Most repurposing advice starts with a tool that turns one video into forty social posts, which gets you forty slightly different ways of saying nothing. AI content repurposing works better as an editing job: one substantial source, one business objective, and a deliberate decision about what each channel needs from that source. The model drafts; you decide what is true and what gets cut.
This guide follows a single example from transcript to blog outline, email, LinkedIn post and short video script, so you can see what changes between channels and what has to stay identical. It covers the editorial transformation only. The plumbing that moves files between apps is a separate decision, and our guide to choosing a first workflow to automate helps you judge whether that is worth building.
The example: one video, one objective
Hollow Oak Joinery is an illustrative business invented for this article: a two-person custom kitchen workshop. The owner recorded a 12-minute video on the question customers ask most, why his quote is so far above the flat-pack price online. The transcript below is illustrative too, deliberately a bit rambling, because real transcripts are.
Before opening any AI tool, write down one objective for the whole set. Hollow Oak’s is: get more people who receive a quote to book a free site visit instead of disappearing. Without it, every draft drifts toward whatever its channel finds interesting, and nobody books anything.
Illustrative source excerpt
[02:10] So the first thing people see is the number, and it is usually about three times what they saw on a flat-pack website. But the flat-pack price is for boxes. Our quote is for the room. We measure every wall, because no wall in a house built in the sixties is straight. [03:05] Last spring we had a customer, Anna, who had already bought a flat-pack kitchen and then called us because nothing fitted. The corner unit was off by about four centimetres. We ended up scribing the panels by hand to the wall. That fix is in our quote from day one. [04:40] The other big chunk is the doors. Solid oak or painted MDF, that is a real difference, maybe a third of the cost. If the budget is tight, keep the carcasses and change the door material. Don’t cut the hinges, the hinges are what you touch every day. [06:15] And honestly the best thing anyone can do is let us come out and look. The site visit is free, it takes about forty minutes, and nine times out of ten we find a way to bring the number down.
The excerpt already contains a named customer (a passing mention in a video is not consent to appear by name in an email campaign), several numbers that are really estimates, and a call to action buried at 06:15. The process has to handle all of it.
Extract before you write anything
Paste a transcript and ask for “a blog post, an email and five LinkedIn posts,” and the model will round estimates into facts and invent connective tissue that was never in the video. So the first pass pulls the transcript apart into claims, examples, audience problems and evidence, each with an ID and timestamp, so every later sentence can point back to its source. That is all source traceability is.
You are helping a small business turn one video transcript into several pieces of content. Business: [one line] Objective for all outputs: [the single objective] From the transcript, extract: 1. CLAIMS stated as fact or advice: ID (C1, C2...), timestamp, short quote. 2. EXAMPLES and stories: ID (E1, E2...). 3. AUDIENCE PROBLEMS the speaker is answering: ID (P1, P2...). 4. EVIDENCE STATUS for each claim: estimate, specific fact to check, or opinion. Quote the transcript. Do not strengthen claims or add facts, numbers or examples. Flag any named person or company. If something is unclear, say so. Transcript: [paste]
A cleaned-up extraction for the excerpt looks like this (sample output, edited):
| ID | Item | Time | Evidence status |
|---|---|---|---|
| C1 | Custom quotes are often around three times the flat-pack price | 02:10 | Estimate; keep the hedge |
| C2 | Older houses rarely have straight walls, so every wall is measured | 02:10 | Opinion from experience |
| C3 | Door material can be roughly a third of the cost | 04:40 | Estimate; owner to confirm |
| C4 | Change door material before cutting hardware quality | 04:40 | Advice |
| C5 | Site visit is free and takes about 40 minutes | 06:15 | Fact; confirm before publishing |
| C6 | Most site visits find a way to reduce the quote | 06:15 | Estimate; never turn into a statistic |
| E1 | Flat-pack corner unit 4 cm off; panels scribed by hand | 03:05 | Named customer: anonymise |
| P1 | Price shock when the quote arrives | 02:10 | Core problem |
| P2 | Fear of hidden extras later | 03:05 | Supporting |
| P3 | Tight budget, unsure what to compromise on | 04:40 | Supporting |
That table is the real asset. Every channel draft is built from it, never from the raw transcript or the previous draft. A draft sentence with no ID behind it is either connective writing, which is fine, or something the model made up.
