ChatGPT Prompts for Video Scripts: Hook, B-Roll & CTA

ChatGPT prompts for video scripts work when you lock platform, length, audience, and offer before you ask for lines. This guide gives copy-paste kits for hooks, B-roll cues, and CTAs for YouTube and Shorts. You leave with ROLE / TASK / FORMAT scaffolds for GPT-5.5 Instant and GPT-5.6 Sol, plus a VIDEO_FACT sheet that keeps spoken scripts separate from AI video generation prompts.

You still read every line out loud before you film. PromptMake /text can scaffold a rough script ask at https://promptmake.net/text before you paste into ChatGPT. This page covers spoken video copy for creators and marketers.

Who ChatGPT prompts for video scripts help

You cut YouTube long-form, Shorts, Reels, or TikTok and need spoken lines that open, hold, and close. You already know the topic and the offer. You need the model to draft a hook bank, mark B-roll beats under each line, and write a CTA that matches the platform. The patterns fit solo creators, product marketers, educators, and agencies that ship weekly scripts.

Skip this page if you want camera, motion, and duration language for Kling, Hailuo, Luma Dream Machine, Pika, or Runway Gen-4. That work lives in AI video prompt structure and text-to-video prompt example guides. Skip it if you want thumbnail or packshot image prompts. Those belong on image workflows.

Treat ChatGPT as a script desk. You supply VIDEO_FACT, the promise, proof points, and the CTA goal. The model proposes spoken lines, on-screen text, and B-roll cues tied to those facts. You reject any draft that invents product claims, fake stats, or a soft close that never names the next step.

Lock VIDEO_FACT before you prompt

Name the video constraints in the first block of every script thread. Platforms punish vague openers and buried CTAs. A polished script from "write me a YouTube video" fails the moment you try to film it in under ten minutes or sell a product the draft never named. Put the constraints in ROLE and repeat the hard ones in a REMINDER line so a long paste does not bury them.

Allowed artifacts: hook banks (3–7 openers), A-roll spoken scripts with time marks, B-roll cue lists tied to each beat, on-screen text overlays, end-screen CTA variants, Shorts / Reels cutdowns from a long script, and retention edits that cut fluff. Banned artifacts: invented reviews, fake view counts, medical or financial claims you did not supply, and motion prompts for text-to-video models.

Keep a VIDEO_FACT sheet outside the chat: platform, target length, audience, one-sentence promise, offer or link, tone, brand words to use, words to ban, and proof you can show on camera. Update it when the offer shifts. Feed it into every prompt. The model should never be your only source of product truth.

Redact unpublished launch dates, private customer names, and affiliate terms that must stay off public scripts before paste. Consumer ChatGPT is a drafting desk, not a secure brand vault. Use your approved workspace tools when legal or compliance requires a locked chat.

Core ChatGPT prompts for video scripts: hook, B-roll, CTA

Strong ChatGPT prompts for video scripts give four inputs before any tone request: video facts, the promise and proof, the script artifact, and the claim boundary. Video facts are platform, length, audience, and CTA goal. Promise and proof are the one-line payoff plus bullets you can show on camera. The artifact names hook bank, full script with B-roll cues, Shorts cutdown, or CTA variants. The claim boundary forbids stats, reviews, and guarantees you did not paste.

Paste those four blocks near the top. Put format rules next. Put banned outputs at the end so they survive a long paste. GPT-5.5 Instant, GPT-5.6 Sol, Claude Sonnet 5, and Gemini 3.5 Flash all follow labeled blocks well. Vague "write a viral script" prompts produce generic hooks because the model has no length, no offer, and no proof.

Run one artifact type per thread. Mixing a hook bank and a full long-form script in the same chat blurs length rules and CTA goals. Close the chat after you finish one pass. Open a new thread for the Shorts cutdown so long-form pacing does not leak into a 45-second cut.

