Once a team starts using AI for emails, quotes, listing copy and replies, the drafts come quickly. That speed is the whole point. The catch is that AI writes with the same confident tone whether it is right or wrong, and a busy person skimming a draft at 4.45pm can easily send something that should never have left the building.
The fix is not to stop using AI. It is to agree on a short, repeatable check before anything AI-written reaches a customer.
What AI tends to get wrong
It helps to know where the problems usually hide. In our experience, the common slips are:
- Made-up details. A price, a date, an address or a policy that sounds plausible but is not yours.
- Promises you did not make. "We will have this fixed by Friday" when nobody agreed to Friday.
- Wrong tone. Too formal for a long-standing client, too casual for a complaint, or oddly American for an Aussie business.
- Missing context. The draft answers the question asked, but ignores the history in the thread.
- Private information. Details from one customer turning up in a message to another, usually because too much was pasted into the prompt.
A five-point check before you hit send
Keep this short enough that people actually do it. Before sending any AI-assisted message, the person sending it reads it once with these questions in mind:
- Are the facts ours? Check every number, date, name, price and policy against the real source, not memory.
- Did we promise anything? Look for commitments on timing, refunds, discounts or outcomes. If nobody agreed to it, take it out.
- Does it sound like us? Read the first and last lines out loud. If it would make a regular customer raise an eyebrow, rewrite them.
- Is anything private in there? Make sure only this customer's details are included, and nothing internal like staff notes or margins.
- Would I sign my name to it? If the honest answer is "mostly", fix the part that is not.
Most drafts pass in under a minute. The ones that do not are exactly the ones you wanted to catch.
Match the check to the risk
Not every message needs the same attention. A friendly reply confirming an appointment is low risk. A quote, a response to a complaint, anything about money owed, or anything touching a contract, lease or legal matter deserves a slower read, and sometimes a second person.
A simple rule many small teams find useful: if it includes a price, a deadline or a complaint, a second set of eyes looks at it before it goes. Everything else, the sender checks it themselves.
Make the AI easier to check
You can also reduce the checking load by setting the tool up well in the first place:
- Give it your real material. Price lists, policies and templates mean fewer invented details.
- Ask it to flag gaps. Tell it to write "CHECK" wherever it is unsure, rather than guessing.
- Keep prompts lean. Paste only what is needed for this message, not a whole inbox.
- Use your own examples. A few past emails you were happy with help it match your voice.
If your team does not have ground rules yet, our one-page AI policy for small teams is a good place to write this check down, and the AI and customer data privacy checklist covers what should never go into a prompt.
Watch for check fatigue
The biggest risk is not a bad draft. It is the slow slide from reading carefully to clicking send on autopilot, because the last fifty drafts were fine. Every so often, pick a few sent AI-assisted messages at random and review them as a team. It keeps everyone honest and shows you where the tool, or the prompt, needs work.
Our honest take
AI drafting saves real time, but only if a person stays responsible for what goes out. A one-minute check is a small price for keeping customer trust, and it quickly becomes habit.
At Cloud Inn we help small businesses and real estate teams across South East Queensland set up AI so it is fast and safe to use. If you would like help building a simple review step into the way your team already works, book a 30-minute call and we can map it out together.
