YouTube AI prompts.
12 copy-paste prompts covering every text asset a video needs: 20 title options under 60 characters, hooks that land in three seconds, descriptions whose first 125 characters stand alone, chapters from a transcript, scripts with retention beats marked, thumbnail concepts, tags and Shorts.
GPTPrompts.AI Editorial
Built from the YouTube questions our own site search recorded and could not answer · Last updated August 2026
The 6 things worth prompting for
In production order. Each links to its prompts below.
- 1Titles. 20 options under 60 characters, ranked
- 2Hooks. 5 openings that land in under 3 seconds
- 3Descriptions and chapters. First 125 characters standalone, chapters from transcript
- 4Scripts. Spoken language, retention beats marked, from your research
- 5Thumbnails. 3 elements maximum, text added separately
- 6SEO and tags. Audience phrasing research, not tag stuffing
How do I write YouTube titles with AI?
Ask for 20 titles under 60 characters, built on a specific curiosity gap rather than a vague tease, then pick the one you would click yourself. Volume matters here: the tenth title is usually better than the first, and asking for one title at a time produces the most obvious option every time.
What decides whether it works
- Under 60 characters so the title is not truncated on mobile, where most browsing happens.
- Specific beats mysterious. I tried X for 30 days outperforms You won't believe what happened, because the first promises a known payoff.
- Front-load the hook. The first three words carry the click. Sidebar and search results cut the rest.
- One idea per title. Two competing hooks split attention and win neither.
Write 20 YouTube title options for a video about [topic]. Rules: - Under 60 characters each. - Each built on a specific curiosity gap: the viewer should know what payoff they are being promised. - Front-load the hook in the first three words. - No clickbait that the video cannot deliver. - Vary the angle: result-led, mistake-led, comparison, number-led, contrarian. Then rank your top 5 and explain in one line each why that one would earn the click. Video content: [describe what actually happens in the video] Audience: [who this is for]
Here is my current YouTube title: [title] Write 8 variants that test one variable at a time so I can learn what works, not just guess: - 2 that change only the emotional angle - 2 that change only the specificity of the number or result - 2 that change only the length, one much shorter - 2 that change only the opening three words Label each with the variable being tested.
What is the best AI prompt for a YouTube hook?
Ask for five hook options that each land within three seconds and open on a concrete image or claim, never on a greeting. The single biggest retention error is spending the first fifteen seconds on introductions and channel branding while the viewer decides whether to leave.
What decides whether it works
- Three seconds, not thirty. The decision to stay is made almost immediately.
- Never open with hello or welcome back. Nothing in a greeting gives a reason to stay.
- Open on the payoff or the conflict. Show the result first, then explain how you got there.
- Name the specific promise. By the end of this you will have X, said in one sentence.
Write 5 opening hooks for a YouTube video about [topic]. Each hook must: - Be speakable in under 3 seconds. - Open on a concrete image, result or conflict, not a greeting or channel intro. - Contain no filler: no hey guys, no welcome back, no in today's video. - Give the viewer a specific reason to stay for the next 10 seconds. Use a different structure for each: result-first, question, contradiction, stakes, direct address. Then write the 10 seconds that follow the strongest hook, so the opening does not collapse after the first line. Video content: [describe]
Here is the first 30 seconds of my video script: [paste] Tell me exactly where the viewer would leave and why. Then rewrite it so: - The first sentence is the hook, not the setup. - Anything the viewer does not need in the first 30 seconds is moved later or cut. - The promise of the video is stated in one specific sentence within the first 15 seconds. Show the cut version and tell me how many seconds you removed.
How do I write a YouTube description with AI?
Ask for the first 125 characters to work as a standalone summary, because that is roughly what shows before the fold and in search results, then a fuller body underneath. Most AI-written descriptions bury the useful sentence under a paragraph of throat-clearing nobody expands to read.
What decides whether it works
- First 125 characters carry the weight. They appear in search and above the fold.
- Write for a reader, not a keyword field. Stuffing keywords has not worked for years and reads badly.
