Product Updates
Inside SupportMe's Draft Rejection Flow
What SupportMe's documentation says about rejecting AI-drafted support replies, which details aren't public yet, and a practical way for small teams to decide between editing a draft and discarding it.
SupportMe lets you reject any AI-drafted reply before it reaches a customer. Nothing sends unless you approve it. Rejecting a draft is one of three choices you get at review time. The other two are sending it as written or editing it first. What you won't find in the public documentation is what happens inside SupportMe after you reject one.
This post covers what's documented, what isn't yet, and a simple way to decide when to throw a draft out instead of fixing it.
What SupportMe documents about rejecting drafts
SupportMe is pre-launch, in its waitlist phase. Its public product information confirms these points:
- Every draft needs your approval. SupportMe drafts replies but never sends them on its own. The product describes itself as "human-in-the-loop by design."
- Reject is one of three options. You can review, edit, or reject any draft before it goes out.
- Learning comes from edits. When you edit a draft and send it, SupportMe runs a diff analysis between its draft and your final version. It then updates your writing style profile and knowledge base.
The documentation doesn't currently say:
- whether a rejected draft gets logged or stored,
- whether a rejection on its own teaches the system anything,
- whether you can give a reason for a rejection,
- whether you can ask for a new draft after rejecting one.
Treat those as open questions, and expect details to change before launch. For the full review step that rejection belongs to, see how SupportMe's human-in-the-loop approval flow works.
Why the ability to reject matters
An approval step only protects you if you're willing to say no. Two well-documented cases show what can happen when AI-generated support answers reach customers without a check.
Air Canada's bereavement fare chatbot. In Moffatt v. Air Canada (2024 BCCRT 149), the airline's website chatbot told a customer he could apply for a bereavement fare within 90 days of booking. The actual policy required the request before the flight. British Columbia's Civil Resolution Tribunal found Air Canada liable for negligent misrepresentation. It rejected the argument that the chatbot was responsible for its own statements and wrote that "it is still just a part of Air Canada's website" (full decision on CanLII; American Bar Association summary).
Cursor's support bot. In April 2025, a front-line AI support bot at the code editor company Cursor told a user that logging in on multiple devices wasn't allowed. No such policy existed. Users posted cancellation threats on Reddit and Hacker News, and the company apologized (eWeek; AI Incident Database entry/)).
In both cases the AI answered customers directly. A tool that drafts but doesn't send gives you a point where you can stop an answer like that. But it only helps if you read drafts closely enough to catch the problem.
NIST makes a related point. Its Generative AI Profile (NIST AI 600-1) lists "confabulation" (confidently stated false content) and "human-AI configuration" risks such as automation bias and overreliance among the main risks of generative AI. If you approve drafts out of habit, the review step stops working. Rejecting drafts when they deserve it keeps it working.
When to reject a draft and when to edit it
This section is our recommendation, not documented SupportMe behavior.
A useful rule: edit when the draft is mostly right, and reject when the basis of the answer is wrong.
Edit when:
- the facts are right but the tone is off,
- a step is missing or the details are slightly wrong,
- the greeting, sign-off, or length doesn't sound like you,
- the answer is right but needs a link or a version number.
Reject when:
- the draft states a policy, price, refund rule, or deadline you can't confirm,
- it answers a different question from the one the customer asked,
- it promises something you haven't decided, like a fix date, a feature, or a refund,
- the message needs a personal reply, such as an angry customer, a security report, a legal or billing dispute, or a bug you haven't reproduced yet,
- fixing it would take longer than writing your own reply.
There's another reason to keep edits for drafts that are close. SupportMe learns from the difference between its draft and what you send. A light edit to a good draft is a clear signal. Rewriting a bad draft until nothing of it is left may be a less useful signal, though SupportMe hasn't published how it handles heavy rewrites. To see how edits get interpreted, read how SupportMe sorts your edits into style and knowledge.
A hypothetical example
This scenario is invented for illustration.
A customer emails a solo developer: "I was charged twice this month. Can you refund one?" The draft apologizes and says a refund will arrive in 3–5 business days. The developer hasn't checked the payment dashboard yet and doesn't know whether the double charge is real or how long refunds take with their payment provider.
The tone is fine, but the draft makes a promise the developer can't back up. Rejecting it and writing a short holding reply ("Thanks for flagging this. I'm checking the charges now and will get back to you today") is safer than tweaking a draft that already commits to a timeline.
What to do after you reject a draft
SupportMe hasn't published what happens next inside the product, so these steps cover your own process:
- Reply yourself, even briefly. Rejecting the draft doesn't answer the customer. A short, accurate reply beats a long, uncertain one.
- Find out why the draft went wrong. Was the right information missing? Is your documentation out of date? Or was the question just unusual? SupportMe builds its knowledge from real replies, so a clear, correct answer from you is the kind of material it's designed to learn from. See how SupportMe builds its knowledge base from replies.
- Watch for repeats. If you reject drafts on the same topic again and again, it's usually a knowledge gap, not a one-off mistake. Write down the correct answer once in your own notes so your replies stay consistent.
- Be extra careful with policy topics. Refunds, pricing, data deletion, and account limits are where both the Air Canada and Cursor problems happened. Read drafts on these topics closely.
How rejection fits the rest of the drafting flow
Rejection sits at the end of a longer pipeline. A message arrives in a connected channel, SupportMe drafts a reply in your style using your knowledge base, and you decide what to do with it. For the steps before review, see inside SupportMe's email reply drafting flow.
Conclusion
According to SupportMe's documentation, you can reject any draft, nothing sends without your approval, and the system learns from the differences between its drafts and your edited replies. It's not yet public whether a rejection teaches the system anything or what happens to a rejected draft. In practice, edit drafts that are close and reject drafts built on facts you can't confirm, especially about policies, money, or promises. Both the Air Canada and Cursor cases show the risk is real, and an approval step only works if you're willing to use the reject option.
References
- Civil Resolution Tribunal of British Columbia, Moffatt v. Air Canada, 2024 BCCRT 149
- American Bar Association, BC Tribunal Confirms Companies Remain Liable for Information Provided by AI Chatbot
- eWeek, Cursor AI chatbot hallucination and fake policy
- AI Incident Database, Cursor AI support bot ("Sam")/)
- NIST, AI 600-1: Generative Artificial Intelligence Profile
Tags
Related posts
Product Updates
How SupportMe Drafts Replies Before It Knows Your Voice
What to expect from SupportMe's first drafts, before it has seen your edits. Why early replies sound more neutral, what the product docs actually say, and how your first reviews shape later drafts.
6 min read
Product Updates
How SupportMe Keeps Your Support Inbox Data Private
What SupportMe has committed to on support inbox privacy, how its learning from your edits works, what's still undocumented before launch, and questions to ask any AI support tool.
7 min read
Product Updates
Inside SupportMe’s Email Reply Drafting Flow
A step-by-step look at how SupportMe drafts email replies in your voice, how you review them before anything sends, and how your edits are used to improve later drafts.
6 min read