Six steps from idea capture to published post, with worked examples, the time each step takes, and the failure modes at each stage. Built from 30+ voice system builds shipped to coaches, consultants, and B2B founders since early 2026 — what actually works at sustainable cadence rather than what looks good in a one-off demo.
Six steps. Capture the idea (your job, not ChatGPT's). Pick the hook formula. Generate the draft using a voice prompt loaded into a Custom GPT. Edit against your voice. Run the 12-point audit. Publish. Total time per post once the voice prompt is built: 15-25 minutes. Without the voice prompt, the same post takes 45-60 minutes because most time goes into rewriting AI output that does not match voice. The infrastructure investment is the gate.
The workflow below assumes you have already built the voice infrastructure. If you have not, the per-post time numbers do not hold; you will spend most of each post rewriting AI output rather than editing it. Three preconditions:
If those three are in place, the six-step workflow below produces voice-matched posts at sustainable cadence. If not, fix the infrastructure first; do not try to compensate at the per-post level.
Step 1 (2-3 min)
Write down the specific thing you want to say. Not a topic ("AI for marketing"); a specific point ("Most ChatGPT users skip the voice prompt because they think prompts are the asset; the prompt is delivery, the voice prompt is the asset"). The specificity is the post's spine. Without it, ChatGPT cannot produce useful output regardless of how clever the prompt is.
Sources for ideas: a thing you said in a meeting today that landed; a pattern you noticed across three client conversations this week; a contrarian observation about your industry; a specific failure you learned from; a question prospects keep asking that nobody answers in writing.
Step 2 (1-2 min)
Match the idea to a hook formula. Not every idea fits every formula; the wrong pairing produces awkward output. Twelve tested formulas covered in best LinkedIn hook formulas: specific number, pain-then-pivot, named scenario, contrarian observation, three-things compression, before-and-after, lesson-from-failure, question-as-claim, data-point, reframe, still-see-this, compressed admission.
Quick mapping: tactical ideas suit three-things or data-point; story ideas suit named scenario or lesson-from-failure; opinion ideas suit contrarian observation or reframe; reflective ideas suit before-and-after or compressed admission.
Variation rule: do not use the same formula in consecutive posts. Three posts in a row using "Most people think X..." reads as templated even if voice match per post is strong. Rotate 4-7 formulas across the week.
Step 3 (1-2 min)
Open your Custom GPT or Claude Project (voice prompt already loaded). Use the post-drafting conversation starter or paste the task prompt. Include the idea from Step 1 and reference the formula from Step 2.
The AI returns a draft. Time elapsed: 90 seconds for the AI to produce, 30 seconds for you to read.
Step 4 (10-15 min)
Three editing passes that compound:
Pass 1 (read aloud). Replace any phrases that do not sound like you. The voice prompt produces 70-85 percent voice match on first draft; this pass brings it to 90+ percent. Common edits: replace AI-default words that slipped past the prompt's banned list; rephrase any sentences that read as generic LinkedIn rather than your specific voice.
Pass 2 (sharpen openings and closings). The first 8-12 words decide whether anyone reads beyond the fold. The last sentence is the most-remembered line. Tighten both. The body can carry slightly looser writing; the bookends cannot.
Pass 3 (cut filler). Remove adverbs that do not earn their place. Replace hedging language ("perhaps", "might", "in some cases") where uncertainty is not real. Cut transition phrases that add words without adding meaning. The post should be tighter than the AI's draft, not longer.
Step 5 (2-3 min)
Run the 12-point audit against the edited draft. Twelve checks across voice match, default vocabulary, structural sameness, point of view, and finish. Each check is pass or fail; the total out of 12 is the audit score.
Score above 80 percent (10-12 / 12): ship.
Score 60-80 percent (7-9 / 12): edit specific failing sections. Identify which checks failed and address them directly rather than regenerating the whole post.
Score below 60 percent: the voice prompt itself is producing drift. Rewrite the post manually for this instance and tighten the voice prompt sections that failed before generating the next post.
Step 6 (1 min + monitoring)
Schedule the post or publish directly. Track engagement at 24 hours and 72 hours. Note which formulas perform best with your audience over 4-6 weeks of consistent posting. Patterns emerge: some audiences engage more with contrarian observations, others with named-scenario stories.
The learning data feeds back into Step 2 (formula selection) over time. After 4-6 weeks you stop guessing which formulas work; you know.
Step-by-step time allocation for one 200-word LinkedIn post once voice infrastructure is built:
Total: 17-26 minutes per post.
Across 5 posts per week: 85-130 minutes. Across 5 posts per week × 52 weeks: 74-113 hours per year. Compared to 45-60 minutes per post without voice infrastructure (195-260 hours per year for the same volume), the infrastructure saves 121-150 hours annually.
For a solopreneur billing £100/hour, the annual time saving is £12,100-15,000 in opportunity cost — orders of magnitude above the £497-997 voice prompt build cost.
Three failure patterns observed across the 30+ voice builds shipped:
1. Skipping Step 1 (idea capture). Users rush to generation and ask ChatGPT for ideas. The output reads as commodity content. The fix is making Step 1 non-negotiable: no draft generation without a 2-3 sentence captured idea.
2. Compressing Step 4 (editing). Users budget 15 minutes for editing and ship at 5 minutes when busy. The unedited drafts compound voice drift across the week. The fix is structural: schedule a 30-minute writing session per post, not a 15-minute one. Buffer ensures the editing pass actually happens.
3. Treating the workflow as linear when it isn't. Some posts need to loop back. A draft that fails the audit at Step 5 should trigger a return to Step 4 or even Step 3, not a forced ship. Loops are part of the workflow, not failures of it.
The six steps adapt to long-form with two changes:
The workflow shape is the same: idea capture, formula or structure selection, draft generation, editing, audit, publish. Only the per-step times scale up.
When the source material exists (a podcast, blog post, internal memo, webinar), Step 1 (idea capture) is replaced by source extraction:
Total time for 5 LinkedIn posts from one podcast episode: 2-3 hours including the source extraction. Detail in repurpose a podcast into 30 LinkedIn posts.
Three things this workflow does not solve:
DFY Voice System ships the voice prompt, Custom GPT, Claude Project, hook library and 5 sample posts in 2-3 working days. £497 founder pricing. Once the infrastructure is in place, the workflow above produces voice-matched posts in 17-26 minutes each.
See The Voice BuildSix steps: idea capture, hook formula selection, draft generation with voice prompt, editing, 12-point audit, publish. Total per post: 17-26 minutes once voice prompt is built.
15-25 minutes per post with voice infrastructure. 45-60 minutes without it because most time goes into rewriting AI output that does not match voice.
No. ChatGPT executes on existing thinking; it cannot produce useful new ideas about your specific domain. You supply the idea; the AI handles drafting.
There is no single best prompt. The strongest output comes from layering a voice prompt with task prompts. Pasting clever prompts into default ChatGPT without voice infrastructure produces generic output.
Yes for occasional use, with friction. Custom GPTs require Plus, so free-tier users paste the voice prompt every conversation. £20/month Plus pays back inside the first week of weekly cadence.
Three passes: read aloud and replace off-voice phrases, sharpen openings and closings, cut filler. Total: 10-15 minutes per 200-word post. Run the 12-point audit as the final gate.