- Include timestamps when the video has sections. They generate chapters and help retention.
- Put links after the summary, not before it. A wall of links above the description costs you the fold.
Write a YouTube description for a video about [topic]. Structure: 1. First 125 characters: a standalone summary that makes sense on its own in search results. Count them. 2. Two to three short paragraphs expanding on what the viewer will learn, written for a human, not stuffed with keywords. 3. A chapters block with timestamps, using the section list below. 4. A short links section at the end. Naturally include these terms without forcing them: [3 to 5 terms] Video sections and timestamps: [paste your list]
Below is the transcript of my video. Generate a YouTube chapters list. Rules: - The first chapter must start at 00:00, which YouTube requires for chapters to appear. - Minimum 3 chapters, each at least 10 seconds long. - Chapter titles under 40 characters, describing what is covered rather than being cute. - Use the actual topic shifts in the transcript, not evenly spaced intervals. Output only the timestamp list, ready to paste. <transcript> [paste with timestamps] </transcript>
How do I write a YouTube script with AI?
Give the model your actual research and ask for a spoken-language script with retention beats marked, rather than asking it to write a video about a topic from memory. Scripts written from a model's general knowledge read like a written article performed out loud, which is the most common reason AI-scripted videos underperform.
What decides whether it works
- Spoken language, not written. Short sentences, contractions, sentence fragments. Read it aloud.
- Mark retention beats every 30 to 60 seconds: an open loop, a pattern change, a payoff.
- Supply your own research. A model writing from recall produces generic content and occasional invented facts.
- Write the ending before the middle. Knowing the payoff keeps the middle from drifting.
Write a YouTube script for a [length] video about [topic], using only the research below. Requirements: - Spoken language: short sentences, contractions, fragments where natural. It must sound like talking, not reading. - Open with a hook that lands in under 3 seconds. No greeting. - Mark a retention beat every 30 to 60 seconds in square brackets, naming which it is: open loop, pattern change, payoff, or visual cue. - State the promise of the video within the first 15 seconds. - End with one specific next action, not please like and subscribe. - If the research does not cover something, write [NEEDS RESEARCH] rather than inventing it. <research> [paste your notes, sources, data] </research>
Below is a long-form video transcript. Find the 5 strongest standalone moments for Shorts. For each: - Quote the exact start and end from the transcript. - Explain in one line why it works without the surrounding context. - Write a 3-second hook to place in front of it. - Write an on-screen text overlay of 6 words or fewer. - Suggest a vertical framing note for the edit. Prioritise moments with a complete idea, not the moments you find most interesting. <transcript> [paste] </transcript>
What prompts work for YouTube thumbnails?
Describe the thumbnail as a photograph with a named lens, one light direction and a maximum of three visual elements, because thumbnails are viewed at roughly the size of a postage stamp and anything busier turns to mush. Then generate the image clean and add text yourself, since image models garble lettering.
What decides whether it works
- Three elements maximum. Subject, one prop or background cue, one text zone. More disappears at small size.
- Leave a text zone empty in the prompt and add wording afterwards where you control the typeface.
- High contrast between subject and background. Thumbnails compete in a crowded sidebar.
- Check it at thumbnail size before committing. Most designs fail this test, not the full-size test.
Create a YouTube thumbnail image for a video about [topic]. Composition: - Maximum 3 visual elements. Nothing else in frame. - Main subject on the [left/right] third, facing into the frame. - Strong contrast between subject and background so it reads at small size. - Leave the [opposite] third visually clean for text I will add myself. - Single light source from [direction], hard enough to create clear separation. - Saturated but not neon colour palette. Do not render any text, letters or numbers anywhere in the image. 16:9 aspect ratio, photorealistic.
Before I generate anything, give me 6 thumbnail concepts for a video titled: [title] For each concept: - The single idea it communicates in under 5 words - The main visual element - The 3-word text overlay I would add - Why it would stand out next to competing thumbnails on this topic Then tell me which one is strongest at postage-stamp size, and why the others are weaker at that size.
