The short version: AI writing tools don't make your content bad, they make it average, because averaging is literally what the model is built to do. If your LinkedIn posts, your emails, and your website copy sound like everyone else's in your industry, that's not a tone problem, it's a training data problem, and there's a specific way to fix it that most people skip because it takes longer than typing a prompt.
The afternoon I realised my clients all sounded like the same person
I was reviewing content for three separate clients in one sitting last autumn. A bookkeeper in Manchester, a wedding photographer in Tel Aviv, and a B2B software reseller in Austin. Nothing to do with each other. Different industries, different countries, different audiences.
All three LinkedIn posts opened with a version of "In today's fast-paced world." Two of them used "when it comes to" within the first two sentences. All three ended with some form of "the choice is yours" or "the future is here." I hadn't written a word of it, they'd each used ChatGPT to draft it themselves, which is fine, that's what I tell people to do. But sat next to each other, the three posts read like they'd come out of the same content mill.
That's when I stopped treating this as a fluke and started treating it as a pattern worth testing.
The test: 40 drafts, one obvious result
I asked 40 small business owners in my network (clients, past clients, people from a mastermind I run) to send me the last AI-assisted piece of content they'd written, no editing, straight out of the tool. LinkedIn posts, email newsletters, About pages.
29 out of the 40 opened with a version of one of these three constructions: "In today's [fast-paced/ever-changing/digital] world," "Have you ever wondered," or "When it comes to [topic]." That's 72.5 percent using one of three templates, from 40 unrelated businesses with no shared writer.
17 of the 40 closed with some flavour of "the choice is yours," "the future is here," or "the possibilities are endless." Nobody told these people to write that way. The tool did it for them, quietly, in every single draft, because it's the statistically likely next word given everything the model has read before.
Why this happens (and it isn't laziness)
Here's the bit people don't say out loud enough: a large language model is a prediction engine trained on huge amounts of existing writing, so its default output is the mathematical middle of everything that's already been said on a topic. It's not trying to be original, it's trying to be the most probable next sentence. That means the "default AI voice" isn't a bug you can prompt away with "write more casually," it's the entire point of how the tool works.
Small business owners aren't getting worse at writing because of AI. They're getting more identical to each other because they're all drawing from the same statistical middle, without adding the one thing a model can't invent: their own specific, lived, slightly odd detail.
I'll say the uncomfortable part plainly. Most business owners who complain that "AI content doesn't convert" haven't got a content problem, they've got a courage problem. They're scared to write the specific, opinionated, slightly risky version of the sentence, so they let the tool smooth it into the safe, generic one, and then they wonder why nobody remembers reading it.
What the brands that don't sound generic are doing
None of this is new if you look at brands that built real distinctiveness before AI existed. When you study how GoPro built a following, the pattern isn't clever copywriting at all, it's that they handed the camera to actual customers and let unpolished, specific footage do the talking, something a language model would never suggest because it's not the "safe" answer. My piece on the GoPro marketing strategy goes into how that specificity became the whole brand.
Same with Instagram in its early years, the growth wasn't from generic messaging, it was from a narrow, specific promise (square photos, one filter tap, done) that I break down in my Instagram marketing strategy piece. Figma did something similar by being loudly, specifically obsessed with one narrow use case (real-time design collaboration) rather than trying to sound like every other software company, which I cover in the Figma marketing strategy post. And Mailchimp's early brand voice was odd on purpose, weird illustrations, a chimp mascot, copy that didn't sound like every other email tool, which you can see laid out in the Mailchimp marketing strategy breakdown.
None of those brands got there by asking a model to "write a LinkedIn post about our values." They got there by being specific about one true, slightly uncomfortable thing, then repeating it until it stuck.
The fix I use with clients, step by step
This isn't "stop using AI." I use it every day, for research, for first drafts, for restructuring a messy paragraph. The fix is adding back the four things a model can't generate on its own.
- Step 1, bank real transcripts. Before you write anything, pull three actual customer conversations, a support email, a sales call, a review. Paste the exact wording customers use, not your summary of it. This is the raw material a model doesn't have access to.
