Jeff Bullas

Nice point — identity stitching and automated UTM normalization are the highest-leverage wins. Nail those and AI becomes a multiplier, not a magic wand. Below is a practical, do-first checklist and a worked example you can run this week.

Do / Do not (quick checklist)

• Do: Capture a hashed email or CRM ID at form submit.
• Do: Start with rule-based attribution to get fast insights, then add ML.
• Do not: Trust last-click alone for budget shifts.
• Do not: Deploy models without explainability (SHAP/LIME).

What you’ll need

• GA4 exports (BigQuery preferred) or CSVs.
• CRM export with lead_time, lead_id, email_hash and conversion flag.
• Environment to run analysis: BigQuery, Colab, or local Python.
• Fields: event_time, clientId/userId, email_hash, campaign/source/medium.

Step-by-step: a practical path you can follow

1. Stitch identities: join on email_hash or userId. Where missing, create probabilistic matches by session timing, user agent and IP proxy. Expect 40–80% match improvement with simple rules.
2. Normalize sources: use an AI-assisted script to suggest canonical mappings (e.g., fb, facebook → Facebook Paid). Review and lock the top 50 mappings.
3. Build touch timelines: for each converted lead, order touches in a 90-day window before conversion. Store touch_index and days_before_conversion.
4. Quick attribution: run a time-decay rule (half-life 7 days) to get fractional credits quickly — this validates major channel patterns.
5. ML step (optional but valuable): train a model (XGBoost or logistic) to predict conversion probability using touch features. Use SHAP to split credit among touches per lead.
6. Integrate: write attributed channel and CPL back to CRM for reporting and budget tests.
7. Monitor: track match rate, attributed CPL, and % direct/unassigned monthly.

Worked example

Lead touches: Organic search (day 0), Email click (day 10), Paid ad click (day 20) — conversion at day 22.

Time-decay (example weights): Organic 15%, Email 25%, Paid 60% → attributed conversion = 0.6 to Paid. An ML model might upweight Email to 35% if historical data shows email drives higher lift before paid converts.

Common mistakes & fixes

• Missing IDs: Fix by adding email_hash at form submit and backfilling where possible.
• Messy UTMs: Use AI to propose canonical names, then lock mappings and backfill historical data.
• Short windows: Test multiple windows (7/30/90) aligned to sales cycle.

Copy-paste AI prompt (use in ChatGPT or your LLM)

“I have two tables: ga4_events(event_time, client_id, campaign, source, medium, event_name) and crm_leads(lead_time, lead_id, email_hash, converted). Join by email_hash and client_id when available. Create ordered touch sequences for each converted lead over 90 days, normalize source strings into canonical channel names, and output BigQuery SQL that returns fractional (time-decay) attribution per touch and the top 10 channels by attributed conversions. Explain each SQL step and show one example output row.”

30/60/90 day action plan

1. 30 days: Export data, stitch identities, run AI-assisted UTM mapping, and run a time-decay attribution on a sample.
2. 60 days: Train a simple ML fractional model, add SHAP explainability, compare results to rule-based.
3. 90 days: Push attributed results into CRM for reporting and run budget tests on 2–3 channels based on new CPLs.

Reminder: Start small, validate on a sample, and iterate. Clean identifiers and UTM consistency are the real game-changers — AI speeds the work and explains the results so business owners can act.

Nice point — repurposing is the heart of cross-platform efficiency. You nailed the concept and the basic steps. Here’s a practical, step-by-step plan to turn that idea into a repeatable system you can run in under an hour a week.

What you’ll need

• One core piece of content (blog post, video, podcast episode).
• A multi-platform scheduler with native posting or reliable API support and basic analytics.
• Simple editing tools (crop/trim, color/logo overlay).
• An AI assistant (chat or built into your scheduler) for captions, hashtags, and variations.
• A place to store post templates and passwords (secure notes + 2FA).

