Write a blog post with AI without it sounding like AI
A working process for drafting with a model and ending up with something a human would actually publish.
Before you start
- An account with any general-purpose assistant (ChatGPT, Claude or Gemini)
- A topic you already know something about
- Somewhere to draft — a doc, not the chat window
What you will be able to do
- Produce a first draft that needs editing rather than rewriting
- Recognise the four phrases that make a post read as machine-written
- Keep your own voice through a model-assisted draft
The complaint about AI writing is almost never that the grammar is wrong. It is that the result is flat — technically fine, saying nothing, in a register nobody actually uses. That is usually a prompting problem rather than a model problem.
The process below is the one that survives contact with an editor. It takes longer than asking for a blog post and pasting the answer, and that is the point.
Decide the argument before you open the chat
One sentence that could be disagreed with. If nobody could argue with it, it is not an argument.
Write down what you actually think, in one sentence, before you prompt anything. "Remote work is complicated" is not an argument. "Remote work fails for junior staff specifically, and companies keep blaming the wrong thing" is.
Everything after this step is arranging evidence around that sentence. Skip it and you get a post that surveys a topic instead of saying something, which is exactly the flatness people recognise as machine-written.
- If you cannot write the sentence, you are not ready to write the post — you are still researching.
Give the model your material, not a topic
Paste your notes, your examples and your half-formed objections. Then ask for structure.
The single biggest quality jump comes from changing what you hand over. Instead of "write a blog post about onboarding remote juniors", paste 300 words of your own messy notes and ask it to propose three possible structures.
A model given raw material arranges it. A model given a topic invents material, and invented material is generic by definition.
Here are my raw notes on onboarding remote juniors.
The argument I want to make is: [your one sentence].
Do not write the post yet. Propose three different structures,
and for each one tell me what it would have to leave out.
[paste your notes]- Asking for the post immediately. You will spend longer fixing the draft than you saved.
Draft one section at a time
Whole-post drafts are uniformly mediocre. Section drafts can be individually good.
Ask for one section, read it, react to it, then ask for the next with your reaction included. The model has your feedback in context and the second section is measurably better than the first.
A whole draft in one shot gives the model no chance to learn what you actually want, so every section is the same distance from right.
Delete the four tells
Search the draft for four specific patterns and remove every instance.
These are the phrases readers have learned to recognise:
- "In today's fast-paced world" and every variant — an opening that says nothing.
- "It is important to note that" — always deletable with no loss.
- "Whether you are a beginner or an expert" — addressing everyone means addressing nobody.
- Three-item lists where the third item is filler — models pad to three. Cut to two if the third is weak.
- Use your editor's find function. This takes four searches and removes most of the machine smell.
Put back what only you know
The specific number, the client who complained, the thing that went wrong. The model cannot invent these.
Go through the draft and add at least three concrete details a model could not have known: a real number, a real conversation, a real mistake you made. These are the only parts of the post that could not have been written by anyone else, and they are what makes it worth reading.
- Letting the model invent the specifics. A plausible fake statistic is worse than no statistic.
The pattern underneath all of this: the model is good at structure and bad at judgement. Give it the judgement — your angle, your examples, your objections — and let it do the arranging.
Downloads & resources
Common questions
Was this guide useful?
88% of readers found this useful
Read next
Edit AI drafts so they stop sounding like drafts
Readers cannot always name what makes text feel machine-written, but they notice. These are the specific tics, and most of them ar…
Research with AI without inheriting its mistakes
AI is genuinely fast at finding leads and genuinely unreliable at facts. The workflow that works keeps those two roles strictly se…
Write ad variants and test them properly
AI removes the cost of writing variants, which makes it very easy to run tests that cannot teach you anything. The discipline is i…
Connect two apps with an AI step in the middle
The genuinely useful automations are not the clever ones. They are a trigger, one AI step that makes a small judgement, and a writ…
