Straight answer: AI hasn't made your emails worse because spam filters can "detect" AI. It's made them worse because thousands of small businesses are now writing the exact same sentences, with the exact same rhythm, and inboxes have got very good at spotting sameness. The fix isn't hiding the AI, it's breaking the pattern.
The email that started this whole rabbit hole
A client of mine, a small skincare brand in Manchester, handed her weekly newsletter over to a virtual assistant in September last year. Sensible move on paper, she was drowning. The VA started using ChatGPT to draft every email: same structure, same opener ("Hope this finds you well!"), same three-bullet benefit list, same "Click here to shop the collection" close.
By January, open rates had gone from 34% to 11%. Click rate had gone from 4.2% to under 0.6%. Nothing else had changed. Same list, same send time, same product range. When she called me in a slight panic, my first question wasn't "is this landing in spam", it was "read me the last five subject lines out loud." She did. They all started with a verb. They all had an emoji in the same position. They were all, frankly, interchangeable with any of the forty other beauty brand emails sitting in the same inbox.
That's the bit nobody wants to say out loud: it's not that AI writes badly. It's that thousands of small businesses are all using the same three tools with the same default prompts, and Gmail, Outlook and Yahoo have got very sharp at pattern-matching that sameness against engagement data. Low engagement plus generic phrasing plus predictable structure equals a slow slide into the Promotions tab, then eventually spam, and once you're there it takes weeks to climb out.
What's happening inside the inbox providers
Gmail, Outlook and Yahoo don't have a magic "written by AI" flag. What they have is engagement tracking at a scale you can't out-think: opens, replies, deletes-without-opening, "mark as spam" clicks, time spent reading, forwarding. When a template becomes common enough that thousands of senders use near-identical phrasing, the inbox starts treating engagement drops on those patterns as a signal, not a coincidence.
- If your subject line follows the same formula as fifty other emails a person gets that week, and most of those get ignored, your email inherits some of that reputation by association.
- If your body copy uses the exact stock phrases ChatGPT defaults to ("in today's fast-paced world", "we're thrilled to announce", "don't miss out"), you're not standing out, you're blending into a pile that inbox providers already know performs badly.
- If your click-through rate drops because the copy feels generic, your sender reputation drops with it, and that follows you into your next send, and the one after that.
This is the uncomfortable bit most marketing advice skips: it was never really about AI detection. It's about the fact that AI, used lazily, makes everyone sound the same at exactly the moment inbox algorithms are optimising harder than ever for signals of genuine reader interest.
The numbers that made this real for me
When I audited the Manchester client's list, three things stood out:
- Her domain's spam complaint rate had crept from 0.02% to 0.19% over four months. Anything above 0.1% is where most email providers start throttling delivery.
- Her list hadn't been pruned in fourteen months. Roughly 22% of it hadn't opened a single email in that window, which meant every send was dragging the average engagement down before the content even mattered.
- Every single subject line in the previous six weeks used a title case format with an exclamation mark. Every one.
None of that shows up if you just glance at "did AI write this." It shows up when you look at patterns across sends. That's the level most small business owners never check because they're too busy running the actual business.
The fix that worked (and it wasn't "stop using AI")
We didn't ditch AI. We changed how it was used, in five specific steps.
- Cut the dead weight first. Removed anyone who hadn't opened an email in twelve months. List went from 8,400 to 6,550. Smaller list, higher average engagement, which is the signal inbox providers reward.
- Made AI write a rough first pass, then a human rewrote every subject line. No formula repeated twice in a row. Sometimes a question, sometimes a fragment, sometimes just a name.
- Banned six stock phrases outright. "In today's world", "we're excited to", "don't miss out", "elevate your", "game-changing", "unlock your" were all struck from the prompt and from the final copy, every time.
- Sent test emails to a seed list across Gmail, Outlook and Yahoo before every send to check placement, not just spelling.
- Varied sentence length on purpose. AI defaults to a steady, slightly formal rhythm. Short sentence. Then a longer one that explains the reasoning. Then short again. That variation is what makes copy read like a person, not a template.
Six weeks later: open rate back up to 27%, click rate to 3.1%. Not back to the original 34%, some of that list damage takes longer to repair, but heading the right way, and the spam complaint rate had dropped back to 0.04%.
The prompt problem nobody flags
Here's the thing most "how to write emails with AI" posts don't tell you: the default prompt most people type in is the problem. "Write a promotional email about our new skincare range" produces the same skeleton every single time, for every single business using that tool. You're not getting a unique voice, you're getting the median output of every training example that shape of prompt has ever produced.
