The short version: ChatGPT can’t watch your footage or edit anything, but it’s brilliant at turning a transcript, a script, or a rough idea into a clear editing brief: hooks, cut points, B-roll suggestions, text overlays, and pacing notes. Feed it specifics from your actual video instead of asking for generic “creative ideas” and the output goes from forgettable to usable in one go.
Need to lighten the mood between edits? Here are some funny prompts to try in ChatGPT.
What ChatGPT can help with (and what it can’t)
Let me get one thing out of the way because it wastes people’s time. ChatGPT has never seen your video. It doesn’t know your footage looks shaky at the 40 second mark, it doesn’t know your voice cracks at 2:15, and it has no idea what B-roll you filmed. Anyone telling you to type “give me video editing ideas” and expect something specific is setting you up to get the same twelve generic suggestions everyone else gets: jump cuts, trending audio, a zoom on the punchline.
Where it earns its keep is turning what you already have (a script, a transcript, a shot list, a rough description of what happened on camera) into a structured plan. Think of it as a very fast, very patient assistant editor who can read a 4,000 word transcript in two seconds and hand you back a clip list. It’s not the editor. It’s the person who preps the brief so your actual editing time gets shorter.
The story that changed how I do this
Last spring I recorded a 47 minute webinar on AI adoption for small business owners. My old process was to sit through the whole recording with a notepad, scribbling timestamps I thought might make good short clips for LinkedIn and Instagram. That took close to 90 minutes on its own, before any actual cutting happened.
I changed the process. I pulled the transcript straight from the recording software, pasted the whole thing into ChatGPT, and asked it to find the five strongest standalone moments, each one a self-contained thought that would make sense with no context. I gave it one rule: each clip had to be under 90 seconds when spoken aloud. It came back with five timestamped sections, a suggested hook line for each, and a one-sentence reason why that moment worked on its own.
That whole process took 18 minutes. I checked each timestamp against the recording (because it does sometimes get the timing slightly off when a transcript has no proper timestamps embedded), adjusted two of them by about 15 seconds, and handed the brief to my editor. Clip planning went from 90 minutes to under 20. That’s the actual number, not a rounded-up marketing figure.
The prompts I use
These are the prompts I keep coming back to, not a wishlist of things that sound clever but never get used. Each one needs an input from you first: a transcript, a script, or a paragraph describing your raw footage.
1. Finding the hook
The first three seconds decide whether anyone watches the rest, and most people write their hook after the video is already cut, which is backwards. Try this:
- “Here is the transcript of my video. Give me 8 possible opening lines that would work as the first 3 seconds, each under 12 words, written to make someone stop scrolling. Rank them by how surprising or specific they are.”
I ran this exact prompt on five different videos in one afternoon and never got a repeated structure. That only works because I fed it a different transcript each time. Paste the same vague brief in five times and you’ll get the same five hooks back with the nouns swapped.
2. Building the cut list
This is the one that saves the most time. It turns a wall of text into a shot-by-shot plan.
- “Break this transcript into sections that could each stand alone as a 30 to 60 second short. For each one, give me the approximate timestamp, a one-line summary, and a suggested caption.”
This is also where the honest limitation shows up again. If your transcript file doesn’t have timestamps baked in, ChatGPT will guess based on word count and speaking pace, and it guesses reasonably well but not exactly. Always check the actual timestamps against your recording before you send anything to an editor. I’ve had a 12 second drift on a 45 minute file. Not a disaster, but enough to annoy an editor who trusted the numbers blindly.
3. B-roll and cutaway ideas
This one needs you to describe what you’ve filmed, not just the topic.
- “I’m talking about [topic] on camera for this section. I have footage of [describe what you filmed: office shots, screen recordings, product shots, stock options]. Suggest where cutaways would help pacing and what specific shot would work at each point.”
Skip the “describe what you filmed” part and you’ll get suggestions like “insert relevant B-roll here,” which tells you nothing you didn’t already know.
4. Pacing and rhythm notes
ChatGPT can’t judge pacing by watching your video, but it can flag where a script drags based on sentence structure and repetition.
- “Read this script and tell me where the pacing likely feels slow based on sentence length and repeated ideas. Suggest where a cut, a text overlay, or a change of shot would help keep attention.”
5. Text overlays and captions
- “Write 5 short text overlay options (under 8 words each) for this moment in the video: [paste the exact line being spoken]. Match the tone: [confident, funny, urgent, etc].”
This one is quick to use inside CapCut, Premiere, or Descript because you can copy the overlay text straight into your timeline without reworking it.
A step by step workflow you can copy
Here’s the exact sequence I use now for repurposing a long video into shorts, start to finish:
- Step 1: Export or generate the transcript from your recording software (most platforms, including Zoom, Riverside, and StreamYard, do this automatically now).
- Step 2: Paste the full transcript into ChatGPT and ask it to identify 5 to 8 standalone moments under 90 seconds each.
- Step 3: Ask for a hook line for each moment, using the exact prompt above.
- Step 4: Check each suggested timestamp against your actual recording. Adjust anything that’s off.
