Straight answer: if you and your competitor both ask ChatGPT to "write a LinkedIn post about our new service," you will get suspiciously similar sentences, because you're both drawing from the same training data with the same default settings. The fix isn't ditching AI, it's feeding it something only you know before you let it write a word.
The experiment that made me stop trusting AI drafts
Back in early 2026 I was working with a small group of clients on their email newsletters, all in completely different industries: a bookkeeper in Kent, a physio in Manchester, a wedding photographer near Brighton. None of them knew the others existed. I asked all three to send me the prompt they'd used to get ChatGPT to draft their "welcome to my newsletter" email.
Two of the three opening lines started with a version of "In today's fast-paced world." The third opened with "We're excited to share." All three had a paragraph in the middle that used the word "journey" to describe their business. None of them had copied each other. They'd all just asked a similar tool a similar question and got a similar answer, because that's what the model does when you give it a generic prompt: it reaches for the most statistically likely next sentence, and the most statistically likely sentence is the one everyone else is also getting.
I decided to test it rather than just trust my gut. I took one prompt, "write an email announcing a 20% off sale to existing customers," and ran it through ChatGPT, Claude, and Gemini, five times each, with no other instructions. Out of fifteen drafts, twelve opened with some variation of "we've got exciting news" and eleven used the phrase "don't miss out" somewhere in the first three sentences. That's not a coincidence, that's the tool doing exactly what it was built to do: predict the safest, most average next word.
Why this happens, and why it isn't going away on its own
Large language models are trained to produce the response that's most likely to be rated as "good" by the humans who scored its training data. Safe, competent, mildly enthusiastic marketing copy scores well. Risky, specific, oddly personal copy scores less predictably. So the model defaults to safe. Multiply that by millions of small businesses all typing roughly the same prompts into roughly the same tools, and you get a flattening effect across entire industries. Estate agents' Instagram captions start to read like each other. Accountants' email subject lines converge. Coaches' "about me" pages all mention a "journey" and a "passion."
I've watched this happen inside my own client work over the past year, and the businesses that stood out weren't the ones who avoided AI, they were the ones who used it as a drafting tool and then did something to it that a competitor couldn't replicate. That's the part most people skip because it takes more effort than hitting generate.
The uncomfortable part nobody wants to say out loud
Here's the bit that makes people uncomfortable when I say it on a call: if your marketing sounds like it could have been written by any business in your sector, that's not because AI let you down, it's because you gave it nothing specific to work with. The tool didn't fail you. You handed it a blank page and asked for "professional and friendly" tone, and it gave you exactly that, in the most averaged-out form possible. The businesses I've seen lose customers to a competitor over the past two years weren't the ones using AI, they were the ones whose entire brand voice had become interchangeable with three other local firms, because everyone was pulling from the same well. Clients rarely leave because your product is worse. They leave because they can't remember why they chose you over the other one.
Compare that to brands that built something unmistakable long before AI existed. Look at how GoPro built a brand you'd recognise blindfolded just from the footage style, or how Instagram's own marketing approach leaned into a very specific visual identity rather than generic "engaging content." None of that came from a prompt. It came from decisions only that business would have made, based on facts only that business had.
What I tell clients to do instead
I'm not telling anyone to stop using ChatGPT, Claude, or whichever tool they've settled on. I use one most days. The fix is a five-step habit I've built into every content session I run with clients now.
- Feed it a fact only you have, before you ask for a draft. Not "write a post about our new pricing," but "we put our prices up by 8% because our supplier costs rose and we absorbed the difference for eleven months first, here's the actual figure." The specific detail forces the model away from the generic template.
- Delete the first two sentences it gives you, every time. In my testing across roughly forty client drafts this year, the throwaway opener ("In today's fast-paced world," "We're thrilled to announce," "Ever wondered why...") shows up in the first two sentences almost every time. Cut them. Start with sentence three or write your own.
- Ban a specific list of words from the final draft. Unlock,, journey, dive in, game-changer, elevate. If you search-and-replace these out of every AI draft before it goes anywhere, your copy already sounds less like everyone else's.
- Read it aloud in your own voice. If it sounds like a LinkedIn influencer you've never met rather than you on a Tuesday afternoon, rewrite the middle third by hand. That's usually where the model's confidence is highest and your voice is thinnest.
