TikTok scripts fail in a predictable place: the AI draft opens with a “hook” that is actually a setup, runs a middle that explains instead of reveals, and bolts a CTA on after the energy is gone. The fix is structural, not stylistic — three sections, each with a single job, each timed against how TikTok’s algorithm actually scores a video in 2026. This tutorial gives you the hook-reveal-CTA template, three reliable variations, the exact six-field prompt, and the retention numbers that tell you whether a script worked. The goal is a batch day that produces five fileable scripts in 30 minutes, not five half-finished ones.
TL;DR
- TikTok’s 2026 completion-rate cliff sits at 85%: videos watched to at least 85% get roughly a 4x distribution multiplier over those below it (Retensis benchmarks).
- The highest-completion sweet spot is 21-34 seconds, not 60. Use 15-30s for reach, 21-34s for hook-plus-payoff, 60-180s only when the topic genuinely needs the room.
- The first 3 seconds are scored separately. Videos holding 70-85% retention at 3 seconds get about 2.2x more total views, so the hook is the single highest-leverage line in the script.
- Use a general model (ChatGPT GPT-5.5 or Claude Sonnet 4.6) for the writing, feed it 3 reference transcripts, and generate 5 hooks before you write a single full draft.
- Average TikTok watch time is roughly 8.4 seconds, which is why no amount of clever middle saves a slow open.
What length should you actually target?
The old “60 seconds is the sweet spot” advice is out of date. As of June 2026, completion rate is the dominant FYP signal, and it falls fast with length. Match the length to the job, not to a default.
| Length | Typical completion (2026) | Best for |
|---|---|---|
| Under 10s | ~81% | Loops, punchlines, single-line hot takes |
| 11-15s | ~76% | Fast tips, trend formats, one payoff |
| 21-34s | Highest FYP rate | Hook + reveal + payoff (the default for this workflow) |
| 31-60s | ~42% | Multi-step how-to, micro-story with tension |
| 60s+ | ~30% threshold | Instruction or narrative that earns the time |
Completion data is from Retensis and OpusClip. The practical read: write for 21-34 seconds by default and only stretch when the payoff genuinely needs more setup. To trigger viral distribution you generally need 70%+ completion, and 85% unlocks the multiplier.
Who this is for
TikTok creators batching weekly content, founders posting under a personal brand, UGC creators producing spec ads, and small marketing teams without a dedicated short-form writer. It is especially useful if your retention drops sharply at 3-5 seconds — that is a hook problem, and this workflow attacks it directly. Skip it for heavily personality-driven content where your unscripted voice is the product.
Before you start
- Pull 3 reference TikToks in the style you want and paste their transcripts or URLs into the AI session. Examples beat adjectives. Tools like TokScript pull transcripts from any TikTok in seconds.
- Pick the template that fits: curiosity loop (best for tips), contrarian take (best for hot takes), or demo payoff (best for product / before-after).
- Decide the CTA before you draft, not after. “Follow for part 2” outperforms “subscribe” because it is native to how TikTok viewers behave.
- Confirm your account’s average watch time in TikTok Studio analytics. Without it, you cannot tell whether the new script is winning. The audience-retention graph there shows second-by-second drop-off and your completion percentage.
Choosing the AI model
Specialized “TikTok script generators” exist, but for control and voice-matching a general model wins. As of June 2026:
- ChatGPT (GPT-5.5) and Claude (Sonnet 4.6) both write strong short-form scripts when the prompt is specific. Both are free to start; paid tiers (ChatGPT Plus $20/mo, Claude Pro $20/mo) lift usage limits and give larger context for pasting multiple references.
- CapCut, owned by TikTok’s parent ByteDance, generates a script from a prompt and then pulls matching stock footage and auto-captions in one flow — useful when you want script-to-edit in a single tool rather than maximum scripting control.
For this workflow, use a general chat model for the script and keep CapCut or your editor for the edit. See our AI short-video script tutorial for the cross-platform version of this prompt.
