AI for the Modern World

Module 4 of 10

Module 04: Content Creation with AI

10 min read1,861 words
What you'll learn
Use ChatGPT / Claude / Gemini workflows for writing that respects brand voice and fact discipline.Build a content strategy checkpoint list (audience, channel, CTA, proof, review).Run quality control: hallucination checks, inclusivity, accessibility, and legal sensitivity.Complete hands-on activities that produce portfolio-safe samples (no confidential data).

"Easy reading is damn hard writing." — Nathaniel Hawthorne (AI can speed drafting; you still own clarity and truth)

Opening scenario: Your nonprofit needs a donor letter, six social posts, a short video script, and a presentation deck outline—by Friday. Leadership wants everything on-brand. Volunteers are exhausted. You are not a full-time creative director—but you can orchestrate AI tools like a producer: writing assistants, image generators, and video helpers, with quality control at every gate. This module is for non-coders who must ship credible content.

Learning Objectives

By the end of this module, you will be able to:

  • Use ChatGPT / Claude / Gemini workflows for writing that respects brand voice and fact discipline.
  • Navigate image generation tools (e.g., DALL·E, Gemini/Imagen-class tools, Adobe Firefly—availability varies) with prompt patterns and rights awareness.
  • Outline video projects using AI for scripts, shot lists, and captions—without confusing "generated" for "filmed."
  • Build a content strategy checkpoint list (audience, channel, CTA, proof, review).
  • Run quality control: hallucination checks, inclusivity, accessibility, and legal sensitivity.
  • Complete hands-on activities that produce portfolio-safe samples (no confidential data).

Concept:
Generative tools are junior creatives — they need a brief (audience, goal, proof, voice). Without a brief, you get generic — and generic erodes trust.

1. AI Writing: From Blank Page to On-Brand Draft

Core idea: Treat the assistant as a junior copywriter who needs a brief. Without a brief, you get generic prose.

Brief ingredients

  • Audience (who, what they fear, what they want)
  • Channel (email vs. LinkedIn vs. flyer)
  • Goal (donate, register, understand, attend)
  • Proof (stats you will supply—never invent)
  • Voice (three adjectives; words to avoid)
  • Constraints (length, reading level, compliance)

Real World Example

A nonprofit uses AI for first-draft donor emails but locks impact statistics behind a single finance-approved bullet list pasted into every prompt. No invented numbers — ever.

Real-world examples

  • Lawyer marketing: AI drafts explainer intro; attorney checks every legal claim; marketing checks tone.
  • Teacher: AI generates quiz wrong answers only—teacher writes correct key.
  • Restaurant owner: AI suggests captions; owner supplies today's specials from the kitchen board.
AI ToolWhat It DoesTry It Here
ChatGPT / ClaudeOn-brand drafts from a pasted briefRun the 150-word literacy-night email exercise
Image tools (vendor varies)Mood boards, headers, stylized artThree prompts for one fictional product; compare composition
Video / audio assistantsScripts, captions, shot lists60-second script + human fixes names in captions

Did You Know?
Accessibility (alt text, captions, reading level) is increasingly expected by audiences and sometimes required by policy — AI can draft, humans verify.

Try This Now

  1. Open your preferred chat tool.
  2. Paste a voice guide (even a rough one).
  3. Prompt: "Draft a 150-word email inviting parents to a literacy night. Include CTA: RSVP link TBD. Do not invent statistics."

Table: Editing passes

PassFocus
1Structure and clarity
2Voice and inclusivity
3Fact and link check
4Legal/compliance (if needed)
5Accessibility (headings, contrast for deck)

Discussion prompts

  • Where should AI drafts be watermarked internally so reviewers know the process?

2. Image Generation: Prompts, Realism, and Rights

Tools (examples—verify availability in your region/account): Vendor offerings change; typical categories include chat-integrated image tools, standalone generators, and creative suite features (e.g., Adobe Firefly in Creative Cloud).