How do I use AI for YouTube SEO and tags?
Use AI for topic and phrasing research, not for tag stuffing, because tags are a minor ranking signal on YouTube compared with title, thumbnail, and how long people actually watch. The highest-value AI task here is finding the specific phrasing your audience uses, then putting it in the title and the first line of the description.
What decides whether it works
- Tags are minor. Title, thumbnail and watch time carry far more weight.
- Match the viewer's actual phrasing, which is usually more casual than industry vocabulary.
- One topic per video. Videos covering three things rank for none of them.
- Retention beats keywords. A video people finish outranks a well-optimised one they leave.
My video covers [topic] for [audience]. List 15 search phrases this audience would actually type into YouTube. Use their casual phrasing, not industry terminology. Group them into: - Problem-aware (they know the problem, not the solution) - Solution-aware (they know the solution exists, want to see it done) - Comparison (they are choosing between options) Then tell me which single phrase should be in the title, and which 3 belong in the first line of the description.
Generate a complete metadata set for a YouTube video titled: [title] Provide: - 15 tags, ordered most to least relevant, mixing broad and specific - A one-line description opener under 125 characters - 3 alternative titles under 60 characters - 5 hashtags for the description Do not repeat the same phrase across tags. Do not stuff. If a tag is only marginally relevant, leave it out and tell me why you did.
Which AI tool for which YouTube task?
| Task | Best fit | Why |
|---|---|---|
| Titles and hooks | ChatGPT or Claude | Volume of options, then you pick |
| Long scripts | Claude | Holds a long transcript and a consistent voice |
| Chapters from transcript | Gemini | Handles long inputs and reads YouTube links |
| Thumbnail images | Gemini or ChatGPT | Both generate; neither renders text reliably |
| Search phrasing research | Gemini or Perplexity | Grounded in live search results |
| Shorts selection | Claude | Long transcript in one pass, picks complete ideas |
Frequently asked questions
What are the best AI prompts for YouTube?
The highest-value AI prompts for YouTube cover six assets: 20 title options under 60 characters built on a specific curiosity gap, five hooks that land within three seconds, a description whose first 125 characters stand alone, chapters generated from an actual transcript, a script written from your own research with retention beats marked, and thumbnail concepts limited to three visual elements. All 12 prompts on this page are copy-paste ready.
Can AI write a full YouTube script?
AI writes usable scripts when you supply the research and ask for spoken language, and generic ones when you ask it to write about a topic from memory. The difference is whether the model is organising your material or inventing content. Give it your notes, ask it to mark retention beats, and tell it to write [NEEDS RESEARCH] rather than filling gaps with plausible detail.
How long should a YouTube title be?
Under 60 characters, so it is not truncated on mobile where most browsing happens. The first three words matter most because sidebar and search listings cut the rest. Generate 20 options rather than one, since the tenth title is usually stronger than the first.
How many characters of a YouTube description are visible?
Roughly the first 125 characters show before the fold and in search results, so that opening needs to work as a standalone summary. Put links below it rather than above it, because a wall of links at the top costs you the only part most viewers read.
Do YouTube tags still matter for ranking?
Tags are a minor signal compared with title, thumbnail and watch time. They help with spelling variants and disambiguation rather than driving discovery. Time spent perfecting tags is better spent on a hook that keeps people watching, since retention influences distribution far more.
Can AI make YouTube Shorts from a long video?
AI is good at finding the moments, and you still edit. Give it the transcript and ask for the five strongest standalone segments with exact start and end quotes, a three-second hook for each, and a short text overlay. The instruction that matters is prioritising complete ideas rather than the moments the model finds interesting.
How do I start a faceless YouTube channel with AI?
The prompts on this page cover the text assets a faceless channel needs: titles, hooks, scripts, descriptions and chapters. What AI cannot supply is a reason to watch your channel rather than the other channels covering the same topic. Pick a narrow subject where you have genuine knowledge or access, then use AI for production speed rather than for the idea.