KoltrixTelling customers an AI helps with your inbox: a plain disclosure
Customers increasingly ask whether a person or a machine wrote the reply. What to disclose, where to say it and plain wording you can adapt for support email.
On this page(10 sections)
- Why say anything at all
- Know what you are disclosing
- The rule of thumb
- Where to say it
- Wording you can adapt
- For a help page
- For an automatic acknowledgement
- For a chat or conversational assistant
- When someone asks directly
- Things not to do
- Make the human review real
- Handling opt-outs and questions
- A one-page checklist
- Key takeaways
Sooner or later a customer writes: "Am I talking to a real person?" Or, worse, they notice a phrase in your reply that sounds like a machine and decide that you have been hiding something.
If your team uses AI to sort mail, summarize threads or draft replies, you do not need to apologize for it. You do need a clear, honest position on what you tell people, and the words ready before anyone asks. This guide helps you decide what to disclose, where to put it and how to phrase it plainly. It is practical guidance, not legal advice. Some places have rules about telling people when they are dealing with an automated system, and those rules differ and change, so check what applies to you with a lawyer or your privacy lead.
Why say anything at all
Three reasons, in order of importance.
Trust. People forgive a tool. They do not forgive being misled. A customer who finds out later that "Sam" was an AI draft nobody read will remember the deception more than the answer.
Accuracy. Telling people that some replies are drafted with AI and reviewed by a person also sets the right expectations: they know a mistake might creep in, and they know how to report one.
Your own clarity. Writing the disclosure forces the team to agree on what the AI actually does. Many teams discover they have not decided.
Know what you are disclosing
Before writing a sentence, list how AI touches customer email in your company. The usual uses fall into four levels, from least to most involved:
- Behind the scenes. Sorting into categories, flagging messages that need a reply, summarizing threads for your staff. The customer never sees AI output.
- Drafting for a person. The AI proposes a reply, and a person reads, edits and sends it. This is the pattern described in AI drafts, human sends.
- Automatic acknowledgement or routine answers. A message goes out without a person reading it first, such as a templated receipt or an FAQ reply. Automated replies vs AI-suggested drafts explains the difference.
- A conversational assistant that talks to customers directly, such as a chat or email agent.
The more a customer's message is handled without a person, the more important and specific the disclosure should be.
The rule of thumb
Match the disclosure to the level:
- Level 1 (behind the scenes): mention it in your privacy notice, not in each email. Say what the system does with the message, who processes it and how long data is kept.
- Level 2 (drafted, reviewed, sent by a person): a general statement in your privacy notice and help pages is enough for most teams. Do not add a footer to every email saying "written with AI", which tells the reader little and makes the person who did review it look absent. What matters is that a human really did read it.
- Level 3 (automatic replies): say so at the point of contact. The message should make clear that it is automatic, and offer a way to reach a person.
- Level 4 (an assistant that converses): say at the start, and never let it claim to be a human. Give a clear route to a person.
Whatever level you are at, never deny it when asked directly.
Where to say it
Put the same truth in several places, each shorter than the last:
- Privacy notice. The complete version: what the AI does with email content, which providers process it, what is kept and for how long, and how to object. Questions about AI and email privacy lists what to cover, and what your AI provider sees when you query your inbox explains the data flow.
- A help-center page titled something like "How we use AI in support". Plain words, a short list.
- Your support auto-acknowledgement, if you send one, with one line.
- The reply itself, only where it matters: automatic or chat-style replies.
- Your team's own policy, so staff know what they may do. A template is in an AI email policy for small teams.
Wording you can adapt
For a help page
How we use AI in support
We use AI tools to sort incoming messages, summarize long conversations for our team and suggest draft replies. A member of our team reads, edits and sends every reply that goes to you, unless a message is marked as automatic.
We do not use your messages to train AI models, and we do not share them with anyone beyond the providers that process them for us. [Add the details that are true for you.]
If you would rather not have AI involved in handling your messages, tell us at [email protected] and we will handle your conversation by hand.
Only keep sentences that are true for your company. If you do not know whether your provider uses messages for training, find out before you write that line.
For an automatic acknowledgement
This is an automatic message to confirm we received your request. A member of our team will reply personally, usually within one working day.
For a chat or conversational assistant
Hi, I am an automated assistant. I can answer common questions, and I will pass you to a person whenever you ask or when I cannot help.
When someone asks directly
Good question. We use AI to help organize our inbox and to suggest draft replies, and a person on our team reads and sends every reply. I wrote this message myself. If you would like to know more, our page "How we use AI in support" has the details.
If it is not true that you wrote it yourself, do not say it. Use the version that matches reality, such as "This reply was drafted with AI and I checked it before sending."
Things not to do
- Do not pretend an AI draft was handwritten. Signing a draft with a human's name is fine if that person reviewed and takes responsibility for it. It is not fine if nobody read it.
- Do not hide it in a footer in tiny grey type and then claim you disclosed it.
- Do not overstate. "Powered by advanced AI" invites questions you may not want to answer and says nothing useful.
- Do not over-disclaim. A warning on every message teaches people to ignore it.
- Do not make AI the scapegoat. "The AI made a mistake" does not help the customer. The company sent the message, and the company is responsible. See apology emails that rebuild trust for how to own a mistake.
- Do not let staff improvise. If everyone answers "are you a bot?" differently, someone will get it wrong.
Make the human review real
A disclosure that says "a person reads every reply" is only as honest as the practice behind it. Make it true:
- Require an explicit send action by a named person for every drafted reply.
- Spot-check sent messages against what the AI drafted, in a regular review. Reviewing AI-drafted replies: a checklist is a good starting point.
- Make sure ownership is clear in a shared inbox, as in ownership of AI-drafted replies.
- Keep a style guide so the replies sound like your company, as in a style guide for AI-drafted email.
If you cannot honestly say a person reviews each reply, change the practice or change the sentence.
Handling opt-outs and questions
Decide in advance what happens when a customer says they do not want AI involved. The simplest promise is that their conversation will be handled by a person without AI-drafted replies or summaries. Record the preference on the contact, make sure everyone who might reply can see it and make sure the tool respects it. A promise nobody can see in the inbox will be broken within a week.
Prepare short answers to the questions you will get: what the AI does, which provider processes the data, whether messages are used to train models, how long data is kept and how to delete it. Human-in-the-loop email automation patterns shows how teams keep a person in charge as they add automation.
A one-page checklist
- We have listed every way AI touches customer email, by level
- Our privacy notice describes it truthfully
- A short help page explains it in plain words
- Automatic and conversational messages say they are automatic
- Nobody signs a message they did not review
- Staff have an agreed answer to "are you a bot?"
- We can honor a "no AI on my conversation" request and staff can see it
- We have checked what our provider does with message content
- A lawyer or privacy lead has read the wording
Key takeaways
- Disclose in proportion to how much AI is involved, and never deny it when asked.
- Behind-the-scenes use belongs in the privacy notice. Automatic replies and conversational assistants must say so at the point of contact.
- Use plain, true sentences. Do not claim a person wrote or reviewed what they did not.
- Make human review real, with a named sender, spot checks and clear ownership.
- Give people a way to opt out and make sure the team can see and respect it.
Start with Koltrix
Your domain, one inbox, and an API that sends.
A team inbox where AI sorts and drafts (nothing is sent without your click), plus the transactional API and SMTP relay your product sends with. 7 days free, no card.