Watch C6. “Nine times out of ten” is how people talk; “90% of site visits reduce costs” is a statistic nobody counted.
What changes by channel and what stays fixed
Channels differ in length, reader attention and realistic next step. The facts do not, so brief the model on the two separately.
| Blog article | LinkedIn post | Short video | ||
|---|---|---|---|---|
| Reader | Searching for a full answer | Has a quote, deciding | Scrolling, not shopping | Scrolling, seconds of patience |
| Job | Explain the quote properly | Get the visit booked | Make one idea stick | One surprising point |
| Material | Nearly everything | P1, P2, C5, C6 | C2, E1 | C4 |
| Length | As long as the answer needs | Under 200 words | Far below LinkedIn’s 3,000-character limit | 30 to 45 seconds, although YouTube allows Shorts up to three minutes |
What stays fixed in all four: the numbers and their hedges, the site visit details, the door advice and the anonymised story. If the email says forty minutes and the video says half an hour, a customer who sees both will wonder which one is the sales pitch.
Voice should not move either. If you have written brand rules, paste them into every prompt. If you do not have any yet, our piece on AI prompts for brand voice and identity is the better first step.
One source, three different adaptations
The same material as three illustrative drafts after one round of human editing. Read across a row: the facts stay put while almost everything else moves.
| Source | Email (customer with a quote) | LinkedIn post (owner) | Short video script |
|---|---|---|---|
| “The flat-pack price is for boxes. Our quote is for the room. We measure every wall.” (02:10) | Subject: About the number on your quote
Hi [name], if the figure made you wince, you’re in good company. Most people compare it to a flat-pack price, which leaves out what we’re actually pricing: your room, measured wall by wall, including the ones that aren’t straight. |
The most expensive part of a kitchen is usually invisible: the wall that isn’t quite straight.
Flat-pack pricing assumes a perfect room. I’ve measured a lot of rooms and haven’t met that one yet. |
[0:00] Spirit level on a wall. VO: “This wall is not straight. Neither is yours.” [0:04] Owner to camera: “That’s why custom kitchens cost more than the box price online.” |
| Corner unit 4 cm off; panels scribed by hand. (03:05) | We often hear from people whose flat-pack corner unit didn’t fit. Hand-fitting to uneven walls is already in your quote, so it won’t show up later as an extra. | Last spring a customer called after buying a flat-pack kitchen. The corner unit was about 4 cm off, so we scribed the panels to the wall by hand.
That work is in every quote we send. Most people only learn it’s needed after the boxes arrive. |
(Not used: the story needs more than five seconds.) |
| “Change the door material. Don’t cut the hinges.” (04:40) | If the budget is tight, door material is usually where the real savings are. We’d rather change that than skimp on hinges you use every day. | (Not used: it would turn a story post into a tips post.) | [0:08] Owner opening a door: “Budget tight? Change the doors, not the hinges.” [0:12] “You touch the hinges about ten thousand times.” [NEW: needs approval] [0:16] On screen: “Free site visit. Link in profile.” |
| “The site visit is free, about forty minutes, nine times out of ten we bring the number down.” (06:15) | The easiest next step is a free site visit. It takes about 40 minutes, and once we’ve seen the room we very often find a way to bring the price down. [Book a time] | If a kitchen quote ever made you wince, that wall is usually why. | (Covered by the on-screen text.) |
The empty cells matter. The LinkedIn post drops the door advice because one idea beats a list, and the short video drops the customer story because it cannot land in five seconds. Deciding what to leave out is most of the editorial work.
Then look at the flagged line in the video script. “Ten thousand times” is not in the transcript. It is the kind of line a model adds for punch, and the traceability check exists to catch it. Either the owner accepts it as a fair rough figure and says it in his own words, or it goes. Anna, meanwhile, became “a customer,” the right default without her consent.