Step-by-step — a simple workflow you can do this week

1. Choose one piece of content and write a one-sentence summary of its main point.
2. List the platforms you’ll post to and the format each needs (e.g., X: short text + link; Instagram: image + 125–150 char caption; Reels: 15–60s vertical clip).
3. Use this AI prompt (copy-paste) to generate variations and a tentative posting schedule:

AI prompt (copy & paste):

“I have a blog post about [insert topic] with the one-sentence summary: ‘[insert summary]’. Create: 3 short captions for X (max 280 chars), 3 medium captions for LinkedIn (professional tone), 3 Instagram captions (conversational with an emoji and CTA), 3 hashtag sets for Instagram (5–8 tags each), and a suggested posting schedule across X, Instagram, LinkedIn and Facebook (times spaced over a week). Also suggest two A/B tests to try.”

1. Edit the AI outputs—keep your brand voice, pick the best captions and hashtags.
2. Create or crop assets: square image for feed, vertical clip for Reels, thumbnail for Facebook/LinkedIn.
3. Batch-schedule with unique captions per platform and set A/B tests (different times or two captions).
4. Track key metrics for 7–14 days: reach, clicks, saves, comments. Record results in a simple sheet.

Example

Take a 3-minute video: extract a 30s highlight for Reels, a 60s cut for Facebook, a static frame with quote for LinkedIn. Use the AI prompt to generate captions and two hashtag sets. Schedule posts across 5 days, test morning vs evening on one platform, compare results after a week.

Common mistakes & fixes

• Mistake: Posting identical copy everywhere. Fix: Adjust length and tone per platform.
• Mistake: Over-automation without review. Fix: Always quick-edit AI suggestions—make them sound human.
• Mistake: No measurement. Fix: Track one or two metrics and iterate weekly.

7-day action plan (quick wins)

1. Day 1: Pick content + run AI prompt.
2. Day 2: Edit captions and hashtags.
3. Day 3: Create assets and resize.
4. Day 4: Schedule posts and set A/B tests.
5. Days 5–7: Monitor, reply to comments, record results.

Keep it small and consistent. The first week is setup; every week after you’ll refine and save hours. Experiment, measure, and let AI handle the heavy lifting while you keep the final, human touch.

5‑minute win: In your Facebook ad editor, open one active ad and add URL parameters so Google Analytics can separate winners from duds. In the URL Parameters box paste: utm_source=facebook&utm_medium=cpc&utm_campaign={{campaign.name}}&utm_content={{adset.name}}&utm_term={{ad.name}}. Save and let it run 48 hours. You’ll instantly see which campaign/ad set/ad drove sales without guessing.

You’re right about the learning phase — let the machines learn. Now add guardrails so they learn the right thing and stop wasting money. The trick is simple: know your profit target, separate “scale” from “tests,” and use AI for creative ideas and weekly triage.

What you’ll need (add these to your list)

• Your average order value (AOV) and rough gross margin %.
• Optional: a CSV export of last week’s ad performance (Date, Campaign, AdSet/AdGroup, Impressions, Clicks, Spend, Conversions).
• 5 product benefits or proof points you can turn into ad angles.

Set your profit guardrail first

1. Break‑even CPA = AOV × Gross Margin %. Example: AOV $35, margin 60% → Break‑even CPA $21.
2. ROAS floor = 1 ÷ Gross Margin. With 60% margin, ROAS floor ≈ 1.67. Anything above that is healthy; below is risky.

Two‑bucket setup (keeps learning clean)

1. Scale bucket (70% of budget): Your best creative + broad audience (Meta) and your proven search/PMax setup (Google). Goal: hit or beat your CPA/ROAS target.
2. Test bucket (30% of budget): New creatives and one new audience at a time. Goal: find the next winner without disturbing your scale campaigns.

Platform basics that save money

• Google: Separate brand from non‑brand. Run a small brand‑only search campaign (exact match) and exclude brand terms from non‑brand campaigns. If using Performance Max, enable Brand Exclusions so non‑brand performance is clear.
• Meta: Use a Sales/Conversions objective with Advantage placements. Start with one broad audience plus one interest/lookalike. Don’t over‑segment on day one.

Creative system (premium shortcut)

• Build a 5×5 angle grid: five angles (problem/solution, social proof, value/price, speed/convenience, gift/occasion) × five formats (product photo, unboxing/texture close‑up, lifestyle, before/after, testimonial).
• Launch three combinations in the Test bucket. Promote winners to the Scale bucket next week.

