Stop Losing Viewers at Second Three: A Claude 3.5 Workflow for Product Video Scripts
You filmed the product, edited the cut, and added trending audio. Then the analytics show a brutal drop-off at second two. For TikTok and Reels creators running e-commerce ads, this is the most expensive problem in short-form video. If the first 3 seconds don’t earn attention, nothing after them gets seen, including your demo, your offer, and your CTA.
The usual fix is to write one hook, hope it works, and start over when it doesn’t. A better approach is to build a repeatable Claude workflow for e-commerce product video script writing that generates many hooks across proven psychological angles, scores them, and expands only the strongest into full scripts with timing, visuals, and on-screen text.
The master prompt below does this in a single run with Claude 3.5.
Table of Contents
Why This Prompt Works
The prompt is built around how short-form ads succeed or fail:
- It separates hook ideation from script writing. Claude generates a range of hooks first, so you aren’t stuck with its first idea.
- It forces distinct hook frameworks. Pattern interrupt, problem callout, curiosity gap, contrarian claim, and the others each trigger a different kind of attention. You get real variety instead of ten rewordings of one idea.
- It treats the hook as audio, visual, and text together. Scrollers react to the first frame and the on-screen text before they process speech, so each hook specifies all three.
- It includes a scoring and rewrite step. Claude critiques its own hooks against clear criteria and rewrites weak ones before you see them.
- It uses XML-style tags. Claude 3.5 follows structured, tagged instructions very reliably, which keeps the output consistent from run to run.
- It sets guardrails. Banned clichés and a no-invented-claims rule keep the scripts fresh and ad-compliant.
The Master Prompt
Copy the prompt below into Claude 3.5 and replace every item in [BRACKETS] with your own details.
You are a senior direct-response creative strategist who writes high-retention short-form video ads for TikTok and Instagram Reels. You specialize in e-commerce products and in winning the first 3 seconds.
</role>
<product_info>
Product name: [PRODUCT NAME]
Product category: [CATEGORY]
Price and any current offer: [PRICE / OFFER]
Key features and benefits: [3-5 FEATURES AND THE BENEFIT OF EACH]
Main problem it solves: [CORE PROBLEM]
Target customer: [WHO THEY ARE, AGE, LIFESTYLE, WHAT THEY CARE ABOUT]
Objections or doubts buyers have: [COMMON OBJECTIONS]
Proof available: [REVIEWS, RESULTS, DEMOS, UGC, GUARANTEE. ONLY LIST WHAT IS REAL]
Brand voice: [E.G. CASUAL, FUNNY, PREMIUM, BLUNT]
Platform: [TIKTOK / REELS / BOTH]
Video length target: [15 / 20 / 30 SECONDS]
Format: [FACE-TO-CAMERA / VOICEOVER + B-ROLL / UGC REVIEW / POV / DEMO]
</product_info>
<task>
Complete these steps in order:
STEP 1 - HOOK GENERATION
Write 10 different hooks for the first 3 seconds. Use at least 6 different frameworks from this list:
- Pattern interrupt (unexpected visual or action)
- Problem callout (names the viewer's exact frustration)
- Curiosity gap (opens a loop the viewer needs closed)
- Contrarian or myth-busting claim
- Bold result or transformation preview
- Social proof or "everyone is talking about this"
- Direct audience call-out ("If you [specific situation]...")
- Mistake or warning ("You're doing [X] wrong")
For each hook, provide:
- Framework used
- Opening visual (what is on screen in the first frame, including motion)
- On-screen text (maximum 8 words)
- Spoken line (maximum 12 words, deliverable in under 3 seconds)
- Why it should stop the scroll (one sentence)
STEP 2 - SELF-CRITIQUE
Score each hook from 1-5 on: (a) pattern interrupt strength, (b) clarity within 1 second, (c) curiosity created, (d) relevance to the target customer. Rewrite any hook that scores below 4 on any criterion. Show the final scores in a compact table.
STEP 3 - FULL SCRIPTS
Choose the 3 highest-scoring hooks. For each, write a complete script for the target video length using this structure:
- 0-3s: HOOK
- 3-8s: PROBLEM / AGITATION (make the pain or desire feel real)
- 8-18s: SOLUTION / DEMO (show the product doing the job, features framed as benefits)
- 18-24s: PROOF / OBJECTION HANDLING
- 24-30s: CTA (one clear action and a reason to act now)
Adjust timings proportionally if the target length is shorter.