The blog outline
The blog uses nearly the whole extraction table, because a searcher wants the full answer. Illustrative output after editing:
| Section | Content | Source IDs |
|---|---|---|
| H1: Why your custom kitchen quote is higher than the flat-pack price | The price shock moment, acknowledged as reasonable | P1, C1 |
| H2: You are comparing boxes with a room | What flat-pack prices leave out; measuring every wall | C2 |
| H2: Fitting problems you pay for either way | Anonymised corner-unit story; scribing quoted upfront | E1, P2 |
| H2: Where the money goes | Door material as the big variable, clearly hedged | C3 |
| H2: If you need to cut the budget | Change doors, keep hardware quality | C4, P3 |
| H2: What happens at a site visit | Free, about 40 minutes, why quotes often come down | C5, C6 |
| Embed | The original video | Source |
Embedding the video keeps the article honest and serves readers who would rather watch. Google’s guidance on using generative AI content on your website centres on usefulness and originality, which a real expert recording with traceable claims supplies.
Prompts for each channel
Each prompt takes the extraction table as input, names the reader, job, material and length, and asks the model to account for its sentences. If briefing a model like this is new to you, our beginner’s guide to prompt engineering for small businesses covers the basics of context and constraints.
Write a follow-up email from [business] to a customer who got a quote last week and has not replied. Objective: book a free site visit. Use ONLY these items: [P1, P2, C5, C6]. Keep every number and hedge exactly as written. No customer names, statistics, guarantees or discounts. Under 200 words, warm, plain. Then list each factual sentence with its source ID.
LinkedIn post
Write a first-person LinkedIn post for the owner of [business]. Voice rules: [paste]. One idea only: [C2], told through [E1], anonymised. The reader is not shopping. No hashtag wall, no emoji bullets, no "Here's what I learned" opener, no sales pitch. Then list any sentence not supported by C2 or E1.
Short video script
Write a 30 to 45 second vertical video script for [business], filmed by the owner on a phone. One point: [C4]. Timestamped lines: what is on screen, what is said. Open on a visual, not a greeting. Every line must be sayable in one take. Mark any new fact, number or claim with [NEW: needs approval].
The self-accounting line at the end does not make the model reliable (models mislabel their own output), but it gives the reviewer a list to check.
The review stage
Someone who knows the facts, ideally the owner, reads each draft with the extraction table open. For a set like this that is roughly fifteen minutes, still far quicker than writing four pieces from scratch.
| Check | Pass when |
|---|---|
| Traceability | Every factual sentence maps to an ID; anything new is cut or explicitly approved |
| Hedges | “About”, “often” and “roughly” have not become precise figures or promises |
| Consistency | Visit length, price claims and advice match across all drafts |
| Permission | No customer named without consent, no private details carried into writing |
| Voice | The owner would actually say it out loud |
Where it usually goes wrong
Sameness is the most common failure. Without separate briefs you get one piece in four lengths. If two drafts could swap channels without anyone noticing, one of them is not doing its job. Then there is drift, where each adaptation moves a little further from the source until “nine times out of ten” becomes “we always save you money.” Building from the table, not the previous draft, prevents most of it.
Thin sources and sheer volume are the other traps. A video without a specific point produces four pieces without one, and asking the model to fill the gap is how invented statistics reach your website. Google’s spam policies describe scaled content abuse as generating many pages primarily to manipulate rankings rather than help users, however they are made. Four good pieces from one good video is a solid week.
Getting a usable transcript
If the video is on YouTube, automatic captions give you a starting point, and you can open the transcript on any video that has captions. Automatic captions mishear names, product terms and numbers; fix the transcript before extraction.
For a recorded experiment paired with a written breakdown, see how a website concept was built with ChatGPT and Claude Code.
To start, keep it small: one recording that answers a real customer question, one objective, an extraction table fixed by hand, and two adaptations rather than four. Direktorium keeps covering this kind of practical AI work for small businesses, and if you would rather have a team build a repurposing process around your own content, BBDirector can help.