Budget guardrails (simple rules)

• If an ad set or ad group spends 1.5× your target CPA with zero conversions after 3 days, pause it.
• When scaling, raise daily budget by 20–30% every 48 hours. If CPA rises above target for 3 straight days, revert and test a new creative.
• Avoid edits during the first 7 days of a new campaign (learning window). If you must change, duplicate into the Test bucket so the Scale bucket stays stable.

Example: candles side business ($400/week)

• AOV $35, margin 60% → Break‑even CPA $21; ROAS floor 1.67.
• Budget split: $280 Scale, $120 Test.
• Scale: Google brand search ($30), Google non‑brand/PMax ($150), Meta broad conversions ($100).
• Test: 3 new creatives × 1 new audience on Meta ($60) + 2 new headlines/descriptions on Google search ($60).
• End of week: keep anything under $21 CPA or over 1.8 ROAS. Pause the bottom two, promote the top one, and add one new creative next week.

Copy‑paste AI prompt: Profit guardrail calculator

“I run ads on Facebook and Google for a small online store. My average order value is [AOV], my gross margin is [X%], and last week my Spend/Clicks/Conversions by campaign were: [paste small table or bullet list]. Calculate my break‑even CPA and ROAS floor, identify which campaigns are profitable vs risky, and propose a 7‑day budget split (Scale vs Test) that keeps me above the ROAS floor. Include simple rules for pausing underperformers and a plan to scale winners without resetting the learning phase.”

Copy‑paste AI prompt: CSV triage + negatives (Google)

“Analyze this CSV with columns Date, Campaign, AdGroup, SearchTerm, Impressions, Clicks, Spend, Conversions, Revenue. 1) List the top wasteful n‑grams (1–2 word phrases) by spend with zero conversions. 2) Recommend exact negative keywords to cut that waste. 3) Suggest 3 new ad headlines that align with the converting themes you find. Finish with a one‑week checklist: what to pause, what to scale, and what to test.”

Copy‑paste AI prompt: Creative angle kit (Meta + Google)

“Product: [what you sell]. Customers buy it because [3 benefits] and [2 proofs like reviews or guarantees]. Create a 5×5 angle grid: five angles and five visual ideas. For each angle, write 2 Facebook primary texts (20–40 words), 3 short headlines, and 2 Google Responsive Search Ad headlines/descriptions. Tone: [pick two: warm, premium, playful, urgent]. Flag which 3 combos you expect to have the best click‑through and why.”

Common mistakes & smart fixes

• Mistake: Mixing brand and non‑brand in one Google campaign. Fix: split them; brand hides non‑brand waste.
• Mistake: Judging results mid‑learning. Fix: wait 7 days or at least 50–100 clicks before big calls.
• Mistake: Changing many things at once. Fix: one new creative or one audience per test cycle.
• Mistake: No tracking consistency. Fix: add UTM parameters (Meta) and keep Google auto‑tagging on.
• Mistake: Scaling too fast. Fix: 20–30% budget bumps; monitor CPA drift.

7‑day action plan

1. Day 1: Calculate break‑even CPA and ROAS floor. Add UTM parameters to Meta ads. Split budget into Scale (70%) and Test (30%).
2. Day 2: Create/confirm Google brand‑only search and exclude brand from non‑brand. Launch your 3 test creatives.
3. Days 3–5: No major edits; collect data. Use the Creative angle kit prompt to prep next week’s assets.
4. Day 6: Export CSVs. Run the CSV triage prompt. Pause anything violating guardrails; move saved budget to winners.
5. Day 7: Lightly scale winners by 20–30%. Queue one new creative for the Test bucket.

Keep it simple: protect profit with guardrails, learn with a small test bucket, and let AI handle ideas and weekly triage. That’s how you turn automation into steady, compounding gains.

Level up: your shot list + style lock gives cohesion. Add two power-ups — an identity anchor and a master grade — and your AI set will feel like a real, single-day shoot.

Why this works: most AI “shoots” fall apart when the subject drifts and colors wobble between images. Lock the person and the palette, then make only tiny moves between rounds. That’s how you get a usable series, not just one lucky image.

• What you’ll add
  • Identity anchor: a short, repeatable description that keeps the same subject across all shots.
  • Master grade: a simple color/contrast recipe you paste into every prompt to hold the look.
  • Packaging rules: aspect ratios, naming, and a quick culling rubric so you can deliver assets fast.

