Is it dishonest to send an AI-written email?

A customer asks whether our valve is suitable for a 90°C glycol loop. I know the answer. I also know that if I open a chat window and describe the product, the model will write a better email than I will at four in the afternoon on a Friday.

Is that dishonest?

I do not think the tool is the problem. The lie, when there is one, is in the sentence "we have tested this at 90°C for 2,000 hours" — if we have not. The model will produce that sentence happily if the prompt is vague enough. That is on me, not on the model.

Three rules, all about claims

Never assert a fact you did not verify. A spec, a certification, a test result, a lead time. If the email contains one, it came from a document I read, not from a model's memory.

Never let the tool invent a person. No "our engineer Klaus reviewed your drawing" unless Klaus exists and actually looked at it. This happens more than people admit, because a named colleague reads as credibility.

Sign your own name only if you read every line. Sounds obvious. It stops being obvious when you are sending forty a day and the fourth one looks like the first.

Where it gets murky

Not text. The things that look like evidence. A product photo generated rather than shot. A voice note cloned from a real salesperson. A "factory tour" video assembled from stock footage. Text can be skimmed and judged; an image reads as proof. I would hold those to a much stricter standard than an email, and I would say so on the page where they appear.

The commercial argument, which is separate

Buyers are getting good at spotting the pattern — the three-paragraph structure, the "I hope this email finds you well", the paragraph that summarises their own inquiry back at them. The tell is not that AI wrote it. The tell is that nobody added anything a human would know. A specific detail about their port, their season, their last order. That is the part no model can supply, and it is the part that gets replies.

So the test I use before sending: if the customer learned exactly how this email was made, would they feel tricked? If yes, rewrite. If they would shrug — because the content is true, specific and about them — send it.

That is a lower bar than "never use AI" and a higher one than "it is just a tool". It is also the only version I can defend in a phone call.

Why there is no ad here. This page carries no ads and no sponsored links, because an answer about honesty should not be for sale. If that is worth something to you, you can support the site here — and if it is not, the argument still stands.

We publish the pricing model, minimum commitment and export-readiness notes for every tool in the library, and we flag it when a vendor changes one without changing the headline price.