s limits and the quality of the human escalation path.
The four questions your disclosure should answer
A practical disclosure framework starts with four plain-language questions.
1. Is AI involved?
Say plainly that the customer is interacting with an AI customer support assistant.
This information can appear in the assistant’s opening message, the support interface, a help page or another visible point where the conversation begins. The wording should be consistent with the assistant’s actual role.
Suggested wording:
You’re chatting with Coffey, our AI customer support assistant.
Or:
Coffey provides AI-assisted customer support and can help answer questions using our approved information.
Avoid hiding the disclosure in dense terms and conditions while presenting the assistant as though it were a human employee. A short introduction is usually clearer and more useful.
2. What can the assistant help with?
Customers should understand the kind of help available. Describe the assistant by its support role rather than by broad claims about intelligence.
For example:
Coffey can help answer customer questions using information approved by our business.
This tells customers what to try first without suggesting that every possible request can be completed automatically.
If the assistant can work with live data through business-controlled capabilities, explain that in outcome-focused language:
Coffey can provide support using current business information through capabilities we have defined and control.
The business should only mention capabilities that are genuinely available in its configuration. Do not describe an action as available simply because it would be useful or technically imaginable.
A good disclosure separates three ideas:
- Answering: providing information from approved business knowledge.
- Acting: using live data or other capabilities the business has deliberately enabled.
- Escalating: involving a human when the question requires judgement or falls outside the assistant’s role.
That separation helps customers form an accurate mental model of the support experience.
3. What information guides its answers?
Customers do not need a technical description of how an AI system generates language. They do need confidence that the business has decided what information should guide support answers.
Coffey is positioned around answering using approved knowledge. That phrase can be translated into customer-friendly language:
Coffey uses support information that our team has reviewed and approved.
Or:
Answers are guided by the information we provide for customer support.
This is more meaningful than saying the assistant is “trained on everything” or “knows all the answers.” It connects the quality of the interaction to information the business can review and maintain.
The business should be prepared to explain what happens when the approved information does not provide a clear answer. For example:
If Coffey does not have enough approved information to answer confidently, your question may be passed to our team.
That sentence sets a realistic expectation. It also avoids presenting an AI-generated answer as authoritative when the relevant business information is missing or unclear.
Approved knowledge is not a one-time task. Businesses change their services, policies and customer guidance. Someone should own the process of reviewing and updating the information used for support. The customer-facing disclosure does not need to describe that internal process in detail, but the business should understand that its public promise depends on maintaining the underlying knowledge.
How should controlled capabilities be explained?
“Controlled capabilities” can sound technical unless it is translated into a simple idea: the business decides what the assistant is allowed to do.
Suggested wording:
Coffey can use selected capabilities that our business has defined and approved.
Or:
Coffey can help with certain support actions, but it operates within the limits set by our team.
This language communicates control without making unsupported claims about permissions, security or compliance.
It is useful to distinguish between an assistant that gives an answer and one that takes an action. A customer may ask for something to be changed, checked or completed, but the assistant should not imply that it has performed the action unless that capability is actually enabled and the action has genuinely occurred.
Before publishing customer-facing wording, ask:
- What can Coffey answer?
- What live information can it use?
- What actions can it take through business-controlled capabilities?
- What should it do when the necessary information is unavailable?
- Which requests must remain with a person?
- What should the customer be told after an action cannot be completed?
Keep the public explanation simpler than the internal control document. Customers need a clear boundary, not a list of implementation details.
How should human escalation be disclosed?
Human escalation should be presented as a normal part of the support design, not as an embarrassing exception.
Customers should know when a person may become involved and what that means. A clear explanation could be:
If your question needs human judgement or falls outside Coffey’s support role, it can be escalated to our team.
Where the business has a specific response process, it can add accurate information about the next step. If no response time or availability commitment has been verified, do not promise one.
Avoid wording that suggests a person is already reviewing the conversation unless that is true. For example, “A team member is looking at this now” should only be used when the business can support that statement.
Better alternatives include:
- “I’ll pass this to the team for human help.”
- “This question needs someone from our team to review it.”
- “I can’t confirm that from my approved information, so a team member should help with the next step.”
The handoff should also be understandable within the conversation. A customer should not need to discover the human-support route by repeatedly asking for an agent.
Where should disclosure appear?
AI support disclosure works best when it appears at the points where customer expectations are formed.
Consider these locations:
The first message
The assistant’s opening message is the most direct place to explain its role. Keep it short enough to read before the customer asks a question.
The support entry point
If customers begin support through a website or customer portal, add a concise explanation near the point where they start the conversation. This prevents the interaction from feeling misleading.
The help or support page
A support page can provide more context about what the assistant handles, how approved knowledge guides answers and when a human may become involved.
The escalation moment
Repeat the relevant expectation when a conversation moves beyond the assistant’s role. The customer should understand why the handoff is happening and what to do next.
Human follow-up
When a person joins the conversation, they can briefly reinforce the transition:
I’m taking over from Coffey so I can review this with you directly.
The goal is not to repeat the same disclosure everywhere. It is to make the assistant’s role clear at each important transition.
Claims to avoid when describing AI customer support
Overstated language creates avoidable disappointment. Businesses should avoid claims that imply capabilities or outcomes they have not verified.
Be cautious with statements such as:
- “Coffey can answer anything.”
- “You will never need human support.”
- “Every answer is guaranteed to be correct.”
- “Coffey understands exactly what you mean every time.”
- “The assistant has unlimited access to your business data.”
- “AI has replaced our support team.”
- “Your issue will always be resolved instantly.”
- “A human is always available,” unless that is genuinely true.
These claims go beyond describing the verified role of an AI customer support assistant. They also blur the distinction between always-on availability and unlimited capability.
A more credible alternative is specific and bounded:
Coffey is available to help with customer questions using approved knowledge. It works through capabilities controlled by the business and escalates to a human when needed.
A pre-publication disclosure checklist
Before launching or updating AI customer support, confirm that:
- Customers are told that AI is involved.
- The assistant’s role is described in plain language.
- The business explains what kinds of questions it can help with.
- Approved knowledge is described accurately.
- The business does not imply that all information is current unless it can support that claim.
- Controlled capabilities are explained without technical exaggeration.
- Customers know what happens when the assistant cannot answer.
- Human escalation is visible and understandable.
- The wording does not promise a response time, outcome or human availability that has not been verified.
- The same basic expectations appear across the customer journey.
- Internal teams know what the public disclosure promises.
- The disclosure is reviewed whenever the assistant’s knowledge, capabilities or escalation process changes.
Conclusion
Customers do not need AI customer support to sound magical. They need it to be understandable.
The clearest disclosure answers four questions: Is AI involved? What can it help with? What information guides its answers? When can a human become involved?
For small businesses and lean teams, this approach makes always-on support easier to introduce without suggesting that automation replaces human responsibility. Coffey fits that model as an AI customer support assistant that never sleeps, answers using approved knowledge, acts on live data through capabilities the business controls and escalates to a human when needed.
That is the message customers can work with: support is available, the business remains in control and human help is still part of the experience when judgement matters.
