A message lands in a customer service inbox. It's long, polite and neatly laid out, the way AI tends to write.
And the last line reads: "Would you like me to rephrase this further?"
The customer had copied the whole thing out of ChatGPT and pasted it in, sign-off and all.
Vik Lakhotia leads the team behind Marshmallow's AI customer agent. At aiblLIVE in London on 20 October, he told the room his team sees exactly this. "A lot of our customers are using ChatGPT, they're using Claude, and then they send us lots of messages," he said. "Our message inbound has exploded."
It's funny right up until you realise it's coming for every business with an inbox.
The side of AI nobody planned for
Most of the day's talks were about what businesses can do with AI. This was a moment about what customers are doing with it.
When writing a long, detailed message costs nothing, people write more of them. When AI can draft a formal request in seconds, more people make formal requests. Vik mentioned a rise in data subject access requests, the formal "send me everything you hold about me" letters, alongside the general flood.
His conclusion was blunt. "It's not possible to defend yourself from it."
So Marshmallow stopped trying. Instead of hoping the volume would go away, they looked at where the problems behind the messages were coming from and fixed those. That was one of the reasons their compliance team, wary at first, came round to using AI on their side too. A human team couldn't keep up with machine-written inbound. Something had to meet it.
Why a small business should care
You don't need to be an insurer with hundreds of thousands of customers to feel this.
Think about what's already landing in a typical small business inbox. Enquiries that read like a tender document. Reviews that are suspiciously well written. Job applications that all sound the same. Complaints that cite consumer law chapter and verse.
None of that is bad, exactly. It's a customer using the tools they have. But it changes what good service looks like. The temptation is to fire an AI reply back. Machine writes to machine, and nobody is any the wiser.
The better answer, I think, is the human one. When a message is long and formal and a bit cold, a short, warm, plain reply from a real person stands out more than ever. "Thanks for this. I've read it properly. Here's what I'm going to do and by when."
The flood of AI-written messages can be a chance to show people there's someone real behind your business.
Your customers use whatever tools they like. You use AI where it helps, to sort, summarise and spot patterns. And every reply still sounds like you.
Your last ten customer emails are a good place to look, along with your replies to them. How many of those replies sound like a person wrote them?
