You recorded a good conversation about your work, and it's still sitting in a folder. Every new post starts from a blank page anyway, because turning 30 minutes of talk into finished pieces feels like a bigger job than writing something fresh.
The usual AI shortcut makes it worse. Paste the transcript, ask for 10 posts, and you get 10 versions of the same point with the specifics sanded off.
Part 3 of the build takes the slower route that works. You'll tag the useful passages, plan 5 pieces for 5 different reader situations, draft each from its passages only, and check every claim against what was said. The example is a 30-minute interview with Maya Okafor of Ridgeline Ops, both illustrative and invented for this series.
Jump to the free Interview-to-Content AI Workbook
Start the AI content repurposing workflow by marking ideas and evidence
An AI content repurposing workflow starts with marking the transcript before any prompt runs. Read it once with timestamps on, tag each passage as a claim, example, story, rule or signal, and flag every number for checking against your notes. The tagged passages become the only material the assistant may use.
Maya's tagged transcript (excerpt)
Maya runs an operations consultancy for 20-to-80-person distribution companies. Part 1, building your own AI content engine, drew a post from her interview's rush-order section; this excerpt covers the daily handoff huddle. The tags in brackets are hers, and a SIGNAL marks how a reader will know the advice is working.
[9:00] Q: What do you set up first with a new client?
[9:10] A huddle. 15 minutes at 8:15, standing, sales lead and shift
lead. It's the cheapest fix I have. [CLAIM]
[10:15] Same 3 questions in the same order every day. What did we
promise yesterday that isn't on the floor yet? Which orders
going out today look risky? Who's changing an order after
noon? [EXAMPLE]
[11:40] At 1 client the huddle grew to 45 minutes because finance
joined with the aging report. It was dead in 3 weeks.
[STORY] [NUMBER: check N-19]
[12:30] My rule now: if it isn't about an order in the building or
promised to the building, it goes somewhere else. [RULE]
[13:20] You'll know it's working when the shift lead stops hearing
about rush orders from the customer's driver. [SIGNAL]
[24:00] Write the 3 questions on a card and tape it to the
whiteboard. People forget the order when it lives in your
head. [EXAMPLE]
[26:40] At the fastener client the huddle caught 9 promised orders
in the first 4 weeks that weren't in the system yet.
[NUMBER: check N-3]
Check the numbers before drafting
Both numbered passages go against the notes first. N-19 confirms the 45-minute huddle and the 3 weeks. N-3's follow-up says the huddle caught 7 orders in its first 4 weeks, so 9 was memory talking, and every piece uses 7.
That's why tagging comes first. A transcript records what someone remembered while talking, and the notes record what happened.
The extraction prompt
Below is a tagged excerpt from an interview with me, then the
corrections my notes made to it.
List each tagged passage as a row: timestamp, tag, the idea in 1 plain
sentence. Apply every correction, keep everything else as said, and
add no ideas that aren't in the excerpt.
Excerpt: [paste the tagged excerpt]
Corrections: [each change, e.g. 26:40: the huddle caught 7 orders;
the transcript says 9]
Save the list it returns. It's the source sheet for all 5 pieces, with the correction already in it.
Create 5 pieces for distinct reader situations
Each of the 5 pieces should serve a different reader at a different moment, which is what stops them repeating each other. Plan that in a repurposing matrix before drafting: 1 row per piece, with its reader situation, its job, the claim it carries, the exact passages behind it and its next step.
The repurposing matrix
| ID | Format | Reader situation | Job | Main claim | Passages | Next step |
|---|---|---|---|---|---|---|
| RL-5 | LinkedIn post | Owner whose sales and warehouse teams only talk when something breaks | Make the huddle sound cheap enough to try Monday | 15 minutes, 2 people, 3 questions | 9:10; 10:15; 24:00; 26:40 with N-3 | Reply "huddle" for the card |
| RL-6 | Article outline | Ops director ready to set 1 up properly | Teach the full method | Format, questions, scope rule, signs it works | 9:10 to 13:20 with N-19; 24:00; 26:40 with N-3 | Download the huddle card |
| RL-7 | Subscriber who tried a meeting like this and watched it die | Explain why it died | Scope creep kills huddles, and a scope rule protects them | 9:10; 11:40 with N-19; 12:30 | Reply with how long yours ran | |
| RL-8 | Short video, under 60 seconds | Owner scrolling between calls | Deliver the 3 questions | The questions, in order | 9:10; 10:15; 24:00 | Card link in profile |
| RL-9 | Carousel outline | Shift lead or ops director saving it for later | Hand over the card | Questions, scope rule, 7 caught orders | 9:10; 10:15; 12:30; 26:40 with N-3 | Save, then reply "card" |
Draft each piece in its own chat with the voice pack from part 2, getting AI to write in your voice, plus only the passages in its row.
Using my voice pack, draft [format] for this reader: [reader situation].
Its 1 job: [job]. Use only these passages: [paste rows from the source
sheet]. End with this next step, word for word: [next step].
The 5 pieces below are Maya's approved versions, written for this article as illustrations. Her edits are covered in the checks that follow.
RL-5: LinkedIn post, approved version 2
Monday, 8:15. Sales lead and shift lead at the whiteboard, standing, 15 minutes.
They answer 3 questions in the same order every day. What did we promise yesterday that isn't on the floor yet? Which orders going out today look risky? Who's changing an order after noon?