Role, task, format skeleton

Copy this skeleton and fill the brackets with your material:

ROLE: You are a video script assistant for [YouTube long-form | YouTube Shorts | Reels | TikTok]. You work from VIDEO_FACT and PROOF only. You write spoken lines a host can say out loud. You never invent stats, reviews, prices, or guarantees.

TASK: Turn VIDEO_FACT and PROOF into a [hook bank | full script with B-roll cues | Shorts cutdown | CTA variant set] for [topic]. Keep every claim tied to PROOF. Mark gaps with [NEED PROOF]. If I ask for AI video generation prompts (camera moves, seed, duration for Kling/Runway), refuse and offer a spoken script artifact instead.

FORMAT: Use the section headers listed under OUTPUT_SHAPE. Time marks in [mm:ss] or beat numbers. Cap each spoken line at [N] words unless I raise the limit.

VIDEO_FACT: [platform, length, audience, promise, offer/link, tone, banned words]

PROOF: [bullets you can show: demo steps, screenshots, quotes you own, specs]

OUTPUT_SHAPE: [list required sections]

RULES: No fake social proof. No medical or financial advice beyond PROOF. No text-to-video motion prompts. If a claim needs a number I did not supply, write [NEED PROOF].

REMINDER: Spoken script only. Hook, B-roll cues, CTA. Never invent proof.

That reminder line stops the model from padding with "studies show" lines you cannot defend on camera. Models love tidy authority. Your rule forces a gap marker you can fill from your product page or demo.

Paste-ready kits for hook, B-roll, and CTA

Hook bank pattern adds: "From VIDEO_FACT and PROOF, write 7 hooks under 18 words each. Label each hook type: Pattern interrupt | Specific outcome | Mistake callout | Curiosity gap | Direct promise. No clickbait that the body cannot deliver. No questions that take more than two seconds to parse on mute with captions. Output as a numbered list only." You pick one hook. The bank gives you options for A/B tests.

Full script with B-roll pattern:

TASK: From VIDEO_FACT and PROOF, write a spoken script for [length]. Sections: Hook (0–15s) | Setup | Proof beats | Soft objection | CTA. Under each spoken line, add a B-ROLL line with a concrete visual (screen recording, hands on product, B-roll location). Add ON-SCREEN TEXT for the three lines that must work on mute. Do not invent proof. Mark missing visuals with [NEED FOOTAGE].

Shorts / Reels cutdown pattern: "TASK: From LONG_SCRIPT, cut a [30–60]s vertical script. Keep one promise. Keep one proof beat. Keep one CTA. Drop setup fluff. Rewrite the hook for silent autoplay with captions. Output: spoken lines with beat numbers, B-ROLL cues, ON-SCREEN TEXT, and a final CTA under 12 words."

CTA variant pattern: "TASK: From VIDEO_FACT, write 5 CTAs. Types: Subscribe + next video | Comment prompt | Soft product trial | Hard product link | Save for later. Each CTA: spoken line, on-screen text, and where it sits (mid-roll | end). Match platform norms. No fake urgency. No invented discounts."

Step-by-step video script workflow

Use one chat thread per video or cutdown. Dumping hooks for three unrelated topics into one long thread blurs offers and claim rules. The loop below keeps a stable VIDEO_FACT sheet while you swap only this week's topic and proof. You spend free ChatGPT or PromptMake runs on structure, then human time on a cold read, a mute caption check, and a film list.

Keep VIDEO_FACT in plain text outside the chat: platform, length, audience, promise, offer, tone, and banned words. Update it after each launch. Feed it into every prompt that touches the script. The model should never store your only copy of product claims.

Step 1: Fill VIDEO_FACT and list proof you can film

Write the sheet in ugly bullets. Example: "Platform: YouTube long-form. Length: 8–10 min. Audience: freelancers who invoice late. Promise: a 3-step invoice follow-up that cuts unpaid invoices. Offer: free checklist link in description. Tone: calm, specific. Ban: 'guaranteed,' 'viral,' fake dollar amounts. Proof: screen recording of template, two anonymized before/after timelines I own."

