A source-checking workflow for creator drafts

A lean creator workflow starts with a deliverable and an evidence trail, not a long shopping list. This tutorial turns one small source packet into a social post, newsletter opening and clip brief while keeping unsupported claims out of the result. You can do the exercise with a text editor, a calculator and a video editor you already know. AI drafting is optional.

Method and limits: The source packet, deliberately flawed draft and corrected outputs below were authored for this tutorial. They are synthetic, not customer records, vendor outputs or results from a hands-on product test. Official documentation was read on 17 October 2026 by an AI assistant. Human editorial review and individual human authorship are not documented.

1. Define the deliverable before choosing tools

Write down the audience, length and decision the content should support. For this example: invite facilitators to try an outline exercise, using a social post under 50 words and a short excerpt from an owned recording. Do not turn a small feedback survey into a claim about long-term productivity.

Keep three separate files: the source packet, an editable draft and a claim ledger. Add a video editor only if you actually need a video deliverable. Before using an external AI service, check whether you may upload the material and whether its current access, retention and export terms fit the task. A directory listing is not an entitlement to a vendor's features.

The Moyan AI Tool Lab is a discovery directory. This workflow does not upload a video to Moyan or extract clips there. Any actual video editing happens in an editor you select separately; a written clip brief is not an edited media file.

2. Build a small source packet

Everything in this packet is fictional. The timestamps are authored placeholders, not timing measurements from a recording.

B1 — Attendance: 12 participants attended a 60-minute outline workshop.
B2 — Feedback: 9 participants returned the form; 7 marked the outline exercise useful or very useful.
B3 — Limits: No comparison group and no later output or time-saved measurements.
B4 — Consent: The sample contains no real people or real recording. Actual reuse permission is not established by this exercise.

T1 00:00–00:19 — Facilitator introduces the outline exercise.
T2 00:20–00:34 — Facilitator: Seven of the nine people who returned the feedback form found the exercise useful or very useful.
T3 00:35–00:44 — Facilitator: We did not measure whether it changed their later work.
T4 00:45–01:00 — Facilitator asks viewers to try the exercise on one draft.

For real work, replace these labels with document names, pages or exact timestamps. Keep the original recording and check the transcript against it. YouTube's automatic-caption guidance warns that automatic captions can misrepresent speech and tells creators to review and edit them. A clean-looking transcript is not proof that names, numbers or qualifiers were transcribed correctly.

3. Audit a deliberately flawed draft

This draft is an authored error specimen, not output from a tested AI tool:

Our workshop improved productivity by 40%. Nearly 80% of all attendees proved the method works. Upload your video to Moyan Lab and let it automatically extract the winning clip.

The arithmetic is 7 / 9 × 100 = approximately 77.8% of respondents. Dividing by all 12 attendees gives 58.3%, but that does not tell you how the three nonrespondents felt. Neither percentage measures productivity. For this small sample, the count “7 of 9 respondents” is clearer than a rounded headline percentage.

A source-grounded draft can still omit a qualifier, choose the wrong denominator or attach a citation to a statement it does not support. Google's notebook source documentation describes using imported sources; its problem-reporting guidance explicitly accommodates inaccurate responses. Using your documents does not eliminate hallucinations. Open the cited passage and check the actual claim.

4. Produce three bounded outputs

These are manually prepared reference answers for the synthetic packet, not captured vendor outputs.

Social post — 40 words:

We tried an outline exercise in a 60-minute workshop with 12 participants. Seven of nine feedback respondents found it useful or very useful. That is feedback, not proof of improved productivity. Try outlining one draft and review what changes yourself.

Newsletter opening:

A small workshop gave us a reason to explore an outline exercise further: 7 of 9 feedback respondents rated it useful or very useful. Twelve people attended. We did not measure later work, so we cannot say whether the exercise saved time or improved output.

Manual clip brief:

Candidate interval: 00:20–00:44, covering T2 and T3.
Purpose: Present the feedback count together with the measurement limit.
On-screen caption: “7 of 9 feedback respondents found it useful; later work was not measured.”
Before export: Confirm actual in/out points in the real recording, retain the complete caveat, check captions and reuse permission.
Status: Planning note only. No video was uploaded, cut or exported.

Do not cut at 00:34 merely because the positive sentence stands alone. In this example that removes the adjacent limit. On real footage, play several seconds before and after each proposed boundary. Check the final export, not only the editing timeline. No “best clip” or engagement outcome has been measured here.

5. Use AI only where you can check its contribution

If you choose to draft in an external tool, give it a bounded request like this. The prompt itself does not guarantee compliance.

Use only B1–B3. Draft a social post under 50 words.
Preserve the distinction between attendees and feedback respondents.
Do not infer productivity, time saved, causality or a comparison group.
Return a claim ledger linking each factual sentence to a source label.
If a requested fact is missing, mark it unsupported rather than filling it in.

Review each factual sentence yourself. Treat a generated ledger as a list of claims to inspect, not as a completed verification report. If two passages conflict, retain both and resolve the conflict before publishing. A citation is useful only when the cited material supports the precise statement, population and time period.

For a separate exercise, try Moyan's same-task comparison lab with one of its built-in source packets. It accepts pasted outputs and manual judgments for that selected exercise; its export includes that packet, not this tutorial's creator packet. It does not call a model or automatically choose a winner. Keep this tutorial's outputs and claim ledger in your own files. For source checking beyond this exercise, see Verify AI output.

6. Try a changed-input exercise

Change B2 so that all 12 participants returned the form but only 7 rated the exercise useful. What changes?

The correct count becomes 7 of 12 respondents, or approximately 58.3%. The old “7 of 9” social post and caption must be updated. B3 still prevents a productivity claim. This catches a common editing failure: changing the source without rechecking every reused output.

Before a real release, record which source version you used, which claims you removed, who actually checked the final output and whether permissions are resolved. Leave an unknown field unknown. Do not use this tutorial's synthetic figures as evidence in your own promotion.

Correction history and source scope

17 October 2026 — This replacement removes the claim that source grounding eliminates hallucinations and the instruction to upload video to Moyan for automatic clip extraction. It replaces a tool-stack pitch with an authored, checkable workflow. This is the preparation/source-check date, not proof of a live deployment. The original article URL and database publication date are retained.

The three official documentation links above were read on 17 October 2026. They support the narrow source/caption limitations described here, not a product ranking or a hands-on performance claim. No paid AI calls, model evaluations or video processing were used to prepare this example.