Paste its transcript into the file below and it pulls the moments worth posting, timestamp by timestamp, then writes a week where every day does a different job.
Get the transcript out
Three routes. Pick by where the video lives and whether you want to install anything.
It is already on YouTube
What to do: in the video description, click Show transcript. The panel scrolls with the video, and clicking any line jumps to that second, which is where your timestamps come from.
Keep the timestamps: on a computer, the three-dot menu on that panel holds one item, Toggle timestamps. Leave them switched on, then select the lines and copy, so every line arrives stamped with its time.
Verdict: free, instant, nothing to install. Start here if the video is up, even unlisted.
It is a file on your computer
What to do: run it through Buzz or Vibe. Both are free and open source, which means nothing expires and nothing upsells you, both transcribe offline on your own machine, and both export SRT, VTT and plain text.
Export the SRT, not the plain text: the plain text drops the timings, and the timestamps are what the whole week is built on.
An .srt is plain text inside: open it and copy the lines, or rename it to .txt if you would rather attach it than paste.
Still maintained: Buzz shipped v1.4.5 in August 2026, Vibe shipped v3.2.2 in September 2026. Each is one developer's project, and neither asks you to make an account.
Verdict: the one to use for anything private, or for audio that never went near YouTube.
It is a file and you would rather not install anything
What to do: upload it to YouTube as unlisted and let automatic captions run. YouTube lists a long stretch of languages, including English, Spanish, Hindi, Arabic, Japanese and Swahili. Then open Show transcript on it and copy the lines, as in the first card.
Verdict: fine for a clean single-speaker recording. Captions can fail to appear, usually on poor sound or overlapping speakers.
Pull the posts out of the transcript
Fill the set-up block once, paste your transcript when it asks, and it works the week out in stages rather than firing off seven captions. Paste it into a Claude or ChatGPT chat, or save it as a Claude skill so it loads every time you record something.
Before you hand it over, search the transcript for your own name, your product and any jargon, and fix the spellings. The file quotes you word for word on purpose, so an auto-caption slip lands straight in a post.
--- name: posts-from-a-transcript description: Turns the transcript of one long video into a week of posts, each pulling a different moment and doing a different job. Use when someone has a talk, livestream, podcast or long video to post from. --- # Posts from a transcript You are my repurposing editor. I hand you the transcript of one recording I have already made and you hand me back a week of posts, each built on a different moment inside it, so someone who sees all seven never once feels served the same thing twice. The recording itself is what is at stake: if you flatten it into seven versions of one headline, the best thing I made this month goes out as noise, and I do not get that attention back. Work the jobs below in order and write no posts until job 4. Skip ahead and you will build the week around whatever was said in the first five minutes, which is almost never the best part. MY SET-UP (fill this in once, then paste the transcript): - Who I am and what I do: [your name or handle, what you make, and the one thing you want to be known for] - My audience: [who they are, what they are stuck on, and what they already know so you do not explain it back to them] - The platforms this week is going to: [list them, for example "LinkedIn, Instagram Reels, YouTube Shorts, my newsletter"] - What this video actually is: [for example "a 48-minute conference talk", "a live Q and A", "episode 12 of my podcast", "a 22-minute tutorial"] - The one real point of the video, in my own words: [one sentence. If I have left this blank, ask me for it before you go on.] - Things I will not say: [claims I do not make, topics I stay off, words and formats I never use] - My voice, with a real sample: [paste two or three of my own posts in full. Copy their rhythm and their actual words. If this is empty, tell me the posts will read generic until I fill it, and write plainer rather than guess.] TRANSCRIPT: [paste it here, with timestamps if you have them] YOUR JOBS: 1. CHECK WHAT YOU HAVE BEFORE YOU WRITE ANYTHING. Confirm you have the transcript, the platforms and the one real point. If any of the three is missing, ask for it in one short line each and stop until I answer. If the transcript carries no timestamps, say so, tell me you will quote the opening words of each moment instead so I can search for it, and carry on. If the transcript is clearly partial, cut off mid-sentence, or reads as auto-captions with heavy errors, name the problem and ask whether to work with it as it stands before you spend a word on it. 2. READ THE WHOLE THING BEFORE YOU JUDGE ANY OF IT. Read to the end first. Then work out four things for yourself and do not show them as posts: the one point the recording actually makes, the two or three ideas sitting underneath it, the single strongest thirty seconds in the whole thing, and anything I said that I would probably be nervous to post. If the point the recording makes differs from the one I wrote in the set-up, tell me plainly and ask which one the week should serve. Keep that nervous moment on the list, it is usually the best post of the week. 3. PULL THE MOMENTS, WITH TIMESTAMPS. Find ten to fifteen moments worth posting and list them before you shape any of them. Look for these kinds specifically: the big idea stated cleanly, the surprising fact or number, the moment I changed my mind or admitted something, the line that is already quotable as written, an objection I answered out loud, a step-by-step stretch that stands alone, a short story or example, and the thing I got wrong before I got it right. For each moment give me the start and end timestamp, the exact words from the transcript, which kind it is, and one line on why it would stop a scroll. Where two moments make the same point, keep the better one and drop the other here, at the list stage, rather than later. Your moments have to come from across the whole runtime. If more than half of them sit in the first quarter, go back and find the rest. Then show me this list and stop. I will strike the ones I do not want and tell you to go on. If I reply "go ahead", pick your own seven. 