List footage you already have or can shoot this week. Hooks that demand a lab demo fail if you only have a laptop screen. Match PROOF to what the camera can show.

Step 2: Generate hook bank, then lock one opener

Run the hook bank pattern first. Read each hook out loud at speaking speed. Cut any line that needs a second pass to understand. Pick one opener that the body can pay off in the first proof beat.

Ask for a mute check next: "TASK: For HOOK and first 20 seconds of SCRIPT, write caption-friendly ON-SCREEN TEXT (max 6 words per card) so a muted viewer gets the promise. Flag any spoken line that fails without audio." Fix those lines before you draft the full body.

Step 3: Draft full script, B-roll list, then cold-read

Run the full script with B-roll pattern. Close the output. Read the spoken lines out loud with a timer. Cut anything you trip over. Then build a shoot list from every B-ROLL line marked [NEED FOOTAGE].

Optional audit: "TASK: Audit SCRIPT against VIDEO_FACT and PROOF. For each claim, reply Keep, Soften, or Remove. Soften means the line overreaches PROOF. List any CTA that never names the next step. Never add stats I did not paste."

Optional scaffold: open https://promptmake.net/text, describe "ChatGPT prompts for video scripts with hook bank, B-roll cues, and CTA variants for YouTube and Shorts," generate once, then paste your VIDEO_FACT into the returned structure. Guests get about three runs per day; free accounts get about five. Use a run to shape the prompt, then finish in your chat model.

Save the winning prompt next to the video title so the next episode reuses the same wrapper. Film only lines you can say without staring at the teleprompter for every word.

Mistakes that wreck video script ChatGPT use

Mistake 1: Asking for a "viral script" with no platform, length, or offer. The model invents hooks the body cannot pay. Lock VIDEO_FACT first.

Mistake 2: Skipping B-roll cues. A spoken-only draft leaves you hunting footage after the edit calendar is full. Require a B-ROLL line under each beat.

Mistake 3: Writing CTAs that say "like and subscribe" with no next step. Name the video, link, comment prompt, or trial. Match platform norms.

Mistake 4: Letting the model invent case studies or dollar results. Ban new numbers in RULES. Use [NEED PROOF] for gaps.

Mistake 5: Mixing spoken script asks with Kling or Runway motion prompts in one thread. Split libraries. Spoken scripts stay here. Camera and duration language stay in video generation guides.

Mistake 6: One mega-prompt that asks for hook bank, 12-minute script, Shorts cutdown, thumbnail text, and email blast together. Split artifacts. Reuse VIDEO_FACT; change TASK and FORMAT.

Mistake 7: Trusting Instant or Flash as the final judge on regulated claims. Fast models fit first drafts. Route health, finance, or legal-adjacent lines through a careful edit pass and your own compliance check.

Mistake 8: Never reading the script out loud. Fluency on screen is not the same as breath and pace on camera. Require a cold read before you shoot.

Model notes for script prompts (mid-2026)

ChatGPT often defaults to GPT-5.5 Instant for fast chat. Instant fits hook banks, CTA variants, and first Shorts cutdowns when you already locked VIDEO_FACT and PROOF. Keep prompts short: ROLE, TASK, FORMAT, VIDEO_FACT, PROOF, RULES. Skip long chain-of-thought slogans.

GPT-5.6 Sol fits harder edit passes: claim audits against PROOF, retention cuts that must keep the promise, and scripts that must not invent stats. Give goal, constraints, and format. Drop "think step by step" padding on reasoning-class models.

Claude Sonnet 5 handles long pastes of an existing transcript you want to rewrite into a tighter script with B-roll cues. Claude Opus 5 fits careful audits when a wrong claim would cost a takedown or an ad rejection. Gemini 3.5 Flash fits volume work: many hook variants, batch CTA sets, and cutdowns from one long transcript. Gemini 3.1 Pro fits hard reasoning over a thick PROOF packet when you need conflict flags against VIDEO_FACT.