1. Build your identity anchor (2–3 minutes)
   • Write 6–10 distinct traits: age range, hair length/color, facial features, skin undertone, one accessory.
   • Include 1–2 recurring wardrobe cues to help consistency (e.g., textured knit, simple hoop earrings).
   • Add “same person across all images” language.
2. Create a master grade (2 minutes)
   • Pick a palette: one base color, one accent, one neutral. Name them in plain English.
   • Define tonal feel in simple words: “soft contrast, gentle highlights, natural skin, no crushed blacks.”
   • State what not to do: “avoid teal shadows, avoid magenta cast.”
3. Run your shoot with three locks
   • Direction (your one-liner), Shot list (from the previous message), Style lock (already defined).
   • Paste your identity anchor and master grade into every prompt. Keep changes tiny round to round.
4. Refine with micro-changes
   • Only one tweak per keeper: crop, warmth, or shadows. Two rounds max.
   • If available, reuse the same random seed or the same top reference each round to steady the look.
5. Package like a pro (10 minutes)
   • Aspect ratios: 4:5 (portrait), 1:1 (square), 16:9 (banner). Export each hero in two ratios.
   • Naming: 01_HERO_L1_C1_F1, 02_TIGHT_L1_C1_F2, etc. Keep the variation codes you already set.
   • Culling rubric: shortlist anything that hits 3 of 4: color match, flattering light, clean composition, on-brief expression.

Copy-paste AI prompt (full generator with identity + grade)

“[Direction one-liner]. Produce a cohesive mini photo shoot using the shot list: 1) HERO waist-up, eye-level. 2) PORTRAIT TIGHT shoulders-up. 3) PORTRAIT CANDID half-turn. 4) DETAIL HANDS. 5) WIDE SCENE with environment. 6) NEGATIVE SPACE composition. 7) VERTICAL SOCIAL. 8) HORIZONTAL BANNER.

Identity anchor: same person across all images; mid-40s; short salt-and-pepper hair; warm brown eyes; light crow’s feet; olive skin undertone; soft natural makeup; textured knit sweater; small hoop earrings; gentle confidence.

Master grade: soft contrast, warm amber bias, natural skin tones, gentle highlights, clean shadows, no teal shadows, no magenta cast.

Use variation codes to label each image: L1/L2 (light), C1/C2 (color), F1/F2 (expression). Keep backgrounds uncluttered. Output 8–12 images. Style lock: soft backlight, warm amber palette, clean skin tones, gentle film grain, 85mm portrait compression, shallow depth of field, subtle vignette. Match the three attached reference images for lighting and palette.”

Copy-paste AI prompt (auto-critique for cohesion)

“You are an art director. Review these 8 images. Score 1–5 for: palette consistency, lighting continuity, subject identity match, composition cleanliness. Identify 3 specific fixes that raise the average by 1 point (micro-changes only: crop, warmth, shadow softness). Then write a one-line refinement prompt for the best 3 images.”

Copy-paste AI prompt (tighten identity if drift appears)

“Ensure the same person appears across all images: mid-40s, short salt-and-pepper hair, warm brown eyes, subtle crow’s feet, olive undertone, small hoop earrings, textured knit sweater. Preserve facial structure and proportions; keep natural skin and minimal makeup. Make no changes to wardrobe or hair.”

Worked example

1. Direction: “Cozy autumn portrait, warm amber, candid.”
2. Identity anchor: mid-40s, short salt-and-pepper bob, soft smile lines, olive skin, knit sweater, small hoops.
3. Master grade: soft contrast, warm bias, gentle highlights, no teal shadows.
4. Round 1: Generate 10 images across the shot list. Label with L1/C1/F1 etc. Flag 4 keepers.
5. Round 2: For each keeper, apply one micro-change: tighter crop (HERO), +7% warmth (TIGHT), soften shadows (CANDID), clean background (NEG SPACE). Export in 4:5 and 16:9.
6. Result: 4–6 images that look like one session, ready for social and banners.

• Insider checks that boost perceived quality
  • Grayscale test: briefly view your set in black & white. If one image is much darker/brighter, fix exposure or shadows.
  • Skin sanity: avoid plastic skin. Ask for “natural texture, gentle grain” in every prompt.
  • Hands and edges: zoom to 100% on hands, hair edges, and jewelry. If odd, rerun that frame with a micro-fix prompt.
  • Negative space: verify room for copy on the NEGATIVE SPACE and BANNER shots before exporting.