Present each script as a table with these columns: Time | Visual / B-roll | Spoken Line | On-Screen Text | Sound / Edit Note.
STEP 4 - EXTRAS
For each script, add:
- One caption (under 150 characters) with 3 relevant hashtags
- One alternate CTA
- One A/B test suggestion that changes only the hook
</task>
<rules>
- Do not invent statistics, reviews, results, or guarantees. If a claim would need proof I have not provided, mark it with [VERIFY].
- Avoid overused openers such as "Stop scrolling," "You won't believe," "Game changer," and "Wait for it" unless you give them a fresh twist.
- The first frame must contain motion or a visual surprise, never a static product shot or a logo.
- The spoken hook must begin within the first 1.5 seconds. No greetings or intros.
- Use simple, conversational language at a 6th-8th grade reading level.
- Write in the specified brand voice.
- Keep the content compliant with TikTok and Meta ad policies: no unrealistic medical claims, no misleading before/after implications.
</rules>
<output_format>
Use clear headings for each step. Keep hooks in a list, scores in a table, and scripts in tables. Finish with a one-paragraph recommendation of which script to test first and why.
How to Customize and Use It
Fill in the product details
The quality of the output depends on how specific your inputs are. Instead of “portable blender,” write “a 20-oz USB-rechargeable blender that blends a smoothie in 30 seconds and fits in a gym bag.” Claude can’t write a sharp hook from a vague product description.
Be honest about proof
The Proof available field controls what Claude is allowed to claim. If you have 400 five-star reviews or a strong before-and-after demo, say so. If you have nothing yet, write “none,” and Claude will build the scripts around demonstration and curiosity.
Match the format to your style
Choose one format. A face-to-camera hook is written very differently from a silent POV demo or a voiceover over b-roll. If you aren’t sure, run the prompt twice with different formats and compare the hooks.
Adjust length and platform
For 15-second videos, the middle sections compress and the hook needs to be even tighter. For Reels, you can lean slightly more polished. For TikTok, lean raw and native. You can add a line such as “Prioritize a lo-fi UGC feel” to the format field.
Tune the scoring criteria
If you’re running cold traffic, add a criterion such as “works with sound off.” If you’re retargeting warm audiences, replace “clarity” with “objection handling.”
Best Practices and Tips for Claude 3.5
- Iterate on the hooks, not the whole script. After the first run, reply with something like: “Give me 10 more hooks in the style of #4 and #7, but for a skeptical audience.” Claude 3.5 keeps context well, so you can refine without restating the product details.
- Feed it real customer language. Paste 5-10 real reviews, comments, or DMs into the product info and add “Mirror the phrases my customers use to describe the problem, but don’t copy full sentences.”* Claude then writes hooks in words your buyers already use, so the problem callout sounds like recognition instead of advertising. Pull this language from product reviews, competitor comments, and your own support inbox. If you already collect customer questions, the Claude workflow for customer service FAQ generation shows how to turn them into a clean objection list for the product info block.
Ask for a sound-off pass. Cold-traffic viewers often scroll with the sound off. Add: “Rewrite my top 3 hooks so they work using only the first frame and the on-screen text.”
Verify every [VERIFY] tag before filming. Claude flags any claim your inputs don’t support. Either supply real proof or delete the claim. Never leave it in the script.
Save the hooks that win. Keep a simple sheet with the hook, framework, product, and result. Paste your top five into future runs with: “Write new hooks in the style of these winners.”
Run one product per prompt. Mixing products dilutes the specifics Claude needs to write sharp hooks.
Example: What One Run Produces
Below is an abbreviated output for the portable blender example, with no proof provided. It is an illustration of the format, not performance data.
Hooks (2 of 10)
Framework
Opening visual
On-screen text
Spoken line
Scores (a/b/c/d)
2
Direct audience call-out
Hand shakes a shaker bottle, clumps visible
Clumpy shake again?
If your shaker bottle always leaves clumps, watch this.
4/5/4/5
6
Pattern interrupt
Blender dropped into a gym bag, whirring from inside
Smoothie. Gym bag. 30 seconds.
I made this smoothie inside my gym bag.
5/4/5/4
Script built on hook 2 (30 seconds)
Time
Visual / B-roll
Spoken Line
On-Screen Text
Sound / Edit Note
0-3s
Shaker bottle shaken, clumps visible
If your shaker bottle always leaves clumps, watch this.
Clumpy shake again?
No music, hard cut on the shake sound
3-8s
Close-up of powder stuck to the bottle wall
Shake, wait, still clumpy. Drink it anyway.
