Create YouTube chapters automatically with AI (2026)

Chapters are one of the best effort-to-result improvements on all of YouTube: they split the progress bar into titled segments, let viewers jump to what interests them and give Google indexable text that can appear as "key moments" in search results. And yet, half of long videos don't have them — because writing them by hand means rewatching the entire video jotting down minutes and seconds.

That's exactly the work AI transcription eliminates. If the video's text already exists with time markers, detecting where the topic changes and titling each block is a job for a prompt, not an afternoon. In this guide you have the full workflow: how to get the timestamped transcript, the prompts to generate and polish the chapters, YouTube's non-negotiable rules and the mistakes that make the list fail.

Why add chapters to your videos

Chapters aren't cosmetic: they change how the video is consumed and how it gets found.

• Retention that goes up instead of down. A viewer who lands on a 40-minute video looking for one specific thing has two options: find it in seconds with chapters, or scrub the progress bar blindly and leave. Every drop-off from not finding something is lost retention that chapters would have saved.
• Key moments on Google. Google can display the video's chapters directly in search results as "key moments", with a direct link to the exact second. That's extra SERP surface only chaptered videos get.
• Text that indexes. Chapter titles are text that YouTube and Google read. Well written, they reinforce the video's keywords with natural language — each chapter is a micro-description of what that block contains.
• The video becomes searchable. A tutorial with chapters works like documentation: someone coming back for step 4 goes to step 4. That generates return visits, and return visits are among the signals a channel benefits from most.
• The clear limit: chapters organize what's there, they don't fix what's missing. A video without structure produces chapters without substance. If the prompt can't find where to cut, the problem is usually in the script, not the AI.

Three ways to create chapters compared

There are three paths to chaptering a video. The difference lies in who controls the cuts and how much each video costs:

Manual chapters written in the description always override YouTube's automatic ones, so the transcript + AI workflow gives you the speed of automatic with the control of manual: the AI proposes, you decide.

Step by step: from transcript to chapters

Step 1 — Get the video's audio

Chapters are calculated on the final version of the video, the one the viewer will watch — not on the raw footage, because every editing cut shifts the timestamps. Export the audio from the final edit in your editor (MP3 or WAV is enough; you don't need the full video to transcribe) or use the file directly if your recording is already the definitive version, like a livestream or an unedited podcast.

Step 2 — Transcribe with timestamps

Upload the audio to an accurate transcription tool like VOCAP and receive the video's full text. The piece that makes the whole workflow work is the time markers: they're what allow each chapter to be anchored to the exact second the topic starts. Our guide on transcribing audio with timestamps explains the available formats. A one-hour video is transcribed in minutes.

Step 3 — Generate the chapters with a prompt

Run the timestamped transcript through Claude or ChatGPT with the first prompt in the next section. The model detects topic changes, anchors each one to the timestamp of the sentence that opens it and returns the list in YouTube format: one chapter per line, timestamp and title, starting at 0:00. The two non-negotiable instructions: cuts only where the topic actually changes, and no timestamp that doesn't exist in the transcript.

Step 4 — Adjust titles and verify the cuts

Review the proposal like an editor: does each cut fall where the topic actually changes? Do the titles say what the block contains or are they filler like "more content"? Rewrite the weak ones —short, concrete, with the keyword if it fits naturally— and check YouTube's three rules: first chapter at 0:00, minimum three chapters, none shorter than 10 seconds. The second prompt in the next section does this audit for you.

Step 5 — Paste the list into the description and check

Copy the final list into the video description (anywhere, though near the top is more useful for the viewer), save and open the video in the player: the progress bar should appear split into segments and each chapter's title should show when hovering over it. If it doesn't split, one of the three rules has failed — the list's format is the first thing to check.

Ready-to-copy prompts

Paste the timestamped transcript and add one of these prompts on top. They work with any current AI model.

