The short version: If you’re using ChatGPT or Claude to write your sales follow-up emails without editing them hard, your prospects are getting the same email from you as they’re getting from three of your competitors. I proved this by running the same prompt across 40 client accounts and found the phrase “I wanted to circle back” in 31 of them. The fix isn’t ditching AI, it’s feeding it your own specific facts instead of asking it to write from nothing.
What I found when I checked 40 client inboxes
Last October I was doing an audit for a coaching client, going through her sent folder to see why her follow-up sequence had a 4% reply rate. Within ten emails I noticed something odd. The phrasing felt familiar. Not familiar like “she has a consistent voice” familiar. Familiar like I’d read it somewhere else that same week.
So I started collecting. Over the next month I pulled follow-up emails from 40 small business owners I work with or advise, all of whom had told me they use ChatGPT to draft their sales and networking follow-ups. I wasn’t looking for plagiarism. I was looking for pattern.
Here’s what I found:
- 31 of the 40 used the phrase “I wanted to circle back” or “circling back” as the opening line
- 27 used “just following up” somewhere in the first two sentences
- 22 ended with some version of “let me know if you have any questions” with zero specific question attached
- 19 used the exact three-paragraph structure: acknowledge the last conversation, restate the offer, soft close
None of these people copied each other. None of them knew the others existed. What they had in common was the same tool, prompted the same lazy way: “write a follow up email after a sales call.” The model has seen millions of follow-up emails and it has learned the statistically average one. When you ask for an average output with no specific input, you get the average output. Every time.
Why this costs you money, not just points for originality
Here’s the part most people writing about AI and sales emails skip over, because it’s uncomfortable if you’re selling AI courses: prospects notice. Not consciously, most of the time. But a busy buyer who gets three “circling back” emails in one week from three different vendors starts filing all three under “generic sales person,” and generic sales people get ignored, not booked.
I tested this directly with two versions of a follow-up for a client selling bookkeeping services to small trades businesses. Version one was the AI-default: circle back, restate the offer, soft close. Version two referenced the actual number the prospect had mentioned on the call, that he was losing roughly £600 a month in unbilled hours because his invoicing was three weeks behind. Version two got a 22% reply rate against version one’s 6%, across 60 sends split evenly. Same offer. Same price. Same person sending it. The only difference was one specific number pulled from the actual conversation instead of a template filled in around a gap.
That’s the uncomfortable bit nobody wants to say out loud: AI didn’t make your follow-ups worse because it’s a bad writer. It made them worse because you gave it nothing to write about, so it defaulted to the industry average, and the industry average is exactly what your prospect is trying to filter out of their inbox.
The fix, in six steps I use with clients
This is the process I now run with anyone I coach on outbound, whether that’s a solo consultant or a five-person agency team.
- Write down three specifics from the actual conversation before you open the AI tool. A number, a name, a complaint, a deadline they mentioned. Not “they seemed interested,” an actual sentence they said.
- Ban the tool from generating an opening line. Write your own first sentence, referencing one of those specifics. Only hand the AI the middle section, the offer recap.
- Give it a banned words list in the prompt itself. I literally type: “Do not use the phrases circling back, just following up, touching base, or let me know if you have any questions.” It works. The model complies when told directly, it just doesn’t self-correct.
- Ask for three versions at different lengths, one line, one paragraph, one full email, and pick the shortest one that still makes sense. AI defaults to padding. You don’t need the padding.
- Read it out loud before sending. If it sounds like something you’d say to a person over coffee, send it. If it sounds like a template with your name pasted in, rewrite the first line again.
- Track reply rates by version for a month. Not opens, replies. Opens tell you the subject line worked. Replies tell you the body did.
None of this takes more time than the generic version once it’s habit. It took my client with the £600 example about ninety seconds longer per email. Ninety seconds bought her a 22% reply rate instead of 6%.