What works instead is giving the tool constraints that force it away from the median: a specific customer complaint to respond to, a real number from your own sales data, a genuine limitation of the product, an actual sentence a customer said to you last week. AI is much better at rearranging specific, weird, true detail than it is at inventing generic enthusiasm. Feed it the boring true fact and it stops sounding like every other AI-written email in the inbox.
This is roughly the same lesson Joe Pulizzi built an entire content marketing career on: specificity and a genuine point of view beat polish every time. AI hasn't changed that rule, it's just made it more obvious who's ignoring it.
Where the same problem shows up beyond email
Once you see this pattern in newsletters, you start spotting it everywhere. Social captions written the same way for every brand using the same tool. Landing page headlines that all promise to "transform" something. Even the interactive tools businesses use to engage visitors run into the same trap when the copy around them is generic, which is one reason interactive calculators work so well as an engagement tactic, the value comes from the specific number a visitor gets back, not from the words around it.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Look at brands that have kept email and content working hard for years. Mailchimp built its entire growth story on making small businesses feel like they were being spoken to directly, not marketed at. That's the opposite instinct to what a lazy AI prompt produces. Same with how Instagram grew engagement by rewarding specific, personal content over generic brand messaging. The platforms have moved on. A lot of small business copy hasn't caught up.
What I'd check in your account this week
- Pull your last ten subject lines. If more than six follow the same structure (same punctuation position, same opening word type), that's your first fix, free, takes ten minutes.
- Check your spam complaint rate in your email platform's analytics. Above 0.1% means stop and fix before you send anything else to that full list.
- Search your last month of emails for the words "unlock", "elevate", "", "game-changing" and "don't miss out." Delete every instance. If your AI tool keeps producing them, that's the prompt's fault, not the tool's, rewrite the prompt with a real detail instead of a general instruction.
- Prune anyone who hasn't opened an email in nine to twelve months, or at minimum move them to a separate, less frequent list.
- Read one email out loud before you send it. If it sounds like it could have your competitor's logo on it instead of yours, it's too generic to send.
None of this is complicated. Most of it takes under an hour a week. But it means treating the AI draft as a starting point, not a finished product, which is the bit that gets skipped when a business owner is exhausted and just wants the email out the door.
The uncomfortable part
The uncomfortable truth in all of this is that AI didn't create a new problem, it exposed one that was already there. Generic marketing copy has always underperformed. It just used to take longer to produce, so there was less of it, and inboxes were less crowded with it. Now anyone can generate a competent-sounding email in ninety seconds, so the market is flooded with competent-sounding, forgettable emails, and inbox providers have had to get better at spotting the pattern because their users were getting fed up. You're not fighting an algorithm. You're fighting every other small business that stopped thinking before they hit send.
The businesses whose open rates are climbing right now aren't the ones that banned AI. They're the ones using it for structure and speed, then putting a real person's judgement, real customer detail, and a varied voice back into the final draft before it goes out.
Free resource: The Re-Engagement Email Template.
Frequently asked questions
Can spam filters detect AI-written text?
Not directly, no. There's no filter checking "was this written by ChatGPT." What filters do detect is low engagement, generic phrasing patterns shared across many senders, and complaint rates, all of which tend to rise when AI copy is used without editing, because that copy converges on the same predictable structure everyone else is producing.
How quickly can I recover my open rate after a drop like this?
In the case above, six weeks brought open rate from 11% back to 27%, still short of the original 34%. List pruning gives the fastest visible lift, usually within one or two sends, because you're removing dead weight from your average. Full sender reputation repair with providers like Gmail can take six to eight weeks of consistently good engagement.
Is it worth using AI for email marketing at all, then?
Yes, for structure, speed and getting past the blank page. The mistake is sending the first draft unedited. Use AI to build the skeleton, then insert a real detail, a real number, or a real customer quote, and rewrite the subject line by hand every time. That combination is faster than writing from scratch and avoids sounding like every other AI-drafted email in the inbox.
What's the single fastest fix if my open rates just dropped?
Check your spam complaint rate first, above 0.1% means stop sending to your full list until it's fixed. Then prune anyone who hasn't opened an email in nine to twelve months. Those two steps alone usually account for most of a sudden drop, before you even touch the copy itself.
Related reading: Why Your Analytics Are Hiding How Many Customers AI Is Sending You and AI Meeting Notetakers Are Quietly Changing What Your Clients Tell You.