- Step 2, write the ugly first line yourself. Never let AI write your opening sentence. It's the sentence most likely to default to "In today's world." Write one true, specific, slightly odd sentence, then hand the rest to the tool to expand.
- Step 3, force in one real number every time. Not "many clients saw growth," but "14 out of 22 clients I worked with in Q3 increased email open rates," even if the number is small or unimpressive. Specific numbers are the single fastest way to break the generic pattern.
- Step 4, cut the closing platitude entirely. If a draft ends with "the choice is yours" or "the possibilities are endless," delete the whole sentence. End on the number, the detail, or a plain instruction instead.
- Step 5, read it against a competitor's post. Put your draft next to a rival's most recent post. If you could swap the company name and nobody would notice, you're not done editing.
This is roughly the same discipline Joe Pulizzi built his entire content marketing career on, long before AI existed, and it's covered well in my business lessons from Joe Pulizzi post: the content that compounds is the content nobody else could have written, because it's built from a real, owned experience rather than a general topic.
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.
The format that AI struggles to copy
There's one type of content that's almost impossible for a competitor to replicate by prompting a chatbot, and it's the one most small businesses ignore because it takes actual build time: interactive tools built around your specific data. A pricing calculator, a quiz that scores a visitor's situation, a quote generator. I wrote about this in the piece on interactive calculators for engagement and conversion, and the reason it still matters now is that a language model can write about your service, but it can't fake a tool that calculates a real answer using your actual pricing logic. That's a moat AI can't shortcut.
A number worth sitting with
Here's the part that should worry you a bit more than it probably does. If 72.5 percent of unrelated small businesses in my small test opened with one of three constructions, and millions of small businesses worldwide are now using the same handful of AI tools with the same default settings, the maths says a very large share of small business content published this year opens with nearly identical sentences. Your prospect isn't comparing you to one competitor anymore, they're scrolling past a feed where a large chunk of it reads the same. Blending in used to cost you a bit of engagement. Now it costs you being noticed at all.
What I changed in my own content after the test
I went back through my last twenty LinkedIn posts and found six that opened with some variation of "when it comes to." I rewrote every one of them starting with either a number, a date, or a direct quote from a client email. Engagement on the rewritten posts (comments and shares, not just likes, because likes are close to meaningless) ran roughly 2 to 3 times higher than the originals over the following month, based on my own Bulk Analytics export, nothing scientific, but consistent enough across six posts that I don't think it was chance.
I didn't stop using AI to draft. I just stopped letting it write the first and last sentence, and I stopped submitting anything for publishing until it had at least one number, one name, or one detail that only I could have known.
What to do this week
Pull your last five pieces of AI-assisted content, whatever platform they're on. Check the opening line of each one against the phrases above. If two or more match, you've found your problem, and it isn't the tool, it's the habit of publishing the first draft it gives you.
Frequently asked questions
Does using AI to write marketing content hurt my results?
Not the tool itself, but publishing the first draft without adding specific numbers, names, or real customer language usually does, because that draft reads almost identically to what your competitors are publishing from the same tools.
How can I tell if my content sounds generic?
Check the opening and closing sentence. If it starts with "in today's fast-paced world," "have you ever wondered," or "when it comes to," or ends with "the choice is yours," rewrite it. Those phrases show up disproportionately often in AI-assisted small business content.
Should small businesses stop using AI for content then?
No. Use it for structure, research, and first drafts, but write your own opening line, add at least one real number or customer quote, and cut generic closing lines before you publish anything.
What's a quick way to build content competitors can't copy with AI?
Build something interactive around your own data, a pricing calculator, a scoring quiz, a cost estimator. A model can describe your service in generic terms, but it cannot fake a tool running your actual numbers.
Related reading: What Your AI Meeting Notetaker Is Doing With Your Client Calls and Why Every AI-Written Newsletter Sounds the Same (And What I Changed In Mine).
More on this here: an AI consultant in Tel Aviv.