Generate the chapters from the transcript

From this timestamped transcript, generate the video's YouTube
chapter list. Rules: (1) one chapter per real topic change, not
per paragraph; (2) each chapter starts at the timestamp of the
first sentence of that topic, using only timestamps that exist
in the transcript; (3) the first chapter is "0:00 Introduction"
(or the actual topic if the video jumps straight in);
(4) titles of 2-6 words that say what the block contains, no
clickbait; (5) output format: one line per chapter,
"M:SS Title" (or H:MM:SS if the video runs over an hour), in
chronological order and unnumbered. Don't invent content: if a
block has no clear topic, merge it into the previous one.
Transcript: [paste the timestamped transcript here]

Audit the list before publishing

Review this YouTube chapter list and verify: (1) that the first
chapter starts exactly at 0:00; (2) that there are at least
3 chapters; (3) that no chapter lasts less than 10 seconds
(compare each timestamp with the next); (4) that the timestamps
are in strict chronological order and none exceeds the video's
duration, which is [duration]; (5) that no title is empty or
duplicated. Return the corrected list and below it one line per
problem found, or "NO ISSUES" if everything passes.
List: [paste the chapter list here]

Rewrite the titles for search

These are the chapters of a YouTube video about [topic] whose
main keyword is [keyword]. Rewrite only the titles that need it
so that they: (1) describe the block's content in search
language, the way someone would type it into the search bar;
(2) include the keyword or a natural variant in 2-3 chapters at
most, without forcing it into all of them; (3) don't exceed
6 words; (4) don't repeat the same formula across all chapters.
Keep the timestamps intact and preserve the titles that already
work. Mark the ones you change with *.
Chapters: [paste the chapter list here]

YouTube's non-negotiable rules

YouTube activates chapters only if the list in the description meets the full format. This is what it checks:

• The first chapter starts at 0:00. Exactly 0:00 — a first chapter at 0:01 or 0:30 disables the entire feature. If your video has an intro, the 0:00 chapter is called "Introduction" and the next one starts where the content begins.
• Minimum three chapters. With one or two, YouTube doesn't split the bar. If your video can't fill three blocks with their own titles, it probably doesn't need chapters.
• At least 10 seconds per chapter. Two timestamps too close together invalidate the list. In practice, useful chapters last minutes, so this limit only gets hit through timestamp errors — the audit in prompt 2 catches it.
• One chapter per line, timestamp first. The format is "timestamp space title", one line per chapter, in chronological order. Both M:SS and H:MM:SS work; the timestamps must be plain text in the description, not in a pinned comment.
• Manual overrides automatic. If you write chapters in the description, they replace any YouTube would have generated. And you can disable the automatic ones in the video settings if you'd rather have none.

Beyond YouTube: SEO, accessibility and other platforms

The same timestamped transcript that produces the chapters feeds everything else:

• Key moments on Google. With the chapters published, Google can link directly to the exact second of each topic from its results. For "how to do X" type searches, that deep link competes at an advantage against unchaptered videos.
• Subtitles in the same pass. The transcript that generated the chapters converts into SRT or VTT subtitles with no extra work — the workflow is in the guide on creating SRT and VTT subtitles with AI. Chapters for navigating, subtitles for watching without sound: both layers come from the same text.
• Real accessibility. Chapters help every viewer, but especially anyone navigating with a screen reader or needing to locate a specific point without scrubbing the progress bar. Clear structure is accessibility.
• Spotify and course platforms. Spotify supports chapters in podcast episodes with the same timestamp + title scheme, and course platforms divide lessons with the same markers. Change the output format in the prompt and reuse the rest of the workflow.
• Repurposing with coordinates. A chaptered video is a potential clip index: each chapter marks a self-contained segment, a candidate for a Short or a social post. The guide on content repurposing: 1 audio, 10 pieces picks up from here.

Common mistakes that break chapters

• Starting the first chapter after 0:00. It's the number one failure: the entire list gets ignored and the bar doesn't split. The first chapter is 0:00 even if it's just the intro.
• Chaptering the raw footage instead of the final edit. Every editing cut shifts the timestamps: chapters calculated on the original recording land misaligned in the published video. Always transcribe the final version.
• Accepting timestamps the AI can't know. If the model works without a timestamped transcript, it makes up the positions. Every timestamp must exist in the transcript — and the verification pass in step 4 isn't optional.
• One chapter per paragraph. Thirty chapters in a twenty-minute video don't help navigation: they bury the structure. One chapter per real topic change; if two blocks share a topic, merge them.
• Vague titles. "Part 2" or "Moving on" tell neither the viewer nor Google anything. Each title must answer "what's here?" in 2-6 concrete words.
• Clickbait in chapter titles. A chapter that promises what the block doesn't contain generates the jump, the immediate drop-off… and distrust for the rest of the list. The honest title retains more than the flashy one.

Frequently asked questions