• Common mistakes & fast fixes
  • Subject drift: add or tighten the identity anchor; reuse the same reference image.
  • Color wobble: repeat the master grade verbatim; avoid adding new adjectives mid-run.
  • Over-iterating: hard-cap at two rounds; log changes so you learn what actually moves the needle.
  • Cluttered backgrounds: ask for “uncluttered backgrounds, soft blur” and crop tighter on selects.
  • One-off hero only: force the full shot list; don’t let the tool give you 10 similar close-ups.
  • Commercial use: confirm licensing and model-release requirements before publishing.

60-minute action plan

1. Minute 0–5: Write the one-line direction. Draft identity anchor and master grade.
2. Minute 5–10: Pick 3 reference images that match light and palette.
3. Minute 10–30: Run Round 1 (8–12 images) using shot list + style lock + identity + grade.
4. Minute 30–40: Cull with the 3-of-4 rubric; pick 3–5 keepers.
5. Minute 40–55: Run Round 2 micro-changes. Export in 4:5 and 16:9.
6. Minute 55–60: Rename with variation codes; save a one-line log of what changed.

Reminder: consistency beats complexity. Lock the person, lock the grade, run the shot list, and make small, intentional tweaks. That’s how a simple direction becomes a convincing, brand-ready photo set.

Quick win (under 5 minutes): Open GA4’s Traffic Acquisition report, filter for the last 7 days, and export 10 recent sessions. Then pull 10 recent CRM leads and spot-check the source/UTM values. You’ll quickly see where source names don’t match — that mismatch is why attribution looks wrong.

Great observation to focus on GA4 + CRM together — that’s where most attribution gaps happen. Below is a practical, low-friction path to use AI to improve multi-touch attribution across both systems.

What you’ll need

• Access to GA4 event exports (BigQuery recommended) or export CSVs
• CRM data export with lead timestamps and source fields
• A place to run analysis: Google Colab, local Python, or BigQuery SQL
• Basic fields: event time, clientId/userId/email-hash, campaign/source/medium, conversion flag

Step-by-step

1. Stitch identities: match GA4 clientId or userId to CRM leads using hashed email or CRM IDs. If you can’t fully stitch, create probabilistic matches by time and IP/IP proxy.
2. Normalize source data: standardize UTM/source/medium names (AI can suggest mappings).
3. Build a touch timeline per user: ordered list of touch events prior to conversion.
4. Choose an attribution approach: rule-based (first/last/time-decay) for speed, or ML-based fractional attribution for accuracy.
5. If ML: train a model (XGBoost or logistic) to predict conversion probability from each touch. Use SHAP or LIME to assign credit per touch.
6. Evaluate: compare model fractional credits to rule-based results and test against holdout conversions.

Example

For a lead with touches: organic search (day 0), email click (day 3), paid ad (day 6, conversion day 7). A time-decay model might assign 20% to organic, 30% to email, 50% to paid. An ML model could adjust those weights based on historical lift.

Common mistakes & fixes

• Missing identifiers: Fix by capturing hashed emails or CRM IDs at lead form submission.
• Bad UTMs: Use an AI-assisted mapping script to normalize source names before modeling.
• Short attribution windows: Test multiple windows (7/30/90 days).
• Over-trusting last-click: Compare with model outputs and A/B test channel spend changes.

Copy-paste AI prompt you can use right now

“I have two tables: ga4_events(event_time, client_id, campaign, source, medium, event_name) and crm_leads(lead_time, lead_id, email_hash, converted). Join by email_hash and client_id where available, build ordered touch sequences for each converted lead over the last 90 days, and generate a BigQuery SQL query that outputs fractional (time-decay) attribution per touch. Also list the top 10 channels by attributed conversions. Explain each step.”

Action plan (30/60/90 days)

1. 30 days: Stitch identities, normalize UTMs, run rule-based attribution and QA.
2. 60 days: Train an ML model for fractional attribution, add explainability (SHAP).
3. 90 days: Integrate attribution outputs back into CRM for smarter lead scoring and budget allocation.

Reminder: Start small, validate with a sample, and iterate. AI helps with matching, normalization, and explaining model decisions — but clean data and consistent identifiers are the real leverage points.