Every. Single. Time.
Quick cuts, light beat
8-18s
Ingredients in, one press, timer overlay
Add everything, press once, thirty seconds later it’s smooth.
30 seconds · USB-rechargeable
Blender sound up
18-24s
Pour the drink, rinse the cup
No clumps, and it rinses in seconds.
Rinse. Done.
Music dips
24-30s
Product in the gym bag, price overlay
Tap the link to get yours. [VERIFY OFFER]
Link in bio
CTA text holds 3 seconds
Notice that the script has no reviews or statistics. With “Proof available: none,” Claude built it entirely on demonstration, exactly as the rules in the prompt require.
How to Test Your Scripts
Generating scripts is half the workflow. The other half is learning which hook actually worked.
Film all three scripts with the same product shots, so only the hook and script differ.
Run them to the same audience with the same budget and offer.
Change only the hook between variants, using the A/B suggestion Claude gave in Step 4.
Read retention before sales. Check the share of impressions that watch past the first three seconds (often called hook rate) and average watch time, then click-through rate and cost per purchase.
Feed the results back to Claude. The next run should learn from the last one:Here are the results from the last test. Hook A: [3-SECOND VIEW RATE, AVG WATCH TIME, CTR] Hook B: [3-SECOND VIEW RATE, AVG WATCH TIME, CTR] Hook C: [3-SECOND VIEW RATE, AVG WATCH TIME, CTR] Explain the most likely reason the winner held attention, then write 5 new hooks for [PRODUCT NAME] that apply the same principle. Do not change the offer.
When the Output Falls Flat
The hooks all sound alike. Add to the task: “Each hook must use a different framework and a different opening visual. Reject any that repeat.”
The script is too long for the video. Spoken delivery runs at roughly 2.5 words per second, so a 30-second script needs about 70-80 spoken words. Add that limit to the rules.
The tone doesn’t sound like your brand. Paste one script you already like into the prompt and add “Match the pacing and word choice of this example.”
Frequently Asked Questions
Can Claude write TikTok and Reels product video scripts?
Yes. Claude can write complete scripts with timing, visuals, on-screen text, and a CTA, as long as you give it specific product details and a defined format. It writes the script, but it does not film or edit the video.
How many hooks should I test per product?
Start with the three scripts the prompt builds, each using a different hook, and test them against each other. Generate new hooks only after you know which framework held attention.
Does this prompt work in ChatGPT or Gemini?
The structure is plain text with tags, so you can paste it into other assistants. Output quality and how strictly they follow the steps will vary, so test it before you rely on it.
Will these scripts pass TikTok and Meta ad policies?
The rules in the prompt reduce risky claims, but they don’t replace a policy check. Review every final script against the current TikTok and Meta ad policies before you publish.
Final Thoughts on Claude Workflows for E-commerce Product Video Script Writing
The workflow is a sequence: generate many hooks, score them, expand only the best into full scripts, test them, and feed the results back. Each run builds on the last, so your scripts improve instead of restarting from zero. Start with one product and one platform, and run the master prompt today.
If you’re running a seasonal offer, pair these scripts with the ChatGPT prompts for Black Friday sales copy and promo emails.
Frequently Asked Questions
Can Claude write TikTok and Reels product video scripts?
Yes. Given specific product details and a defined format, Claude can write complete scripts with timing, visuals, on-screen text, and a CTA. It writes the script, but it does not film or edit the video.
How long should a product video script be?
Match the script to your target length. Spoken delivery runs at roughly 2.5 words per second, so a 15-second video needs about 35-40 spoken words and a 30-second video about 70-80. If Claude’s draft runs longer, add the word limit to the rules and run it again.
How many hooks should I test per product?
Start with the three scripts the prompt builds, each using a different hook, and test them against each other. Generate new hooks only after you know which framework held attention.
What does the [VERIFY] tag mean in the output?
Claude adds it to any claim your inputs don’t support, such as a result, statistic, or guarantee. Before filming, either supply real proof or delete the claim. Never publish a script with an unresolved tag.
Can I use this prompt in ChatGPT or Gemini?
The prompt is plain text with tags, so you can paste it into other assistants. How strictly each one follows the steps varies, so test the output before you rely on it.
Will these scripts pass TikTok and Meta ad policies?
The rules in the prompt reduce risky claims, but they don’t replace a policy check. Review every final script against the current TikTok and Meta ad policies before you publish.