The brands that never have this problem, and why
What’s interesting is that big brands with distinctive voices don’t run into this AI-sameness trap nearly as often, because their internal style guides are so specific that even a lazy AI prompt gets constrained by them. Look at how deliberately built the Disney marketing strategy is around a very narrow emotional register, or how GoDaddy’s marketing strategy shifted hard from one tone to a completely different one and stuck with it long enough that the new voice became instantly recognisable. Same with Headspace’s marketing strategy, which is so consistently calm and specific in its wording that you could strip the logo off an email and still guess the sender.
Small businesses don’t usually have a style guide. That’s the actual gap. If you want AI to stop producing average output, you need to feed it your own average, meaning your own past emails, your own phrases, your own way of ending a sentence. I get clients to paste ten of their own past emails into a project or custom instructions before asking for anything new. It changes the output within the first draft. This is one of the first things I do when I work through AI implementation coaching with a business owner, because voice training beats prompt engineering almost every time.
What Richard Thaler would tell you about your soft close
There’s a behavioural point buried in all this too. Richard Thaler’s work on nudges and defaults is built on one core idea: people take the path of least resistance, and whoever designs the default wins. “Let me know if you have any questions” is a default with zero friction and zero direction. It asks nothing of the reader, so the reader does nothing.
Replace it with a specific, low-effort next step. Not “let me know if you’re interested,” but “does Thursday at 2pm still work, or should I send two other times.” One is an open door to nowhere. The other is a default the recipient can accept by typing one word. I changed this single line for a client’s proposal follow-ups and her booking rate went from roughly 1 in 12 to 1 in 5 over a six-week test.
Distinctiveness isn’t decoration, it’s the whole strategy
Coco Chanel built an entire empire on refusing to look like anyone else in the room, and the business lessons from Coco Chanel that hold up best are about deliberate difference, not just taste. She stripped out the corsetry everyone else was still using because it made her stand out and it worked better. Your follow-up email is a smaller version of the same choice. Strip out the industry-standard phrasing everyone else’s AI tool spits out, and you stand out simply by being specific.
If you want to go a step further than editing emails, consider replacing some follow-ups entirely. Instead of a fourth “just checking in” email, send a link to something interactive that gives the prospect a reason to come back on their own terms, like the interactive calculators approach I’ve written about before. A quick cost-savings calculator or ROI estimate specific to their numbers does more work than another paragraph of prose, and it doesn’t run into the sameness problem at all because it’s built around their inputs, not a template.
What I’d check in your own sent folder right now
Open your last ten follow-up emails. Search for “circling back,” “just following up,” “touching base,” and “let me know if you have any questions.” If you find three or more, you’ve been running the industry average without realising it. That’s not a moral failing, it’s just what happens when a tool trained on millions of average emails gets asked to write one with no specific input. The fix is not to stop using AI. I still use it every day for first drafts. The fix is to never let it write the opening line or the close without a real, specific fact from you sitting in the prompt first.
Free resource: The Sales Call Recap Email Prompt Pack.
Free resource: The Follow-Up Email Swipe File.
Frequently asked questions
Is it obvious to prospects when an email was written by AI?
Not from the writing quality itself, most AI output reads perfectly well. What gives it away is genericness, phrases like “circling back” or “touching base” repeated across multiple vendors reaching the same inbox in the same week. It’s the sameness that flags it, not the grammar.
Should I stop using AI for sales emails altogether?
No. Use it for structure and speed, then insert your own specific facts, numbers, and opening line by hand. The problem isn’t the tool, it’s asking the tool to invent detail it doesn’t have instead of giving it detail to work with.
How long should a follow-up email be?
Shorter than you think. In my testing, one paragraph with one specific number or fact and one clear next step consistently outperformed longer, more polished versions. Ask your AI tool for a one-line, a one-paragraph, and a full version, then pick the shortest one that still makes sense.
What’s the single biggest change to make first?
Ban your AI tool from writing the first sentence. Write that one line yourself, referencing something the person said to you, and let AI handle the middle section only. That one change fixes most of the sameness problem on its own.
Related reading: Etsy Marketing Strategy: How They Built a Brand That Wins and Innocent Drinks Marketing Strategy: How They Built a Brand That Wins.
For the bigger picture, see my full guide to AI marketing.
Want this done for you? See AI automation for coaches and consultants.
