You can spot AI writing from across the room now, can't you.
The tidy three-point lists. The "in today's landscape." The relentless, polite blandness that reads like a brochure nobody asked for. I used to produce exactly that, and I couldn't understand why my AI drafts felt so lifeless when other people's didn't.
The answer turned out to be embarrassingly simple. It wasn't the model. It was me. I was asking for the average, and getting it.
Here's the one habit that fixed it.
AI output looks generic because AI gives you the statistical average of everything ever written on a topic unless you tell it not to. The fix is to feed it specifics only you have — your real examples, your actual opinions, your weird constraints — so it writes from your world instead of the internet's. Generic in, generic out. Specific in, you out.
A language model is, at its heart, a beautiful averaging machine. Ask it for "a blog post about productivity" and it returns the center of mass of a million productivity blog posts. Smooth. Competent. Forgettable.
That average is exactly what makes it sound like everyone else — because it literally is everyone else, blended together. You asked for the most likely sentence, and the most likely sentence is, by definition, the one everyone's already read.
The model isn't being lazy. It's giving you precisely what you asked for. The genericness is a reflection of a generic request. The whole game is learning to ask for something the average can't produce — which is the same uncomfortable point I keep coming back to in the honest truth about AI productivity tools: the output is only ever as specific as what you bring to it.
This isn't just a vibe. The Stanford HAI AI Index documents how rapidly these models converged on fluent, confident, broadly average prose — fluency is solved, distinctiveness is not.
Here's the shift. Stop asking AI to write for you. Start asking it to write from you.
Before I draft anything now, I dump in the raw material only I have:
Then I tell it to use those and nothing else as the spine. Suddenly the output can't be generic, because the generic version doesn't contain my Tuesday, my client's weird complaint, or the number I'm slightly embarrassed by.
AI gives you the average of the internet. Your job is to give it the part of the internet that doesn't exist yet — you.
The difference is night and day. A prompt that says "write about email automation" gets you beige. A prompt that says "here's the exact email that wasted my morning, here's why it made me angry, write about that" gets you something nobody else could have written. It's the same structural fix behind the prompt pattern that fixed my AI output — feed it the part of the world only you have access to.
Even with good material, AI smuggles in its dialect. So I keep a kill-list and bolt it onto every drafting prompt.
The phrases I ban outright:
And one structural rule: no paragraph longer than three sentences, and vary the lengths. AI loves to produce identical, evenly-weighted blocks. Real writing has rhythm — short, short, then a long one that breathes, then a fragment. Like that.
Generic writing is afraid to take a side. It hedges. "There are many perspectives." "It depends."
So I force a position. I tell the AI: "Take the strongest defensible opinion on this and commit to it. No hedging. If there's a counterargument, address it and then say why you still believe your side."
Opinion is the fastest cure for blandness there is. A confident wrong take is more useful — and more human — than a perfectly balanced non-statement. You can always dial a strong draft back. You can't make a beige one interesting.
I ran the same topic two ways to prove it to myself.
| Generic prompt | Specific prompt | |
|---|---|---|
| Opening | "AI is transforming how we work." | "I deleted my to-do app at 11pm out of spite." |
| Detail | abstract, no names | real tools, real numbers |
| Stance | balanced, hedged | one clear opinion |
| Felt like | a press release | a person |
Same model. Same topic. The only thing that changed was what I brought to it.
It's worth understanding why this works, because once you do, you stop needing tricks.
A specific detail does two jobs at once. First, it constrains the output — every concrete fact you add removes a thousand generic directions the AI could have wandered. Second, and more subtly, specifics carry texture that generic prompts can't fake. "I deleted my to-do app at 11pm out of spite" tells the AI about emotion, timing, and stakes all at once. The average can't reverse-engineer that from "write about productivity tools." It was never in the training data, because it's yours.
This is also why personal stories work so well, even tiny ones. A story is just a dense bundle of specifics — a who, a when, a problem, a feeling. Drop one in and the AI has no choice but to write around it, and writing around a real moment is structurally incapable of being beige.
The same logic applies to opinions. "Most inbox-zero advice is wrong because it just makes you do manual work faster" is a specific claim — it has an enemy, a reason, an edge. The average has no enemies. It agrees with everyone. The moment you hand the AI a position to defend, it stops hedging and starts sounding like a person who believes something.
So the real skill isn't prompt engineering. It's noticing what you uniquely have — the moment, the number, the strong take, the weird constraint — and refusing to start writing until you've handed it over. Generic output is almost always a symptom of a writer who skipped that step.
Here's a check I run on anything AI helped me write, and it catches beige output every time.
Read it back and ask one question: could a competitor have published this exact piece with their name on it?
If the answer is yes, it's generic, and you already know what's missing — you. There's no specific moment in it, no number only you'd have, no opinion sharp enough to belong to one person. It's the average wearing your logo. Swap the logo and nobody would notice.
When something fails the test, I don't reach for fancier prompts. I go back and add the things only I have. The real client complaint. The actual hour I gave up. The number I'm slightly embarrassed by. The take I'd defend in an argument. Almost always, two or three specific injections drag the whole piece from "anyone could've written this" to "only I could have."
The test also works in reverse, as a filter on your inputs. If you can't find a single specific to add — no story, no number, no real opinion — that's a sign you don't actually have anything to say on the topic yet, and no amount of clever prompting will hide it. Genericness is sometimes the symptom; emptiness is sometimes the disease. The cure for both is the same: go get something real to say, then let AI help you say it.
Next time you draft something with AI, try running the "could a competitor publish this" test before you hit save — it's the fastest way to catch beige writing while it's still fixable.
Q: Is this just about adding personal stories? Stories help, but it's bigger than that. It's about feeding any specific only you have — a number, an opinion, a constraint, a real objection. Specificity is the raw material; story is one form of it.
Q: Won't editing out the "AI tells" take forever? A kill-list in your prompt prevents most of them up front. The cleanup is minutes, not hours. And you get faster at spotting the remaining ones as you go.
Q: Doesn't a strong opinion risk being wrong? Sometimes. But a clear, slightly-wrong take starts a conversation and gets corrected. A perfectly safe non-take just gets scrolled past. Wrong is recoverable. Boring isn't.
Q: Does this work for any kind of content? Yes — emails, posts, reports, scripts. Anywhere AI is producing beige, the cause is the same: you asked for the average. The cure is always more of you in the prompt.
Your AI doesn't sound like everyone else because it's broken. It sounds like everyone else because you asked for the most likely words, and the most likely words belong to everyone.
Give it your Tuesday. Your number. Your too-strong opinion. The objection only your audience makes. Feed it the part of the world that isn't on the internet yet — and the output stops being generic the instant the input does.
So here's the question that matters: what do you know that the average doesn't?
No following, no network, no luck. Just an unglamorous system I ran for eighteen months. Here's exactly what I did.
I went from 200 to 11,000 subscribers without hiring anyone. AI didn't write my newsletter — it did everything around it.
One person, output that looks like five. It isn't about working more hours — it's about a kind of leverage teams rarely have.
Comments
Sign in to join the conversation
No comments yet. Be the first to share your thoughts!