Straight answer: AI text doesn’t send your emails to spam. Bad authentication, dead weight on your list, and low engagement do. I found this out the hard way when my own open rate fell from 38% to 11% in six weeks, and the fix had nothing to do with rewriting a single sentence.
The story everyone’s telling you, and why it’s wrong
There’s a rumour going round small business Facebook groups and half the “AI marketing tips” newsletters: Gmail and Outlook can smell AI-generated text and they punish it. I believed this too, briefly, in about October last year, when my email numbers fell off a cliff and I’d just started batch-writing my newsletter with ChatGPT to save time.
It felt like a neat explanation. It’s also not what happened, and I can prove it, because I fixed the real problem and the AI-drafted emails started performing exactly like the ones I’d written by hand for a decade.
Spam filters at Google, Microsoft and Yahoo don’t run your copy through an “is this AI” detector. They can’t reliably tell anyway, nobody can, not even the tools that claim to. What they score is behaviour: how many people mark you as spam, how many never open you, whether your domain is authenticated, and whether you make it easy to leave. That’s it. That’s the whole game.
My own crash, in numbers
Here’s exactly what happened, because vague warnings help nobody.
- Late September: weekly newsletter to just under 9,000 subscribers, open rate steady at 36 to 40%, sent through the same platform I’d used for years.
- Early October: I moved to sending two emails a week instead of one, using AI to draft the second one faster so I could keep up.
- By mid-November: open rate down to 11%, and worse, three separate subscribers replied asking why my emails had gone to their spam folder.
- Spam complaint rate (visible in the platform’s own reporting) had crept from 0.02% to 0.31%, which sounds tiny until you know that anything over 0.3% is the point Gmail and Yahoo start treating your whole domain with suspicion, not just that one email.
The AI drafting wasn’t the cause. The extra volume was. I was mailing people twice a week who’d signed up expecting once, a chunk of them hadn’t opened anything from me in over a year, and I hadn’t touched my domain authentication since I set it up. Three separate, boring, entirely fixable problems, none of which had anything to do with a robot writing the words.
The technical bit nobody explains in plain English
If you’re sending email for a business, you need three records set up on your domain, and if you’ve never heard of them, that’s very likely part of your problem right now, AI-assisted copy or not.
- SPF tells receiving servers which mail services are allowed to send on behalf of your domain.
- DKIM attaches a digital signature to your emails so receivers can confirm they weren’t tampered with in transit.
- DMARC tells Gmail and Yahoo what to do if a message fails those checks, and gives you a report so you can see it happening.
Since February 2024, Google and Yahoo require all three for anyone sending more than 5,000 emails a day to their users, and honestly the smaller senders should treat it as required too, because the trust scoring applies regardless of volume. If you’re using a platform like Mailchimp, most of this gets set up for you when you connect your domain, which is one of the quiet reasons a proper Mailchimp marketing strategy still beats sending from a personal Gmail account with a spreadsheet of addresses. But “most of this” isn’t “all of this.” I found my DKIM record had a typo from a domain migration two years earlier. Nobody had noticed because nobody checks these things until something breaks.
If untangling DNS records isn’t something you want to do yourself, it’s a small, cheap job for the right person, and it’s exactly the sort of thing an AI implementation coach can sort in an afternoon rather than you losing a weekend to Stack Overflow threads written for developers, not marketers.
The list hygiene problem, which is duller and matters more
Here’s the uncomfortable bit most email advice skips over because it means admitting your list isn’t as big or as good as you’d like: a list full of people who never open your emails is actively harming the emails that go to people who do.
Inbox providers watch engagement per sender, not per email. If 60% of your list hasn’t opened anything in eight months, every single send teaches Gmail that you’re someone worth ignoring, and that penalty then lands on the 40% who want to hear from you.
What fixed my numbers faster than anything technical was brutal and took twenty minutes:
- I pulled everyone who hadn’t opened an email in 90 days.
- I sent that group one re-engagement email with a clear subject line: “Should I stop emailing you?”
- About 4% clicked to stay. The other 96%, roughly 3,100 people, I removed from the list entirely.
My list shrank by a third overnight. My open rate, measured against the smaller, more honest list, jumped to 42% within two sends, higher than it had ever been before the crash. Fewer people, better numbers, more actual buyers. This is the trade nobody wants to make because a smaller list feels like admitting failure, but a list full of ghosts isn’t an asset, it’s a liability sitting in your account quietly wrecking your deliverability.
So where does AI cause problems
Not never, to be fair. There are two genuine ways AI drafting can hurt you, and neither is the “detection” myth.
