Why Your AI Drafts Sound Generic (And the 4 Constraints That Fix It)
This note shows the operating context, the decision path, and the record behind it.
Evidence trail: Products
You can tell. Three sentences in, you can always tell.
"In today's fast-paced world." "It's worth noting." "This transformative approach." "Let's delve into." None of these phrases are wrong, exactly. They are just unowned — sentences that could have been written about any subject, by anyone, for anyone.
Search for a fix and you will land on a wall of humanizer tools and detector-score checkers, all promising to rewrite the draft until some meter turns green. That approach has a structural problem: it treats the symptom as the disease. A paraphraser swaps generic phrasing for awkward phrasing. The draft still has nothing to say.
The actual fix happens before the draft exists, in the constraints you set.
Why AI drafts sound generic
A language model produces the statistical center of everything it has read on your topic. Ask for "a blog post about project management software" and you get the average of ten thousand blog posts about project management software. That average is, by definition, the most generic possible output.
Generic writing is not a style failure. It is a specificity failure. The draft is vague because your instruction contained no information the model could not have guessed.
Which means the lever is not "write better," it is "give it something only you know."
Technique 1: Negative constraints
The most underused instruction in AI writing is the list of things it may not do. Models have verbal habits, and habits are bannable.
Do not use any of the following: "in today's world," "fast-paced," "delve," "it's worth noting," "transformative," "robust," "comprehensive," "unlock," "leverage" as a verb, "game-changer," "landscape" as a metaphor.
Additional bans:
- No sentence may begin with "In conclusion" or "Ultimately."
- No paragraph may open with a rhetorical question.
- Do not end sections with a summary sentence restating what you just said.
- Do not use the pattern "It's not just X, it's Y."Read the full operating note
That last one is worth its own line. The "not just X, but Y" construction is the single most recognizable tell in AI prose, and models reach for it constantly because it produces the feeling of insight without the substance of one.
Keep this list in a notes file and paste it into every drafting prompt. It costs you nothing and removes about half the problem.
Technique 2: Force a point of view
Banning phrases makes a draft less obviously machine-written. It does not make it worth reading. For that, the draft needs an argument someone could disagree with.
Before writing, state in one sentence the specific claim this piece is making — a claim a reasonable expert in this field could argue against.
Then write the piece defending that claim.
If the claim is something no one would dispute, reject it and choose a sharper one. Do not present "both sides" unless I asked for a balanced overview.
Balance is the default failure mode of AI drafting. The model hedges because hedging is safe, and the result reads like a committee wrote it. Forcing a defensible claim up front is what turns a summary into a piece.
Technique 3: Feed it what it cannot know
This is the technique that separates client-ready work from filler, and it is the one nobody automates around.
The model cannot know your client's actual numbers, the objection their sales team hears every week, the thing that went wrong on the last project, or the sentence a customer said in an interview. Those specifics are the entire difference between "content" and something worth someone's attention.
Read the full operating note
Use only the source material below for all factual claims, examples, and figures. Do not add illustrative examples of your own invention.
Source material: [PASTE INTERVIEW NOTES, CALL TRANSCRIPTS, INTERNAL DATA, SUPPORT TICKETS]
Where the source material does not cover a point I asked for, write [GAP: what is missing] instead of filling it in.
The [GAP] instruction is the important half. Left to itself, a model will paper over a missing fact with a plausible generic sentence, and you will not notice — because plausible generic sentences are exactly what you were trying to eliminate. Making the gaps visible turns an invisible quality problem into a visible to-do list.
Technique 4: Constrain the shape, not just the words
AI drafts have a recognizable rhythm: paragraphs of near-identical length, every section symmetrical, every list exactly three items. Human writing is lumpy.
Vary paragraph length deliberately. At least two paragraphs must be a single sentence. Do not make every section the same length. Lists may have two items or seven — do not default to three.
Small instruction, disproportionate effect. Structural monotony is what people register as "AI-sounding" even when they cannot name why.
The order matters
Run these as a sequence, not a pile:
- Point of view — decide the claim before any prose exists.
- Source material — load the specifics only you have.
- Negative constraints — ban the verbal habits.
- Shape constraints — break the symmetry.
- Read it aloud — the one step with no prompt. Anything you stumble over is a sentence the model wrote for itself rather than for a reader.
Read the full operating note
Do this and you stop editing AI slop out of drafts, because you stopped generating it.
Read the full chapter free
This is the short version of Chapter 4, "Curing AI Slop" — from The One-Person AI Content Agency. The full chapter is published free, in its entirety, with no email required: the complete system prompts, the persona frameworks, and the negative-constraint patterns that strip generic phrasing out of client-ready drafts.
👉 Read Chapter 4 in full, free
If it earns its place, the complete book covers the rest of the operation: positioning, client intake, production workflow, quality control, and pricing.
👉 See the full book — $39, one-time
📺 One useful AI signal every day
What changed in the tooling, and what it means for real work.
Comments
Post a Comment