- Add one number, one name, or one detail nobody else would have. A client's name (with permission), a real date, a number that's oddly specific rather than rounded. "We fixed it in three days" beats "we fixed it quickly" every single time, and no competitor's AI draft will land on your three days by accident.
None of this is about writing longer copy or working harder. It's a ten minute edit pass, done consistently, on every piece before it goes out. The businesses I work with who do this see it show up in reply rates and comments, not just vanity metrics, because people respond to specificity even when they can't articulate why.
Where this shows up worst: email and social captions
Email is the format where I see the most damage, because most small business owners batch-write a month of newsletters in one sitting using one long AI session, which means every email in that batch inherits the same tics. If you want to see how a company with actual scale handles this, it's worth studying Mailchimp's marketing strategy, which leans hard into playful, specific, slightly odd copy rather than the safe corporate middle ground AI defaults to. Social captions have the same problem but it's more visible because people scroll past five similar captions in thirty seconds. If your last ten Instagram captions could be swapped with a direct competitor's and nobody would notice, that's worth auditing this week, not next quarter.
One format that's naturally harder to homogenise is interactive content, because it requires actual inputs from your business and your audience rather than a generic prompt. I've written before about using interactive calculators to engage and convert visitors, and the reason they still convert well in 2026 is that AI can't fake the underlying data. A calculator built on your actual pricing, your actual delivery times, or your actual case study numbers can't be replicated by a competitor typing a similar prompt, because the substance behind it is yours.
What this means for content strategy, not just single posts
Joe Pulizzi has spent years making the point that content only works long-term when it's built around something a business owns, whether that's a niche, a format, or a point of view, rather than chasing whatever topic is trending that week. I go back to some of the business lessons from Joe Pulizzi whenever a client tells me their AI-assisted content plan is "everything our industry talks about." That's precisely the trap. If your content calendar reads like a summary of what everyone in your sector already covers, AI will only make you faster at producing more of the same, which is worse, not better, because now you're publishing the sameness at volume.
The same logic applies to product-led brands. Look at how Figma built a brand around a very specific way of talking about collaboration rather than generic "productivity" language every design tool uses. That specificity is a decision, made by people, and it's the thing AI drafting tools will never generate for you unprompted, because it requires a stance, and stances don't come from averaging what's already out there.
When AI-as-is is fine
To be fair to the tools, there are jobs where the generic draft is exactly what you need and no editing pass is worth the time. Internal meeting summaries, first-pass customer service replies to routine questions, a rough outline before you write the real thing, a subject line variant to A/B test against your own. Nobody needs their internal Slack update to have a distinctive voice. Save the editing effort for anything a customer or prospect will read and remember, and let the tool do the boring, disposable stuff at full speed.
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.
If you're not sure whether your own content has drifted into the generic middle, or you want a second pair of eyes on where AI has quietly flattened your brand voice over the past year, that's a fairly quick audit an AI consultant working with small businesses can do in a single session, usually by comparing your last twenty pieces of content against three direct competitors' output side by side. It's an uncomfortable hour, but it's cheaper than another six months of content that isn't landing.
Free resource: The AI Agent Test Plan Template.
Frequently asked questions
Is it bad to use AI to write my marketing content?
No, using AI isn't the problem, using it without adding anything specific is. AI drafts are trained to produce safe, average copy, so if you publish the first output without editing in a fact, number, or detail only your business has, it will read like everyone else's AI draft, because it largely is.
How can I tell if my content sounds too AI generated?
Read your last five pieces of content next to a direct competitor's. If you could swap the opening two sentences between them and nobody would notice, that's the sign. Words like "unlock," "journey," "game-changer," and phrases like "in today's fast-paced world" are the clearest giveaways.
What should I do differently before asking AI to write a draft?
Give it a specific, real detail before you ask for the draft, a number, a name, a genuine reason behind a decision, rather than a general topic. The more specific the input, the less the output resembles what everyone else typing a similar prompt is getting back.
Should small businesses stop using ChatGPT for marketing?
No. It's fine for internal notes, first drafts, and routine replies. The problem only shows up when the first draft goes straight to customers without an edit pass to remove the generic phrasing and add something only your business could have written.
Related reading: Why Every AI-Written Newsletter Sounds the Same (And What I Changed In Mine) and The AI Sameness Problem: Why Your Small Business Marketing Sounds Like Everyone Else's.
For the bigger picture, see my full guide to AI marketing.
This overlaps with my main post on the topic: ai marketing sounds like competitors.