Hedge on exact menu names in each vendor UI. They shift. Re-check the model picker when you open a new thread. Prompting split that holds: Instant and Flash get RTF plus a short sample of your house script format when you need matching headers. GPT-5.6 Sol, Opus 5, and Gemini 3.1 Pro get goal plus constraints plus format, with an explicit refuse-to-invent-proof rule and a [NEED PROOF] token. All need your VIDEO_FACT in the message. None replace a cold read or a mute caption check.

Build script prompts with PromptMake /text

Write the rough ask in plain words: platform, length, hook vs full script vs CTA set, and whether you need B-roll cues. Open https://promptmake.net/text and generate a structured prompt once. Expect labeled sections you can fill with VIDEO_FACT and PROOF.

Edit offers, banned words, and proof bullets yourself. PromptMake cannot know your product claims. Paste the filled prompt into Instant or Flash for first drafts, or GPT-5.6 Sol / Opus 5 / Gemini 3.1 Pro for claim audits. Keep free-tier runs for scaffolding, not five synonym retries of the same weak "make it viral" ask.

Workflow that sticks: VIDEO_FACT, list filmable proof, PromptMake scaffold, fill facts, fast model draft, cold read, reasoning model audit, you lock the shoot list. Store one template per artifact type so you do not rewrite ROLE and RULES from scratch each week.

FAQ

What are the best ChatGPT prompts for video scripts in 2026?

The best ChatGPT prompts for video scripts lead with ROLE and claim fences, paste VIDEO_FACT and PROOF, then demand FORMAT for a hook bank, full script with B-roll cues, Shorts cutdown, or CTA variants. Add a second audit prompt that marks Keep, Soften, or Remove against your proof and lists any CTA that never names the next step. Match GPT-5.5 Instant or Gemini 3.5 Flash for first drafts and GPT-5.6 Sol, Claude Opus 5, or Gemini 3.1 Pro for claim audits.

How do I write a hook with ChatGPT for YouTube or Shorts?

Lock platform, length, and promise in VIDEO_FACT. Ask for a numbered hook bank with labeled types (pattern interrupt, specific outcome, mistake callout, curiosity gap, direct promise) and a hard word limit per opener. Read each hook out loud, then run a mute caption check so the first seconds work without audio. Pick one opener the body can pay off in the first proof beat.

How should ChatGPT mark B-roll in a video script?

Require a B-ROLL line under every spoken beat with a concrete visual you can film or already own. Add [NEED FOOTAGE] when the draft asks for a shot you do not have. Build your shoot list from those cues before you schedule the edit. Vague "show B-roll here" lines waste a filming day.

How do I write CTAs in ChatGPT prompts for video scripts?

Name the platform and the next step in VIDEO_FACT: subscribe plus a specific next video, a comment prompt, a free trial, a product link, or a save-for-later cue. Ask for mid-roll and end variants with spoken line and on-screen text. Ban fake urgency and invented discounts. Test one CTA per cut so analytics stay clean.

Are ChatGPT prompts for video scripts the same as AI video generation prompts?

Spoken script prompts draft what a host says, what appears on screen, and which footage supports each beat. AI video generation prompts describe subject, motion, camera, and duration for models like Kling, Luma, Pika, or Runway. Keep the libraries apart. Use this guide for YouTube and Shorts copy. Use the video generation guides when you need motion language.

Should I use GPT-5.5 Instant or GPT-5.6 Sol for script prompts?

Use Instant for hook banks, CTA sets, and first Shorts cutdowns when VIDEO_FACT and PROOF are already in the message. Use GPT-5.6 Sol when you need a careful claim audit, a retention cut that must keep the promise, or a rewrite of a long transcript with strict proof rules. Run the same facts through both only when you measure quality for a recurring series workflow.

Can PromptMake help with ChatGPT prompts for video scripts for free?

Yes. PromptMake /text turns a rough script idea into a labeled prompt you can aim at ChatGPT, Claude, or Gemini. Guests get about three generations per day; registered free users get about five. Fill in your own VIDEO_FACT and proof, then paste into Instant for drafts or GPT-5.6 Sol for audits.