At a fastener distributor I worked with, that huddle caught 7 promised orders in its first 4 weeks that weren't in the system yet.
It's the cheapest fix I have. Reply "huddle" and I'll send you the card we tape to the whiteboard.
RL-6: article outline, approved version 1
Working title: How to Run a 15-Minute Sales-to-Warehouse Huddle
- Who attends, when and for how long (9:10)
- The 3 questions and why their order matters (10:15; 24:00)
- What stays out: the scope rule, and the 45-minute huddle that died (11:40; 12:30)
- Signs it's working, including 7 orders caught in 4 weeks (13:20; 26:40 checked against N-3)
Next step: download the huddle card.
RL-7: email, approved version 3
Subject: Why your morning meeting died
At 1 client, the sales-to-warehouse huddle started at 15 minutes and grew to 45 once finance joined with the aging report. 3 weeks later it was gone.
The fix was a scope rule: if it isn't about an order in the building or promised to the building, it goes somewhere else.
If you've tried a huddle before, hit reply and tell me how long yours ran before it stopped.
RL-8: short-video script, approved version 2
| Beat | Spoken | On screen |
|---|---|---|
| Hook | "Your sales team and your warehouse need 15 minutes a day. Here's what goes in them." | Whiteboard, 8:15 |
| Questions | "First: what did we promise yesterday that isn't on the floor yet? Then: which orders going out today look risky? And last: who's changing an order after noon?" | Each question appears as it's said |
| Card | "Same 3 questions in the same order daily. Write them on a card and tape it up, because people forget the order when it lives in your head." | Card taped to the board |
| Next step | "The card's at the link in my profile." | Profile link |
Read it aloud with a timer before filming, and trim until it runs under 60 seconds.
RL-9: carousel outline, approved version 2
- Cover: The 15-minute huddle that keeps sales and the warehouse talking
- When and who: 8:15, standing, sales lead and shift lead
- Ask first: what did we promise yesterday that isn't on the floor yet?
- Then ask: which orders going out today look risky?
- Finish with: who's changing an order after noon?
- Scope rule: if it isn't about an order in the building or promised to it, it goes elsewhere
- Proof: at 1 fastener distributor, 7 promised orders caught in the first 4 weeks
- Save this, then reply "card" for the printable version
The Interview-to-Content AI Workbook has the full excerpt, matrix, claim map and all 5 pieces, plus a blank workbook for your recording.
Check meaning, repetition and each next step
Checking repurposed pieces takes 3 passes. Confirm each claim still means what the speaker meant, make sure no 2 pieces tell the same reader the same thing, and confirm each next step fits its format. Run all 3 with the matrix and the tagged transcript open beside the drafts, and fix the draft whenever it disagrees with the source.
Meaning check, claim by claim
The draft claims below come from illustrative example drafts written for this article; no tool produced them. The last column shows what Maya approved in their place.
| Claim in an example draft | Piece | Source passage | Verdict and approved wording |
|---|---|---|---|
| "The huddle fixes late orders" | RL-5, version 1 | 9:10: "the cheapest fix I have" | Overstated. Approved: "It's the cheapest fix I have." |
| "caught 9 promised orders" | RL-9, version 1 | 26:40, corrected by N-3 to 7 | Wrong number: the draft ran in a chat holding the raw transcript. Approved: "7 promised orders caught", redrafted from the source sheet |
| "Finance ruined the meeting" | RL-7, version 1 | 11:40: finance joined with the aging report | Blame added. Approved: "grew to 45 once finance joined with the aging report" |
| "3 questions in the same order" | RL-8 | 10:15 | Matches |
| "tape it to the whiteboard" | RL-5, RL-8 | 24:00 | Matches |
Repetition check
Lay the 5 main claims side by side. Every piece except the email carries the 3 questions, which works because each reaches a different reader in a different format: a post read once in the feed, an article read in full, a video caught mid-scroll and a carousel saved for the next huddle.
RL-5 and RL-7 had both opened with the 45-minute story in version 1. The story belongs to the email, whose reader has watched a meeting die, so 11:40 came off RL-5's matrix row and the post now opens at 8:15 at the whiteboard. The matrix also gains a Used in column, so the next overlap shows up before drafting. Spread publishing across 3 weeks too, so the same followers don't meet the huddle 5 times in 5 days.
Next-step check
| Piece | Next step | Why it fits |
|---|---|---|
| RL-5 | Reply "huddle" | A reply opens a conversation Maya can follow up by hand |
| RL-6 | Download the huddle card | An article reader has time for a longer resource |
| RL-7 | Reply with how long yours ran | The answer tells her who has tried this before |
| RL-8 | Card link in profile | A viewer mid-scroll won't stop to type a keyword |
| RL-9 | Save, then reply "card" | Saving suits a reference format, and its own keyword keeps its replies apart from RL-5's |
In the AuthorityOS Sparklers and Bonfires model, the article from RL-6 is the Bonfire, the deep piece, and the post, video and carousel are Sparklers. The model has every Sparkler point back to its Bonfire, which these 3 can't do until the article is live, so once it's published, make it the next step for the Sparklers you cut from this interview.
Pick 1 recording you already have, tag 1 topic's worth of passages, and build the matrix before you open the assistant. If 5 pieces feels like a lot for a first pass, make 3 and keep the matrix; part 4 of this series automates the drafting once the manual version works.