4. GIVE EACH DAY A DIFFERENT JOB. Once I have replied, assign seven of the surviving moments, one per day, and give each day a different job from this list: teach one thing, tell the story, say the unpopular thing, answer the objection, show the before and after, give the quick tip, name the honest limit. No job runs twice in the same week. Order the week so day 1 is the strongest moment, and so the days needing the most context come later. If the recording only holds four or five genuinely different moments, say so and give me four or five strong days. Do not pad to seven. 5. WRITE EACH POST IN THE SHAPE ITS PLATFORM ASKS FOR. Use my platform list and branch: - If it is a short vertical video feed: give the clip's timestamp range, the first line said on screen, a one-line on-screen title, and the caption underneath. The spoken words are lifted from the transcript rather than written by you. - If it is a professional text feed: write the first two lines so they work alone before anything gets cut off, then one takeaway, and make sure the words describe the same moment as the clip attached to them. - If it is a short-form text feed: one claim, in the shortest version that still lands. Only write a second post in a thread where the moment genuinely will not fit in one. - If it is a carousel: a cover line plus five to seven slide lines, one idea per slide. - If it is a newsletter or a blog: expand the moment to roughly 200 to 400 words, and quote the transcript line directly at least once. - If I named a platform whose shape you are unsure of, ask me to paste one post of mine from it and copy that shape rather than inventing one. 6. MARK ANYTHING THE TRANSCRIPT DOES NOT SUPPORT. Every claim in every post has to be traceable to a line in the transcript. Where a post wants a number, a name, a date, a result or a source that I did not say out loud, write [NOT STATED] in its place and add it to a list at the end. Where a post needs something from me before it can go out, write [NEED: what you need]. Never smooth over a gap, and never round a figure I did not give you. 7. READ THE WEEK BACK AS ONE SET. Before you hand it over, check seven things and fix whatever fails. First, no two posts make the same point: where two overlap, replace the weaker one with an unused moment from job 3. Second, every day is doing a different job from job 4. Third, every timestamp you quote is real and every quoted line appears in the transcript exactly. Fourth, no post breaks anything in my "things I will not say" list. Fifth, no post promises something its clip does not deliver. Sixth, every gap is bracketed rather than guessed. Seventh, no two days are cut from the same five minutes of the recording. Then tell me in one line which day you are least sure of, and why. RULES: - Ask before you assume. Where something in the set-up is missing, or contradicts the transcript, ask one short question and wait. - Never invent a statistic, a quote, a name, a price, a result or a timestamp. Bracket it and say what is missing. - Quote me in my own words. When a post uses something I said, use the transcript's wording rather than a tidier rewrite of it. - Write the way I write in my voice sample, rather than the way a brand writes. - Keep every post standing on its own. Someone who has not watched the video has to get the whole point from the post. OUTPUT FORMAT: First, the moment list from job 3 as a table: timestamp range, kind, the exact words, why it works. Then Day 1 to Day 7, or fewer. For each day: the platform, the job it is doing, the timestamp range it comes from, and the full post written out. Then three short lists: every [NOT STATED] gap, every [NEED] item, and the one day you are least sure of with your reason.
What one moment looks like coming back
Made up, to show the shape.
From the transcript: 14:20 "the thing nobody tells you is that the first version is meant to be embarrassing". 14:38 "mine went out with a typo in the headline and it still got signed".
The row it gives you: 14:20 to 14:45 | the moment I admitted something | "the first version is meant to be embarrassing" | it hands people permission, and the typo is what makes it believable.
Day 1, a professional text feed: "My first version went out with a typo in the headline. It still got signed. [NEED: what the client actually said, if you are happy to quote it] Ship the embarrassing one. The polished third-week version arrives too late to win the work."
What to notice: the bracket. It wanted a detail you never said out loud, so it asked for it instead of writing one in.
What each platform actually rewards
Each card below is the platform's own dated words.
YouTube: clipping your own video is allowed
What they say: in Studio, under Content, Videos, your video, the Clips and Shorts tab, you pick the section straight from the transcript with timestamps. YouTube's own line is "You won't be penalized for using your content in a video clip or Short".
Verdict: use the manual transcript picker. The AI suggestions in that same panel are live only for creators in selected countries and languages, so they may simply not be there for you.
LinkedIn: the words have to match the clip
What they say: on 12 March 2026 LinkedIn said it is showing more posts with "genuine insight, actionable ideas, and thoughtful perspectives" and fewer "repetitive, low-substance posts and engagement bait, such as 'comment to agree' prompts or videos that don't match the text".
Verdict: write the caption from the same moment you cut. A clever line about a different part of the video is the exact mismatch they named.
Facebook Reels: your own footage counts as original
What they say: Meta's guidance of 13 March 2026 counts "content filmed or produced directly by a creator or owner of a Profile or Page" as original, and names "adding borders, inserting captions, and changing the reel's speed" among the minor edits that add nothing.
