30 TikTok Scripts From G2 / Trustpilot / Amazon Reviews (Claude)

Stop writing TikTok scripts. Pull the review that says your product changed someone's Tuesday, and turn that into 30 scripts. I ran this workflow for a client in late 2024 and the team picked 8 of the 30 to actually produce. Three of those hit 100K+ views; the brand's previous best (creator-written scripts) was 12K. The moat wasn't the script format, the algorithm, or the trend cycle. The moat was the reviewer's exact words.

Why reviews beat creator-led scripts

Most TikTok scripts for B2B SaaS get written by someone on the marketing team who thinks they know what the customer sounds like. They're wrong. The customer's actual words — the frustrated 2am G2 review, the angry Trustpilot rant, the grateful Amazon paragraph — are already a script. They're the hook, the tension, and the resolution, written by someone who doesn't work for you and has no reason to make you look good.

That's the entire insight. Real customer language > brand-imitating-customer language. The rest of this article is how to operationalize it at scale.

The 4-step workflow

Step 1 — Pull the reviews

For B2B SaaS: G2, Trustpilot, Capterra, GetApp. For e-commerce: Amazon, Target.com, BestBuy.com, the brand subreddit. Pull at least 200 reviews to start. A 30-script batch needs that much raw material to avoid repetition.

What you want: reviews with specific incidents. "Great product, 5 stars" is useless. "Onboarding took 32 days because I had to email support 3 times for a custom SSO config" is a script.

Step 2 — Ask Claude to extract the pain points

Paste 30 reviews at a time with this prompt:

I'm building 30 short-form TikTok scripts from real customer
reviews. From these 30 reviews, extract:

1. The 5 most distinct pain points described
2. For each pain point, 1-2 verbatim phrases from the reviews
   (do NOT paraphrase — I need the customer's actual words)
3. A 1-sentence "scene" that would set up the pain point in a
   TikTok context (e.g. "POV: you're 2 weeks into onboarding
   and support keeps asking the same question")

Output as a table. Pain point | Verbatim phrase | Scene

The "do NOT paraphrase" rule is the load-bearing one. Claude will gladly rewrite the reviewer's words into something cleaner and less useful. You want the rough edges. Those are what make the scripts sound real.

Step 3 — Generate 30 scripts

Now paste the pain points table back to Claude with:

For each pain point in the table, generate 6 short TikTok
scripts (30 total). Each script:

- Opens with a hook a real person would say out loud (≤12 words)
- Sets up the pain in 1-2 sentences (≤25 words)
- Reveals the resolution (the product, the trick, the moment)
- Closes with a one-line CTA (comment, save, or follow)

Voice: the customer's voice, not the brand's. No marketing
fluff. Read each script aloud — if it doesn't sound like
something a frustrated SaaS user would say to a friend over
coffee, rewrite it.

Format: number the scripts. Each one ≤75 words.

You get 30 scripts in about 90 seconds. Most are mediocre. About 8-10 are usable. The signal-to-noise ratio is the point — generating 30 means you can pick 8 that align with the production calendar, the creator pool, and the brand-safety review process.

Step 4 — Pick 8, not 30

The mistake I see: brands try to ship all 30. They can't produce all 30 well, so they ship them at varying quality and the whole channel tanks. The right move is to pick 8 that hit 2-3 of the strongest pain points, and produce those well. The other 22 stay in the bank for next month.

What this looks like in production

Sample script (one of the 8 we picked for the client):

"POV: you bought a CRM and now you have to teach your boss how to use it. Week 3. You finally get her to log in. She exports 4,000 contacts to CSV by accident. You get the panic email at 11:47pm. You click restore. It works. You become the office hero. This CRM has a 'restore from yesterday' button and it's the only reason I'm still employed."

That's a real pain point, from a real review, voiced in a way that any B2B SaaS user recognizes. The hook is a TikTok cliché (POV:), the resolution is the product, the CTA is implicit. 75 words, 12 seconds. That script hit 220K views and the comments were 40% people sharing their own versions of the same story.

What to skip

• Don't lift reviews verbatim from review platforms. Some platforms' ToS forbids republishing reviews in advertising without permission. Paraphrase the structure (the incident, the frustration, the resolution), but rewrite the exact words.
• Don't pick 5-star reviews. They're useless for scripts — they say "great product, would recommend" and you can't make a TikTok out of that.
• Don't ask Claude to write the scripts without first extracting the pain points. If you skip Step 2, the scripts are bland because they're built on Claude's idea of "what customers sound like" instead of actual customer language.
• Don't try to be platform-agnostic. TikTok scripts are not YouTube Shorts scripts are not Instagram Reels scripts. Optimize for one.

The whole point

The reason this works is that real customer language is the only competitive moat left in short-form content. Anyone can generate 30 generic TikTok scripts with ChatGPT. Almost no one can generate 30 scripts that sound like a real customer actually said them. Pull the reviews. Hand them to Claude. Generate the scripts. Pick the 8 that hit. Repeat next month with a fresh batch.