Right on — separating quick qualitative checks from small paid tests keeps you honest and speeds up learning. Let’s add a premium layer: build a modular “creative brick set” so AI can remix hooks, proof points, and CTAs into fast testable ads.

Before you start — simple guardrails that pay off

• Do: cap voiceover at ~150–165 words per minute (15s ≈ 35–40 words, 30s ≈ 70–80).
• Do: limit on-screen text to 2 lines, ~24–32 characters per line for mobile readability.
• Do: change something every 3–5 seconds (shot, angle, text, sound) to reset attention.
• Do: use a 1:1 map — every script line gets its own frame.
• Don’t: claim results you can’t substantiate. Feed real proof into the prompt.
• Don’t: bury the hook; make the first 3 seconds unskippable and specific.

What you’ll need

• One-paragraph brief (product, audience, pain, main benefit, tone, CTA, 15s or 30s).
• Assets: logo, product images, brand hex colors, verified proof (ratings, count of customers, awards).
• Tools: an LLM for scripts/storyboards, optional image generator for frames, a simple editor or slides for an animatic.

High-value trick: the Creative Brick Set

• Hook bank (5–7 lines): each mentions a time, place, or emotion (“3pm slump?”, “Stiff back after Zooms?”).
• Proof chips (3–5 lines): “4.6★ from 2,300+ buyers”, “Used by 120+ teams”, “Backed by a 30‑day trial”.
• CTA variants (2–3): “Try it for 14 days”, “Get yours today”, “See it in action”.
• Ask AI to recombine: 5 scripts × 3–6 frames × swap hooks/proof/CTA = fast, meaningful variations.

Step-by-step (beat-based so it stays tight)

1. Draft the brief (40–60 words). Add one concrete audience detail (age, context, pain) and one proof item you can actually use.
2. Generate 5 scripts with a word budget, VO tone, and 3–5s hooks. Ask for both 15s and 30s timestamps.
3. Split to scenes: each line becomes a scene with framing, on-screen text, motion, background, and asset notes.
4. Build a shot list: scene number, duration, VO, on-screen text (keep to 2 lines), assets.
5. Create a rough animatic: slides or quick phone shots with temp VO. Check pacing against timestamps.
6. Run the two-step test: 10–20 people for qualitative feedback, then paid micro-test (50–200+ views per variant). Judge on hook hold (0–3s), 15s hold, CTR, and early CPA.
7. Iterate: swap hook/proof/CTA bricks on the winning structure; keep the pacing that worked.

Copy-paste AI prompt — script with pacing and brick set

“You are a senior performance ad writer. Brief: [paste your 40–60 word brief]. Output 5 distinct short video ad scripts for both 15s and 30s. For each script, include: 0–3s hook (specific and emotional), problem line, solution + emotional benefit (1–2 lines), one proof/credibility line, direct CTA. Provide timestamps for 15s and 30s, VO word count (max 40 words for 15s; 80 for 30s), and three alternative on-screen hook texts (each <32 characters, 2 lines max). Keep language clear and human. Deliver in numbered lists.”

Copy-paste AI prompt — storyboard visuals to shot list

“For this script: [paste selected script]. Break into 3–6 scenes. For each scene, provide: visual description, camera framing (close/medium/wide), background/setting, on-screen text (2 lines max, <32 chars/line), motion direction (e.g., push-in, quick cut), suggested SFX or music cue, asset needed (logo/product/hand model), and exact duration so total time matches 15s or 30s. End with a one-paragraph shot list summary I can hand to a videographer.”

Worked example (use as a pattern)

Example brief: “Ergonomic footrest for home-office workers 40–65 who feel leg fatigue after long Zoom days. Benefit: sit more comfortably and feel fresher by 5pm. Tone: warm, practical. Length: 15s. CTA: Try it risk-free for 30 days. Proof: 4.6★ from 2,300+ buyers.”

Example 15s script with timestamps (≈38 words)

• 0–3s Hook: “Legs tired by 3pm?”
• 3–6s Problem: “Hours of sitting strain your legs and lower back.”
• 6–11s Solution + benefit: “Lift your feet to a natural angle. Feel the pressure ease in minutes.”
• 11–13s Proof: “4.6★ from 2,300+ buyers.”
• 13–15s CTA: “Try it for 30 days — free returns.”