Prompt pattern

  • Subject + context + style + lighting + composition + negatives
  • Example: "Editorial illustration, community garden volunteers planting seedlings, morning light, inclusive group, watercolor flat style, no text overlays, avoid logos."

Rights and policy

  • Read terms: commercial use, training opt-out, indemnity limits.
  • Trademarks: avoid prompting protected characters or logos.
  • People: some platforms restrict likenesses—assume consent required for real people.

Real-world examples

  • Publisher: Uses stylized non-photoreal covers to avoid deepfake confusion.
  • HR: Creates generic onboarding graphics—no faces resembling staff.
  • Agency: Keeps human photographers for hero brand assets; AI for mood boards only.

Try This Now

  • Generate three images of a fictional product "EcoBottle" with consistent color palette; note which prompt produced the most usable layout.

Table: When to use AI images vs. stock vs. custom shoot

NeedAIStockCustom shoot
Blog headerOftenYesIf brand-critical
Product packshotRiskyYesYes
News photo of real eventNoMaybeYes

Discussion prompts

  • How will you label AI imagery on your website for trust?

3. Video Creation: Scripts, B-Roll Ideas, and Captions

Pattern: AI excels at scripts, storyboards, caption files, and editing checklists. It does not film your event.

Workflow

  1. Outline in chat: hook, problem, proof, CTA.
  2. Shot list: wide, medium, close-up, B-roll suggestions.
  3. Teleprompter script with short sentences.
  4. Captions: export plain text; use editor tools (CapCut, Descript, Premiere—your choice) for timing.

Real-world examples

  • University: AI drafts 30-second student services spot; students on camera; facts verified by staff.
  • Consultant: AI suggests analogy for explainer; presenter records voice; editor cuts.

Try This Now

  • "60-second script: why volunteer for park cleanup; upbeat; include on-screen text suggestions; no fake statistics."

Step-by-step: caption accessibility

  1. Draft script in chat.
  2. Paste into video tool.
  3. Auto-generate captions; human fixes names and technical terms.
  4. Check reading speed (not too fast).

Discussion prompts

  • When is synthetic voice acceptable in your organization?

4. Content Strategy: Calendars, Pillars, and Measurement

Pillars: Education, proof, community, product, behind-the-scenes—pick three, not ten.

Calendar: AI can propose themes; humans align with inventory, legal, and world events.

Measurement: Pick few KPIs: click-through, conversion, attendance, qualitative feedback.

Table: Channel fit

ChannelAI strengthHuman must own
BlogOutlinesExpertise
EmailDraftsList hygiene
Short videoScriptsAuthenticity on camera

Try This Now

  • "Monthly calendar skeleton for a city library: four posts/week; mix events, literacy tips, staff picks; include risk column."

Scenario: A school posts AI-generated historical scene without label; community confusion ensues. Fix: label synthetic media; link to sources.

5. Brand Voice and Quality Control

Voice doc (one page): mission sentence, do/don't words, three exemplar paragraphs, inclusive language note, legal disclaimers if needed.

QC checklist

  • Truth: every stat traced to source
  • Claims: no guaranteed outcomes unless approved
  • Inclusivity: avoid stereotypes; check representation
  • Accessibility: alt text for images; captions for video
  • Consistency: terminology matches style guide

Try This Now

  • Paste a draft you wrote. Ask: "Flag clichés, vague claims, and exclusionary phrases; suggest fixes without changing meaning."

Discussion prompts

  • Who is the final voice approver when AI is involved?

Activities

  1. Voice lock: Write a 12-line voice guide; generate three pieces; score how on-brand they feel after your edits.
  2. Image triathlon: Same prompt in two image tools; compare composition and artifacts.
  3. Video prep pack: Script + shot list + 10 caption lines for a hypothetical workshop promo.
  4. Hallucination hunt: Ask for five statistics about your city; verify each online—log errors.
  5. Accessibility pass: Take one AI draft; improve headings and plain language for a public notice.