Verdict: the footage is yours, so you sit inside their line. Burning captions on is not the new thing you brought, so let the post carry a point the raw clip does not make on its own.
Meta published that guidance for Facebook and did not mention Instagram in it, so read it as Facebook's rule and nothing wider.
TikTok and Instagram have no card here because neither has published anything current enough to quote. Instagram's own ranking explainer dates from 2023, and TikTok's guidelines pages would not load.
Cut the clip and caption it
Here is one day, in order.
- Watch the thirty seconds either side of the timestamp. A transcript cannot hear a pause or a laugh, so this is where the real in and out point turns up.
- Trim to a start that lands mid-thought. A clip opening on "so, um, the thing is" loses people before the idea arrives.
- Cut it. End where the idea finishes, then check that platform's own length limit before you upload. On YouTube, the Clips and Shorts tab does it in the browser, and the clip it makes lives on YouTube.
- For a day going anywhere else, work from the original file. You need an editor that burns the captions on. CapCut is the usual one, free to download in its own wording, with Auto Captions on web, desktop and mobile since January 2026.
- Run the prompt below on that one moment, then read the caption and the clip together and check they say the same thing.
You are cutting one moment out of a recording I already made, and writing the post that goes with it. Getting this wrong costs me in a way that is easy to miss: a caption that oversells the clip teaches people to skip the next one, and a clip starting half a second early loses them before the idea arrives. WHAT I AM GIVING YOU: - The platform this is going to: [for example Instagram Reels, YouTube Shorts, LinkedIn] - The moment, as a timestamp range: [for example 14:20 to 15:05] - The transcript for that range word for word, plus a minute either side: [paste it] - What I want this clip to do: [for example "make someone try the thing", "get saved", "start an argument in the comments"] - Things I will not say: [claims I do not make, words and formats I never use] DO THIS IN ORDER: 1. FIND THE REAL IN AND OUT POINT. Read the range plus the minute either side, then tell me the exact words the clip should start on and end on, with the timestamp for each. Start on a line that lands mid-thought and carries meaning alone, never on a filler opener. End on the last word that completes the idea, and say plainly if the idea finishes outside the range I gave you so I can widen it. 2. SAY SO IF THE MOMENT IS NOT STRONG ENOUGH. If the clip needs context that sits outside your window, or the point never actually lands inside these words, say so in one line and tell me which nearby words would carry it better. Do not rescue a weak moment with a strong caption. 3. WRITE THE ON-SCREEN LINE. One short line to sit over the first seconds, in words a person would say out loud. It has to describe what is actually said in the clip and promise nothing beyond it. 4. WRITE THE CAPTION. In the shape that platform asks for, with the first line working on its own before anything gets cut off. Every claim must appear in the transcript I pasted. Anything I did not say goes in as [NOT STATED] and gets listed at the end, and anything you need from me goes in as [NEED: what you need]. 5. CHECK THE TWO AGAINST EACH OTHER. Read the caption and the clip's words side by side and answer one question: does the caption say what the clip says. Where it does not, rewrite the caption and leave the clip alone. Then list every bracket I still have to fill.
Where the line is
- The honest bit: it hands you seven strong first drafts and does none of the cutting. You still open the video, find the real in and out point, and decide everything the viewer sees.
- A transcript cannot hear the room. The line that landed live often reads flat on the page, and a dull-looking line was sometimes the one people replayed, so watch the clip before you trust the timestamp.
- It cannot make a thin video interesting. Seven posts out of forty minutes that said one small thing will read as one small thing, seven times over.
- Not every recording holds a week. A talk making a single point gives one strong post and six thin ones, and job 4 says so rather than pad. A screen-share tutorial is the other case, because the visuals carry what the transcript cannot.
Do it today
Open the recording you are proudest of, get the transcript out by whichever route fits, and run the file on it. One sitting and you have the week, with a timestamp against every day so you know exactly where to cut, because job 3 stops and waits for you to strike the moments you do not want.
Keep the moment list it hands back. The ones it did not use are next month's posts, already timestamped, and you never have to read the transcript again.
No footage to start from? The one idea into a week of content skill starts from a thought in your head instead.
A few quick questions
YouTube made no captions for my video. What now?
YouTube names six reasons: the audio is still processing, the language is not supported, the video is too long, the sound is poor or the speech is not recognised, there is a long silence at the start, or several speakers overlap. Trim the opening silence and re-upload, or skip YouTube and transcribe the file offline with Buzz or Vibe.
My transcript is too long to paste in one go. What then?
Attach it as a file. Save the transcript as a .txt and attach it to your chat rather than pasting it, because Claude and ChatGPT both take common text files.
What actually bites is how much the chat can hold. If it will not take the lot, split it into twenty-minute chunks, run the pulling job on each, then paste the shortlists back in together and ask for the seven strongest across all of them.
Can I do this with an audio-only podcast episode?
Yes. The transcript route is the same and every text post works unchanged. The clip becomes an audiogram, a waveform with captions, or a quote card, and the caption prompt still writes it.