On-screen hook options: “Beat the 3pm slump”, “Happy legs, happier you”, “Comfort for long sits”. VO tone: calm, confident.

Storyboard (3 scenes)

• Scene 1 (0–5s, close-up): Under-desk shot of feet fidgeting; quick cut to footrest sliding under desk. Text: “Legs tired by 3pm?” Motion: quick cut, light click SFX. Mood: candid.
• Scene 2 (5–11s, medium): Feet resting at angle; small overlay shows pressure easing (soft gradient). Text: “Ease pressure in minutes.” Motion: slow push-in. Color: warm wood + brand accent.
• Scene 3 (11–15s, wide): User stretches and smiles at desk; logo and star rating appear. Text: “4.6★ | Try 30 days.” Motion: gentle dissolve to end card.

Insider calibration prompts

• Voiceover pace check: “Count words in this 15s script and rewrite to 35–40 words without losing meaning. Keep the same timestamps.”
• Hook stress-test: “Write 10 alternative hooks that name a time/place/emotion relevant to [audience]. Keep under 6 words. Make 3 surprising.”
• Proof sanity check: “Highlight any proof lines that could be misleading or unverifiable. Replace with conservative, accurate wording.”

Common mistakes and fast fixes

• Wall of text → Trim to 2 short lines on screen; move detail to VO or landing page.
• Pretty but slow → Add a pattern interrupt at 3–5s (new angle, overlay, SFX).
• Vague proof → Replace with concrete, sourced facts you can stand behind.
• Mismatch in tone → Specify VO tone and color mood in the prompt (e.g., “warm, natural light”).
• Testing apples vs oranges → Change one variable at a time: hook only, or proof only.

7-day quick plan (pragmatic and light)

1. Day 1: Write the brief, assemble assets, list 2 KPIs.
2. Day 2: Generate 5 scripts using the pacing prompt; pick 2.
3. Day 3: Turn each into 3–6 scene storyboards + shot lists.
4. Day 4: Build rough animatics; check word counts and readability.
5. Day 5: Qualitative check with 10–20 audience-adjacent people.
6. Day 6: Paid micro-test (50–200+ views/variant); record hook hold, 15s VTR, CTR.
7. Day 7: Keep the winning structure; swap in 2 new hooks and 1 new proof chip. Scale modestly.

What to expect: one variant usually wins on hook hold and CTR; keep its pacing and rotate hooks/proof to edge up performance another 10–20%. When you’re ready, paste your one-paragraph brief and I’ll generate the five scripts and two storyboard options to test first.

Nice brief — aiming for ATS compatibility without sounding robotic is the exact sweet spot. That balance wins interviews: machines find your resume, people connect with it.

Quick checklist — do / do not

• Do: Match keywords from the job description, use standard section headings, keep simple formatting, quantify achievements, use active verbs.
• Do not: Use images, headers/footers for critical info, complex tables, or keyword-stuffed, generic sentences that read like a bot.

What you’ll need

1. Your current resume (DOCX or clean text).
2. One or two target job descriptions.
3. An AI writing tool (chat-style or editor) and a plain-text editor.

Step-by-step: make it ATS-friendly, keep it human

1. Extract keywords: copy the job description and ask the AI to list the top 8–12 skills/phrases (hard skills, certifications, tools, core verbs).
2. Map them: mark which keywords you already have on your resume and where they should appear (Summary, Skills, Experience bullets).
3. Rewrite bullets with context + metric + keyword: use the formula (Challenge + Action + Result) and include the keyword naturally.
4. Simplify formatting: single-column, standard headings (Summary, Experience, Education, Skills), plain bullets, standard fonts.
5. Test: paste your resume into a free ATS checker or into the AI and ask for an ATS-scan; correct missing keywords or format issues.
6. Save: submit as DOCX unless the employer explicitly requests PDF (some ATS parse DOCX better).

Worked example

Before (robotic): “Responsible for improving website traffic using SEO best practices.”

After (ATS-friendly + human): “Improved organic website traffic 42% in 9 months by implementing SEO strategy, keyword optimization, and content calendar aligned with target-audience research.”

Mistakes & fixes

• Too many keywords stuffed: fix by weaving 1–2 keywords per bullet naturally.
• Tables or columns: convert to single-column bullets.
• Vague verbs: swap “responsible for” with specific verbs like “led,” “optimized,” “reduced.”

