Start with an operating brief, not a giant prompt
A one-off prompt can produce a script. It does not create a reliable channel operation. A workflow needs stable channel rules, named inputs, a reviewable output format, decision gates, and a measurement loop. Without those pieces, every upload begins from a blank page and the owner cannot tell what the AI assumed.
ChatGPT and Claude can both help assemble the operation. The specific model matters less than the handoff: it should say what the channel is trying to achieve, which work the AI may prepare, which facts require evidence, and which actions remain human-only.
The six-step review-first workflow
- 01
Define the channel brief
Give the AI one stable source of truth: audience, promise, topic boundaries, primary format, publishing cadence, and the metrics that matter. Do not restart these facts in a new prompt every day.
- 02
Create a source packet
Collect the links, documents, notes, and rights information behind the next video. Ask the AI to separate verified facts, assumptions, and missing evidence before it drafts a script.
- 03
Generate a production packet
Request a structured output: candidate hook, outline, script, shot or asset list, title options, description, and a checklist of claims that still need human verification.
- 04
Review the exact revision
Approve a version only when its revision identifier and contents are visible together. If it is rejected, record a concrete reason such as an unsupported claim, weak hook, wrong tone, or unclear source right.
- 05
Authorize upload separately
Script approval is not account access. Keep channel credentials private and treat the final upload or scheduling action as a separate decision after rights, factual, and platform checks pass.
- 06
Feed results into the next brief
Record exposed metrics such as views, retention, click-through rate, subscribers gained, and comments. Leave unavailable values blank. Use the result to change one meaningful variable in the next production packet.
What the AI should prepare
Planning packet
Topic candidates, audience fit, source list, claim-risk notes, and a recommendation the owner can accept or reject.
Production packet
Hook, outline, script, asset plan, title and description candidates, plus a list of unresolved facts or rights questions.
Review packet
The exact revision, what changed, which rejection reasons were addressed, and the checks required before approval.
Measurement packet
Only metrics actually exposed by the platform, the observation time, and one evidence-based variable to test next.
What should stay human
The owner remains responsible for source rights, factual accuracy, account security, platform rules, and the final decision to upload or schedule. A rights-aware checklist can surface risks, but it cannot guarantee copyright clearance. A performance template can organize results, but it cannot guarantee views or subscribers.
Build it yourself or use a prepared kit?
You can create this structure from scratch with a long system prompt, folders, templates, QA rules, and a small local interface. That is reasonable if you already understand content operations and want to maintain the system yourself.
ChannelOps is the prepared option for someone who already uses ChatGPT or Claude and wants to stop reconstructing sources, decisions, approval state, and next actions across sessions. It combines 28 English files, planning and rights checks, QA gates, measurement templates, and an offline Generate / Approve / Reject console in one download. It does not connect to YouTube or publish on its own.