This resource is built for marketing teams, founders and agency producers who already record long-form material, webinars, interviews, podcasts and internal briefings, and who watch almost all of it die in a folder after one publish. It contains a 12 stage production line, 12 paste-ready prompts, a 6 criteria clip scoring model with a defined cut line, a 16 field clip database, 9 production statuses, 3 platform adaptation profiles, 3 human approval gates, 4 named failure checks and one fully worked fictional example.
<aside> ⚠️
Before using these prompts you should omit any PII (personal identifiable information) and check with your compliance team and/or your AI policy.
</aside>
<aside> 🥷
Run this line once on a recording you already have and you will know inside a week whether your content problem was ever a production problem. Talk to UGC Ninja
</aside>
Every number on this page describes the structure of the system itself, not a client's measured results.
There is nothing to download. Every prompt, table and status list on this page is meant to be copied straight off the page into your own tools.
Most teams treat a long recording as a thing to publish. It is not. It is a seam of material with roughly eight to fourteen usable minutes buried inside sixty, and the only question that matters is which minutes, in what order, cut how.
This page describes a production line with 12 numbered stages. Every stage has one input, one output and one condition that has to be true before the material moves on. Nothing is left to taste except the three points where a human is deliberately asked to use theirs.
| Stage | Name | Input | Output | Moves on when |
|---|---|---|---|---|
| 1 | Recording preparation | A booked recording | A recording brief | The brief names 6 to 10 target questions and the speaker has seen it |
| 2 | Capture | The recording brief | A clean master file | Audio is per speaker, framing is stable and the room log is filled in |
| 3 | Transcription | Master file | Timecoded, speaker-labelled transcript | Every speaker turn carries a timecode and the transcript has been spot checked at 5 random points |
| 4 | Topic detection | Transcript | A topic map of 8 to 15 topics | Each topic has a start and end timecode and a one line summary |
| 5 | Clip identification | Topic map plus transcript | 25 to 40 clip candidates | Each candidate has in and out timecodes and a verbatim transcript extract |
| 6 | Clip scoring | Candidates | Scored candidates, ranked | Every candidate has 6 criterion scores and a weighted total |
| 7 | Hook rewriting | Clips above the cut line | 3 hook options per clip | Each hook is speakable in under 3 seconds and is supported by the clip |
| 8 | Platform adaptation | Approved clips and hooks | Per platform cut sheets | Each clip has a named platform set and a reason for every platform it is not going to |
| 9 | Edit spec and B roll | Cut sheets | A shot by shot edit instruction and a B roll list | An editor who was not in the room could execute it without asking a question |
| 10 | Captions, titles, thumbnails | Edited cuts | Caption, title and thumbnail brief per platform | Nothing is generic, nothing restates the hook word for word |
| 11 | Scheduling | Publish ready assets | A dated month schedule | No two clips from the same topic sit within 5 days of each other |
| 12 | Performance review | Published posts plus data | A ranked learning note | The note changes at least one instruction in stage 1, 5, 7 or 10 for the next recording |
Stages 1 to 3 are the ones teams skip, and skipping them is why the other nine feel hard. A recording captured without a brief produces clips that need context. A transcript without timecodes turns clip identification into scrubbing. Fix the front of the line and the back of the line stops being expensive.
<aside> 🥷
If your last recording produced one post and a folder nobody opened, the line is the missing part, not the ideas. Talk to UGC Ninja
</aside>
A recording brief is not an agenda. An agenda organises a conversation. A brief plants the answers you already know you want to cut out, and it does it by asking the speaker to answer in a particular shape: claim first, reason second, example third, and never starting an answer with "so, as I was saying".
Give the speaker the brief at least 24 hours ahead. Ask for three things at the top of the recording: a stated name and role, a one sentence framing of the topic, and permission to use the material. Then never use that opening segment as a clip, it exists to make the rest of the transcript unambiguous.
Prompt 1, recording preparation brief
You are preparing a long-form recording so that it can later be cut into short-form clips.
Context:
Audience: {{audience_description}}
Speaker and role: {{speaker_and_role}}
Recording format: {{format_interview_or_webinar_or_podcast}}
Planned length in minutes: {{planned_length_minutes}}
Commercial goal of the content: {{commercial_goal}}
Topics the business wants to be known for: {{priority_topics}}
Things the speaker must not say for legal or commercial reasons: {{restricted_topics}}
Produce a recording brief with these parts and nothing else:
1. Ten target questions. Each one must be answerable in 40 to 90 seconds and must
invite a specific claim rather than a description of a category. Reject any
question that could be answered with a list of generalities.
2. For each question, one line telling the speaker the shape of the ideal answer,
in the form: lead with the claim, give the reason, give one concrete example.
3. Five "sharpening follow-ups" the interviewer can use when an answer stays abstract.
4. A list of six phrases the speaker should avoid because they make a clip
dependent on earlier context, for example back-references and pronouns with
no antecedent.
5. A 60 second opening script for the speaker to read that states name, role and
topic framing, to be used for transcript clarity only and never published.
6. A room and capture checklist covering audio per speaker, framing, lighting
consistency and anything that would make a vertical crop fail.
Write in British English. Do not use em dashes or en dashes.
Output as plain text under numbered headings.