Write the post in five ordered steps: lock the angle and outline yourself, feed the model your source text and voice samples, draft section by section with hard constraints, revise for specificity, then verify every claim before you publish.
The generic feel comes from asking a model to write the whole post from a one-line prompt. When you give it nothing but a topic, it reaches for the most common phrasing on that topic, because that's what its training rewards. Give it your own material and your own structure, and the output stops sounding like everyone else's.
Step 1: Lock the angle and outline yourself. Before you open any AI tool, write two things by hand: the specific argument of the post, and a five-to-seven line outline. "How to save money on cloud hosting" is a topic. "Why small teams overspend on cloud hosting in their first year, and the three habits that fix it" is an angle.
The model can't pick your angle for you, because the angle is the part that carries your judgment. Once the outline exists, the model's job shrinks from "invent a post" to "expand this line into a paragraph" — a much smaller job, and one it does far better.
Step 2: Feed it source text and voice samples. This is the step most people skip, and it's the one that changes the output most. Paste in the raw material the post should be built from — your interview notes, a transcript, a competitor page you're responding to, your own earlier drafts.
Then paste in two or three paragraphs you've written before, with the instruction: "Match this voice. Do not use words I wouldn't use." According to our internal AI tool database, which tracks 360 AI tools with pricing and capability snapshots verified as recently as 2026-09-18, almost every mainstream writing tool supports some form of reference-text input, but the quality of what you get back tracks the quality of what you put in.
Vague input returns vague prose. Specific input — real names, real numbers, real quotes — returns prose that has something to say.
Step 3: Draft section by section with hard constraints. Don't ask for the whole post at once. Ask for one section at a time, and attach constraints to each request. Useful constraints include: "Use at least one named example," "Include a specific number from the source text," "No sentence longer than 25 words," "Do not use these words: [list]."
Keep a running ban list of the phrases that make writing sound machine-made — "in today's fast-paced world," "it's important to note," "unlock the potential," "game-changer." Add to that list every time you catch one. A concrete example: for a post about onboarding emails, instead of "Write a section on timing," try "Write one paragraph arguing that the first email should go out within ten minutes, using the signup data in the attached notes.
Name the metric. No more than 120 words." You'll get something usable, not something you have to rewrite from scratch.
Step 4: Revise for specificity. Read the draft and hunt for sentences that could appear in any post on this topic. Those are the generic ones. Replace each with something only you could have written — a number from your own data, a mistake you've seen, a customer quote. Rewrite the opening and closing paragraphs entirely in your own words; models are weakest at beginnings and endings, because those are where a human voice matters most. A quick test: cover the byline and ask whether a reader could guess who wrote it. If not, it's still generic.
Step 5: Verify every claim. This is the step that keeps you out of trouble. Models will state a confident fact that turns out to be wrong, and it will sit in your draft looking perfectly normal. Check every number, date, name, and quote against a real source before publishing. If you can't find the source, cut the claim. According to our AI tool database, capability snapshots are recorded at a specific verification date precisely because tool behavior changes — the same logic applies to any fact you publish.
Where this process breaks down: it costs time. Steps 1, 4, and 5 are human work, and they're the slow part. If you're producing fifty thin posts a month, this workflow won't scale, and you should probably produce fewer posts.
It also fails when you have no source material — if you're writing about something you don't actually know, no prompt fixes that. The honest limit is that AI speeds up the middle of the process, the drafting, and does little for the beginning and the end, which are the parts readers actually remember.
Tools like a zero-prompt AI content generator can handle the first-draft stage, but the angle, the voice, and the verification stay yours.