The six-field brief
Long briefs produce bland scripts. Constrain the model to exactly six fields:
Niche: [e.g. solo SaaS founders]
Target viewer: [who, what they want, what frustrates them]
Template: [curiosity loop | contrarian take | demo payoff]
One-sentence promise: [the single payoff this video delivers]
Voice adjective: [e.g. dry, blunt, warm]
CTA: [e.g. "comment TEMPLATE for the prompt"]That is the whole brief. More context dilutes the hook.
Step by step
- Paste the six-field brief plus your 3 reference transcripts into ChatGPT or Claude.
- Generate 5 hook variants matching the chosen template. Never use the first one — it is almost always the safest and weakest. Force distinct patterns: specific number, contrarian claim, calling out the viewer, before-after gap, a question on a specific pain point. (Question hooks on a pain point outperform generic openings by roughly 23% on retention, per OpusClip’s hook data.)
- Pick one hook. Generate the full script in three labeled sections sized for 21-34 seconds: HOOK (0-3s), REVEAL (3s to ~3s before end), CTA (final 2-3s).
- Read it aloud and time it. AI scripts overrun by 15-25%. Cut filler — “really”, “actually”, “kind of”, and any sentence over 12 words.
- Add direction notes inline:
cut on word X,B-roll: hand holding phone,text overlay: keyword,zoom on payoff word,music drops here. The model never adds these. - Extract a shot list from the direction notes. The script is for audio; the shot list is what you film against.
- Film, post, and check the 3-second hold and completion in TikTok Studio. If the 3-second hold is below ~70%, the hook is the failure point — regenerate hooks only, not the whole script.
The three template variations
- Curiosity loop — open a question in the hook, withhold the answer through the reveal, close it at the payoff. Best for tips. It tolerates a softer hook because the open loop does the holding work.
- Contrarian take — lead with a claim that contradicts the niche’s common wisdom, then defend it. Best for hot takes. Highest risk and highest ceiling; a weak defense reads as bait.
- Demo payoff — show the before state, run the transformation in the reveal, land the after at the payoff. Best for products and before-after content. The visual carries it, so the script stays lean.
One template per content type. A curiosity loop will not save a product demo, and a demo structure wastes a hot take.
Quality check before you film
- Does the hook land in 3 seconds? Read only the first line aloud. If it is setup, it is too slow.
- Does the reveal earn the watch? Each line should add information, not restate the hook.
- Is the CTA platform-native? “Follow for part 2” or “comment X for the template” beats “subscribe”.
- Are direction notes present? Without them the script is half-built.
- Does the whole thing speak well? Sentences over 12 words die on camera.
Common mistakes
- Using the first hook the model writes. Always generate 5 and reject the safe one.
- Treating the reveal as exposition. Each line must move the payoff forward.
- Generic CTAs. “Subscribe” signals not-of-the-platform on TikTok.
- One template for all content. Curiosity loops do not work for product demos.
- No retention tracking. Without the 3-second hold and completion numbers, you cannot tell which hooks work.
- Writing to 60 seconds by default. The completion math punishes length; aim for 21-34s unless the topic earns more.
- Leaning on AI voiceover for the main delivery. Your real voice outperforms it; reserve AI voice for caption-style overlays.
FAQ
- What length should a TikTok script target?: Default to 21-34 seconds, the highest-completion range on the For You Page in 2026. Use 15-30s for pure reach, and 60-180s only when instruction or a story genuinely needs the room. Completion rate falls fast with length, so longer scripts have to earn it.
- Which template should I start with?: The curiosity loop. It tolerates a soft hook and lands payoffs reliably, which makes it the most forgiving first attempt.
- How do I make AI sound like me?: Feed 3-5 of your best past scripts as voice references in the same session. Generic AI tone is the default; voice transfer requires concrete examples, not a “sound natural” instruction.
- Should I use ChatGPT or Claude for this?: Either. As of June 2026, GPT-5.5 (ChatGPT) and Sonnet 4.6 (Claude) both handle short-form well; both have free tiers, and the $20/mo paid plans mainly buy higher limits and room to paste references. Pick whichever you already pay for.
- Do I need AI voiceover?: No. Your real voice outperforms AI voice for the main delivery. AI voice is fine for caption-style overlays only.
- How long until the workflow is fast?: First script: 45-60 minutes while you build the template. After about 10 scripts with saved templates: 10-15 minutes each.