Copy-paste AI prompt (use this directly)

“You are a resume editor. I will paste my resume and a job description. Extract the top 10 keywords from the job description. Then rewrite 5 experience bullets from my resume to include relevant keywords naturally, use active verbs, add metrics where possible, and keep each bullet under 28 words. Preserve truth and avoid buzzwords. Output only the rewritten bullets and a short list of missed keywords to add to the Skills section.”

Action plan (next steps)

1. Run the prompt with your job description and resume.
2. Pick 6–8 bullets to prioritize for tailoring per application.
3. Submit as DOCX and track responses—iterate quickly.

Small edits, targeted keywords, and clear metrics: that’s the quick win. Keep testing and tailoring — your resume should pass the robot and speak to the human recruiter.

Good point: You nailed the key idea — AI is fast at first drafts when you give clear context. I’ll add a practical checklist, a short worked example (one week + one detailed lesson), common mistakes with fixes, and a copy-paste prompt you can use right away.

Do / Don’t checklist

• Do: give grade, lesson length, standards, materials, and student needs.
• Do: start with one week or a single lesson and iterate.
• Do: ask for timings, checks for understanding, and differentiation.
• Don’t: accept the first draft without checking accuracy and pacing.
• Don’t: ask for overly broad output — be specific about format (teacher notes, student handout, sub plan).

What you’ll need (quick list)

• Your curriculum map or unit goals (topics & target standards).
• Grade level and lesson length (e.g., Grade 6 — 45 minutes).
• Materials & tech limits (textbooks, Chromebooks, lab kits).
• Student profile notes (ELLs, IEPs, gifted learners).

Step-by-step (do this now)

1. Paste a one-page summary of the unit into your AI tool.
2. Ask for a weekly breakdown (which standard/topic each day).
3. Request one fully detailed sample day (timed segments: do-now, teach, guided practice, independent work, assessment, homework).
4. Review and edit for accuracy, pace, and materials. Ask for simpler student language or extra differentiation.
5. Export into your planner and pilot in class; refine after classroom feedback.

Worked example — Grade 6 Math, 45-minute lessons (one-week view)

• Day 1: Integer operations — intro & guided practice
• Day 2: Order of operations with integers — practice stations
• Day 3: Word problems — partner tasks + strategy journal
• Day 4: Mini-assessment + targeted reteach groups
• Day 5: Project/apply: real-life budgeting task

Sample detailed Day 2 (45 minutes)

• Do-now (5 min): 3 quick integer problems on whiteboards.
• Mini-lesson (10 min): 2 worked examples modeling order of operations with negative numbers.
• Guided practice (12 min): partner stations — 6 mixed problems; teacher circulates and gives 1-minute checks.
• Independent task (12 min): 4 scaffolded problems — one extension for early finishers.
• Exit ticket (4 min): 2 problems for quick formative check. Homework: 6 practice questions.

Common mistakes & quick fixes

• Too generic activities → Fix: specify examples, numbers, or text excerpts.
• No timing → Fix: require minute-by-minute or block timings.
• Unusable language for students → Fix: ask for student-facing version at a specific reading level.

Copy-paste prompt (use this)

“I teach Grade 6 math. Unit: Integers and order of operations. Lesson length: 45 minutes. Materials: whiteboards, 1 Chromebook per pair, textbook page 88-90. Student needs: 25% ELLs, 2 students with IEPs (extra time), mixed math levels. Create a one-week plan (5 lessons) mapping standards to each day. Then provide a detailed Day 2 lesson: include learning objective, do-now, mini-lesson script (teacher language), guided practice (with 6 specific problems), independent task (4 scaffolded problems plus 1 extension), exit ticket (2 problems), materials list, timings, and differentiation strategies for ELLs and IEPs. Output two versions: teacher notes (detailed) and student-facing checklist (simple).”

Action plan — next 15 minutes

1. Pick one unit and open your curriculum map.
2. Copy the prompt above, paste it into your AI tool, and run it for one week.
3. Review the sample day, tweak timings, and try it in class next lesson.

Reminder: AI speeds up planning — but your classroom judgement makes it work. Start small, iterate, and you’ll get quick wins.