Classes 57 through 61 gave you the instruments — images, video, voice, audio, music. This class is about conducting them. A complete AI media piece is not five good generations stapled together. It is five generations that agree on audience, promise, proof, tone, and action.

Why complete media fails without a brief

Most AI media projects fail the same way: the script promises one thing, the visual shows another, the voice sounds like a third, and the caption links nowhere. Each piece was generated well in isolation. Together they confuse the viewer.

The fix comes from Class 42 (Specifications) and Class 43 (Plan First, Code Second), applied to media. Before you spend credits on images, voice, and video, write down what every piece must agree on:

• Audience: one specific person, not "everyone interested in AI."
• Promise: the one outcome or insight the piece delivers.
• Proof: the evidence that makes the promise believable.
• Tone: how it feels — calm explainer, energetic launch, dry humor.
• Action: the single thing the viewer should do next.

If two pieces disagree on any of these five, the audience feels it instantly, even if they cannot name it. A calm evidence-led script read in a hype-trailer voice is a disagreement. A thumbnail promising "3 steps" on a video that teaches seven is a disagreement.

The brief: nine fields, one page

A campaign brief is your SPEC.md for media. Keep it to one page. Nine fields are enough:

1. Objective: what success looks like in one sentence. 2. Audience: who it is for, where they are, what they already believe. 3. Message: the one sentence the viewer should repeat afterward. 4. Evidence: the verified claims, numbers, and sources you are allowed to use. 5. Formats: which deliverables you will ship (article, visual, clip, post, link). 6. System: the visual and voice constraints every piece shares. 7. Voice: reading level, pace, banned phrases, pronunciation notes. 8. CTA: the single call to action, stated verbatim. 9. Deliverables / owner / timeline / gates: who makes what, by when, and who approves it.

Here is the shape, using a running example for this class — a 30-second explainer called "What an agent remembers":

# CAMPAIGN-BRIEF.md — What an agent remembers

- Objective: Get solo builders to save the 30s explainer and click through to the memory lesson.
- Audience: Solo builder who has used chat but never built an agent; skeptical of hype.
- Message: An agent is useful because it remembers state between steps.
- Evidence: (1) Agent loop diagram from Class 30; (2) watchlist example from Research Desk — both already published, no new claims.
- Formats: 30s video + thumbnail + caption + link. No blog version in v1.
- System: Dark navy background #0F1B2D, off-white text, one accent (amber); sans-serif; same presenter voice as Class 61 exercise.
- Voice: Plain language, 8th-grade reading level, calm, no "revolutionize / game-changer."
- CTA (verbatim): "See how the loop works — link below."
- Deliverables: script (you, day 1) → shots + voice (you, day 2) → edit + captions (you, day 3).
- Owner: You. Gate: no generation until evidence lines are sourced.

Notice what the brief does: it makes generation cheap because every prompt afterward copies from it. The image prompt copies the system colors. The voice prompt copies the voice line. The script copies message, evidence, and CTA verbatim.

One verified claim, five deliverables

The content-engine idea from Class 66 only works if every repurposed piece traces back to one verified claim. Pick one claim, verify it once, then spin it into five forms:

1. Article: 300 words explaining the claim with its source linked. 2. Visual: one diagram or still that states the claim without new numbers. 3. Clip: a 30-second script that says the claim in spoken language. 4. Post: two sentences plus the visual, ending in the verbatim CTA. 5. Link: the canonical URL where the evidence lives.

For our example, the verified claim is: "An agent repeats think → act → observe → decide, carrying state between steps (Class 30)." That is it. The article explains the loop. The visual draws the four-step loop. The clip narrates it. The post summarizes it. Nothing adds "agents never forget" or "agents are 10x faster" — those are new claims needing new evidence.

This discipline is also a truthfulness guard. AI generators love to embellish. When the brief lists exactly two allowed evidence lines, any third fact in a draft is automatically flagged in review (Lesson 62.3).

The research-story habit

Treat the brief as a research story: every factual sentence in it must survive three questions:

• Where did I see it? Name the source, not "I think I read."
• How strong is it? Published lesson and docs beat a remembered tweet.
• What qualifier does it need? "Often," "in this setup," "for short tasks" — keep the qualifier in the brief so generators do not strip it out.

Write the claim exactly as you will allow it to appear. If the source says "agents can carry state when the system stores it," do not shorten the brief to "agents remember everything." Generators amplify whatever you hand them; a loose brief becomes a false video.

Exercise: write CAMPAIGN-BRIEF.md

Create a file called CAMPAIGN-BRIEF.md for your own 30-second explainer. Constraints that keep it gradable:

• One audience. Name a real person and situation ("freelance bookkeeper triaging receipts on Mondays"), not a demographic.
• Three deliverables minimum. Pick three of: 30s script, thumbnail spec, caption + CTA, visual still, or short article.
• Every claim sourced. Each evidence bullet needs a source you can open right now — a lesson, doc page, or dataset. No "common knowledge."
• Copy-paste voice and CTA. Write the CTA verbatim and a three-line voice spec (level, pace, banned words) that your Lesson 62.2 prompts will reuse.

Finish line: a CAMPAIGN-BRIEF.md with all nine fields filled, one audience, three named deliverables, and at least one sourced evidence line.

Verify fast: hand the brief to someone (or a fresh AI session) and ask, "what should the thumbnail say and what should the viewer do?" If they cannot answer from the brief alone, the brief is not done. Common failure: listing four audiences and three CTAs "to keep options open." Options open at brief stage become contradictions at edit stage.

Check your understanding

1. What five things must every piece of a campaign agree on? 2. Why does the brief list the CTA verbatim instead of letting each generator paraphrase it? 3. Your script cites a statistic not in the brief's evidence section. What should happen? 4. Rewrite "this video is for founders who want growth" as a one-audience brief line.

Next

The brief locks what the campaign says. Next, Lesson 62.2 turns the brief into a pipeline — brief to script to system, assets, voice, shots, edit, captions, and exports — with handoffs clean enough that another AI session can pick up any stage cold.