Your Challenge

Build a Content QA One-Pager for your team:

  • Brand voice snippet
  • Pre-publish checklist (10 items)
  • Labeling rule for AI-generated images/video
  • Escalation path when legal PR risk appears

Run one real piece of content through it end-to-end.

Key Takeaways

Try This!
Take one published paragraph you wrote. Ask: “Flag clichés and vague claims only.” Delete any suggestion that changes facts — practice being editor-in-chief.

  • Briefs beat clever prompts for reliable writing.
  • Image and video AI need rights, labeling, and consent thinking—not just aesthetics.
  • Strategy still chooses pillars, timing, and proof—AI proposes; humans decide.
  • Quality control is a multi-pass discipline, not a single click.
  • Accessibility and inclusivity are professional standards, not optional polish.
  • Transparency with audiences builds trust; hiding AI rarely ages well.

Resources

Extended Scenario: Crisis-ish Social Moment

A negative comment thread spikes. Do not let AI respond unsupervised. Use AI to draft three empathetic responses; leader chooses one; comms verifies facts; legal if needed.

Try This Now: Simulate with a fictional brand; compare tone across ChatGPT and Claude.

Comparison Table: Writing Tools for Longform

ToolTypical strengthYour verification
ChatGPTIteration, plugins varyFacts
ClaudeLong pasted contextFacts
GeminiGoogle ecosystemFacts

Deep Dive: Newsletter Production

Monday: AI proposes three story angles from your bullet list.
Tuesday: Draft sections; experts verify quotes.
Wednesday: Line edit by human; subject lines A/B in sandbox.
Thursday: Accessibility check; link test.
Friday: Send; post-mortem metrics captured in one page.

Prompt Starters for Editors

  • "You are a developmental editor. Diagnose structure only—do not change voice."
  • "List 10 stronger verbs than 'utilize' for a nonprofit annual report."
  • "Rewrite for grade-8 reading level without talking down to adults."

Discussion Prompts (Advanced)

  • How do you prevent AI from flattening distinctive organizational voice over time?
  • Should donors know AI helped write thank-you notes?
  • What watermarking or metadata will you store in your DAM?

Rubric: Score an AI Draft in 2 Minutes

Dimension012
Fact riskInvented statsMostly sourcedFully sourced
VoiceGenericCloseOn-brand
ClarityConfusingOKCrisp
CTAMissingWeakClear

When Not to Use Generative Image AI

  • Court exhibits, insurance claims, journalistic evidence
  • Medical education where accuracy of anatomy matters without expert review
  • Any depiction that could harm a real person's reputation

Reflection Journal

  1. My brand's non-negotiable style rule: ___
  2. The riskiest content type we publish: ___
  3. One QC step I will never skip: ___

Appendix: Shot List Template (Copy to Notes)

  • Scene 1 — Hook — wide — 3s
  • Scene 2 — Problem — interview — 10s
  • Scene 3 — Proof — B-roll hands working — 8s
  • Scene 4 — CTA — on-screen text + voice — 5s

Synthetic Media Disclosure — Sample Language

"This image was created with generative AI to illustrate [concept]. It does not depict a specific real person or event."

Adapt with legal review for your sector.

Collaboration Norms for Creative Teams

  • Credit human creative direction explicitly.
  • Version prompts alongside final assets in your DAM.
  • Review AI outputs in pair sessions for sensitive campaigns.

Measuring Creative Quality Beyond Clicks

  • Qualitative feedback from frontline staff
  • Reduction in revision cycles
  • Brand tracker questions (if available)
  • Accessibility audit pass rate

Closing

AI content tools reward clear thinking. The winners are not prompt wizards; they are editors and stewards of trust.

Key Takeaway

  • Briefs (audience, channel, proof, voice) separate pro output from slop.
  • Images and video need rights, consent, and labeling — not just “cool.”
  • Multi-pass QC: structure → voice → facts → compliance → accessibility.
  • Synthetic media disclosure builds trust; hiding it usually backfires.
  • You are judged on truth and taste, not on how fast the first draft appeared.