- Step 5: Ask for 3 to 5 text overlay lines per clip, based on what’s said at that timestamp.
- Step 6: Write the final brief (timestamp, hook, overlay text, any B-roll notes) into a single document.
- Step 7: Cut the video yourself or hand the brief to an editor. Either way, the editing itself now takes a fraction of the time because there’s no guesswork left.
That’s the whole system. No plugin, no paid tool, just a transcript and a set of prompts that ask for specifics instead of vibes.
Where people go wrong with this
The mistake I see most often, including from people who’ve been doing this for months, is asking ChatGPT to be the creative director instead of the assistant editor. “Give me some fresh, unique video editing ideas” produces exactly what you’d expect from a prompt with no input: a list that could apply to literally any video, which means it applies meaningfully to none of them. The same problem shows up in photo editing prompts for Instagram, where the people getting decent results are the ones feeding it their actual photo, not asking for abstract trend ideas.
The other mistake is trusting the tool to know what’s currently trending. ChatGPT’s knowledge of what’s popular on TikTok or Reels this week is patchy at best, because trends move faster than most model training cycles and it has no live feed of what’s performing right now. If you want current trend data, that’s a job for the platform’s own trending page or a tool built specifically to track it, not a chat prompt. I’ve seen people ask for “the trending transition effect right now” and get an answer describing something that was popular eighteen months ago, delivered with total confidence and zero indication it might be stale.
And here’s the uncomfortable bit nobody likes admitting: the ideas ChatGPT gives you are only as good as what you fed it, which means the quality ceiling is set by your own script, your own footage, your own transcript. If your raw material is thin, no amount of clever prompting turns it into something sharp. I’ve watched people spend 40 minutes refining a prompt trying to squeeze creativity out of a five-line brief when the fix was simply to film more, or better, footage in the first place.
Prompts for planning the whole video before you film
This works the other way round too, before you’ve filmed anything.
- “I’m planning a 60 second video about [topic] for [platform]. Give me a structure: hook, 3 main points, and a closing line, written for someone with a 3 second attention span.”
- “Suggest 3 different visual styles for this same script: one talking head, one screen recording with voiceover, one B-roll heavy with text overlays. Tell me which platform each style tends to suit best.”
This is closer to what I’d call content strategy than editing, but the two overlap more than people admit. If you’re building out a wider content system, this is the same thinking I’ve written about in how to use ChatGPT for content creation, where the transcript-in, structure-out approach applies just as well to blog posts as it does to video.
The tools worth pairing this with
ChatGPT plans, but you still need something to cut. Descript is worth knowing about because it lets you edit video by editing the transcript directly, which pairs neatly with the workflow above since you’re already working from text. CapCut has a built-in caption and auto-cut feature that’s decent for short-form once you’ve got your clip list sorted. Premiere Pro and Final Cut remain the standard for anything longer or more polished. None of these need ChatGPT to function, but combining a text-based planning step with a text-based editing tool cuts out a lot of back and forth. If you want a wider list of what pairs well with ChatGPT day to day, I’ve rounded up the ones I use in 11 ChatGPT tools that will turn you into a productivity powerhouse.
Getting past generic output
If the ideas you’re getting feel flat, it’s almost never the model’s fault, it’s the input. Three things fix it fast:
- Paste the actual transcript or script, not a summary of the topic.
- Tell it the platform (a LinkedIn video and a TikTok need completely different pacing and openings).
- Give it a constraint (length, tone, number of options) because open-ended requests produce open-ended, forgettable answers.
I wrote more on this pattern, why the same tool produces such different quality of output depending on how much you’re willing to put into it, in breaking free from the ChatGPT mould. The short version of that longer piece: the mould exists because most people ask lazy questions and accept the first answer.
Related reading: ai photo editing strengths.
Related reading: chat gpt portrait.
See also What ChatGPT Prompts Work Best for Planning Travel?, which picks up where this leaves off.
For the rest of the ChatGPT prompts questions, see my ChatGPT prompts guide.
Frequently asked questions
Can ChatGPT edit my video for me?
No. ChatGPT has no way to view, cut, or export video. It can only work with text, so it’s useful for planning (hooks, cut points, B-roll notes, overlay text) that you then apply in an actual editing tool like CapCut, Descript, or Premiere Pro.
What’s the best ChatGPT prompt for video editing ideas?
The best prompt is one that includes your actual transcript or script, not a general question about the topic. Something like “here is my transcript, find 5 clips under 90 seconds that could stand alone, with a hook line for each” consistently outperforms a vague request for creative ideas.
Does ChatGPT know current TikTok or Reels trends?
Not reliably. Trend cycles move faster than the model’s knowledge updates, and it has no live feed of what’s performing right now, so it can confidently suggest a transition or audio trend that’s already months out of date. Check the platform’s own trending tab for anything time-sensitive.
How much time does this save on a real project?
On my own webinar-to-shorts process, clip planning went from about 90 minutes of manual scrubbing and note-taking to under 20 minutes using a transcript-based prompt workflow. The saving comes almost entirely from skipping the manual timestamp hunt, not from the editing itself getting faster.