First, sameness. If every email reads like a slightly different flavour of the same generic AI voice, tone, structure, three bullet points, a call to action, repeat, people stop opening because there’s nothing distinct to open for. That’s an engagement problem that then becomes a deliverability problem, the mechanism I described above, just triggered by boring writing rather than by the tool that produced it.
Second, volume without judgment. AI makes it easy to write five emails in the time it used to take to write one, and the temptation is to send all five. More emails to people who didn’t ask for more emails is the single fastest way to push your complaint rate over that 0.3% line.
The fix for both is the same: use AI to draft faster, then spend the time you saved on the parts that build a relationship, a specific story, a real number, an opinion someone could disagree with. Think about how Airbnb built trust into every touchpoint of their marketing by making things feel personal and specific rather than templated. Your newsletter needs the same instinct. Generic is what gets ignored, and ignored is what gets you flagged.
What to do this week
In order, because order matters here:
- Check your SPF, DKIM and DMARC records exist and are correct. Most email platforms have a one-click checker in settings, or you can use a free online DMARC lookup tool.
- Pull your list and sort by last-open date. Anyone past 90 to 120 days of silence gets one re-engagement email, then gets removed if they don’t respond.
- Look at your last ten subject lines. If you can’t tell them apart at a glance, that’s your engagement problem, not your spam filter.
- Add a one-click unsubscribe link if your platform doesn’t already include one. Gmail and Yahoo now require it, and making it hard to leave is one of the fastest ways to get reported as spam instead.
- Slow your send frequency back to whatever you promised people when they signed up. If you said weekly, send weekly.
None of this is glamorous. It’s also, almost entirely, the whole answer.
The bigger point about AI and marketing that this whole mess proves
Small business owners keep asking me whether AI will get their content penalised, whether it’s Google search rankings, social reach, or email deliverability. The honest pattern I keep finding, across every channel, is the same one from this story: the platforms are scoring behaviour and quality signals, not authorship. Alex Hormozi says something similar about offers, that the market doesn’t care how you built the thing, it cares whether the thing works for the person receiving it, and that lesson from his business lessons applies directly here. Nobody’s inbox provider cares that ChatGPT helped you draft paragraph three. They care whether the person who received it wanted it, opened it, and didn’t report it.
That’s a much more useful thing to obsess over than whether your writing sounds “too AI,” a worry that’s cost small business owners a lot of hours rewriting perfectly good emails for a problem that was never really there.
A note on repurposing, since you’re already doing the work
If you’ve fixed your list and your authentication, the email content itself becomes worth reusing rather than throwing away after one send. A well-performing newsletter section can become a carousel for Instagram, or the seed for a lead magnet quiz along the lines of the interactive calculators that convert visitors better than a generic PDF ever will. The point of fixing deliverability isn’t just cleaner inbox stats, it’s that the content becomes worth building a system around, instead of quietly vanishing into folders nobody opens.
Frequently asked questions
Does Gmail detect AI-written emails and mark them as spam?
No. There’s no evidence Gmail, Yahoo or Outlook detect AI authorship at all, and their own published guidance focuses entirely on authentication (SPF, DKIM, DMARC), spam complaint rates, and unsubscribe behaviour. The AI-detection theory is a myth that’s spread because the timing of AI adoption and deliverability drops often coincide, but the real cause is almost always volume, list hygiene, or missing authentication.
What spam complaint rate is too high?
Google and Yahoo both flag anything above 0.3% as a serious risk to your domain’s sending reputation. Most established senders sit well under 0.1%. If your platform’s reporting shows you above 0.3% even occasionally, treat it as urgent, because it affects deliverability for every email you send after that, not just the one that triggered complaints.
How often should I clean my email list?
Every 90 to 120 days at minimum. Pull anyone who hasn’t opened an email in that window, send one re-engagement email asking if they still want to hear from you, and remove anyone who doesn’t respond. A smaller, engaged list will consistently outperform a large, stale one, both in open rates and in actual sales.
Should I stop using AI to draft my newsletter?
No, use it to draft faster, then spend the time saved making the email specific rather than generic, a real story, a real number, an actual opinion. The risk with AI drafting isn’t detection, it’s that it makes it easy to send more, more often, to people who didn’t ask for more, and that behaviour is what damages deliverability.
Related reading: Why Your AI Content Sounds Like Everyone Else’s (And It’s Not the Tool’s Fault) and Why Your AI Customer Replies Are Quietly Losing You Repeat Customers.
