Get Started
Long-form YouTube for builders

Mission 1 of 6

Plan a long-form channel that makes sense: the promise, the formats, ten videos.

A reel that goes big is a lottery ticket. Long-form is a different game with different rules. Plan it properly before the camera, then leave with ten videos mapped out.

55 minBeginnerFreeFree mission

What you'll do

Learn how long-form gets watched

See how search, suggested and subscribers each send viewers, and why retention matters more than views for a new channel.

Write your channel promise

Fill in a one-sentence template and test it against a rule that catches vague promises.

Build a ten video plan

Use outlier research and three faceless formats to fill a plan.md, then choose and justify your first video.

  • Pick up a script that shows which videos beat a channel's own median
  • Learn the three faceless formats that work for builders
Before you start
  • A YouTube channel, even an empty one. Create it in five minutes at studio.youtube.com if you don't have one
  • Python 3 and yt-dlp installed for the research step: pip install yt-dlp or pipx install yt-dlp
  • Honesty about what you can actually build and show on screen. Long-form exposes bluffing in a way 30-second reels don't
Your progress0/6
01/06Why long-form is a different game from shorts0 done
STEP 01 of 06~10 min

Why long-form is a different game from shorts

If you carry short-form instincts into long-form you'll make the wrong videos. Know the three differences before you plan anything.

A reel that does a million views is a lottery ticket. A 12-minute video that ranks for 'best local AI coding tool' pays out every week for two years. They are different games with different rules.

The creators who try to play long-form like short-form post six videos, get 200 views each and quit. This mission is the plan before the camera, and it starts with how the game is different.

How shorts get distributed

Shorts are distributed to strangers by a feed that tests a clip on a few hundred people and widens it if they stay. You've lived this: the hook is everything, the viewer has no idea who you are, and a video is dead or viral within a day.

How long-form gets distributed

Long-form is distributed three ways and none of them work like that.

  • Search. Someone types 'ollama vs lm studio' into YouTube or Google. Your video shows up because the title, the first 30 seconds and the transcript match what they typed, and because people who clicked it before stayed. Search traffic arrives slowly, lasts for years, and is the reason a builder channel can earn from a video made eighteen months ago.
  • Suggested. Your video appears next to or after another video. YouTube picks it because viewers of that other video tended to watch yours too. This is where most big views come from and it depends on two things you control: the thumbnail and title (do people click) and the watch time (do they stay).
  • Subscribers and the home feed. Your existing audience is shown the video. For a new channel this is nearly zero, which is why a brand new long-form channel should not expect anything from the first few uploads and should judge them on retention, not views.

Session time

YouTube wants the viewer to keep watching YouTube. A short that gets swiped past cost nothing. A 12-minute video that someone leaves after 40 seconds is a bad signal for your whole channel, because it ended a session.

The flip side: a video that keeps people for 8 of its 12 minutes and sends them to your next one gets pushed harder than its view count suggests. Retention is the currency. Views are a consequence.

Subscriber intent

A subscriber from a reel wanted more reels. YouTube Studio separates Shorts and long-form audiences in Analytics for exactly this reason: they typically don't overlap much, and a channel with thousands of Shorts subscribers often sees a small fraction of them open a long upload.

Treat the long-form channel as starting from zero and be pleasantly surprised if the shorts audience helps. Don't build the plan on it.

Warning
Do not mix Shorts and long-form on the same channel without a reason. If your existing page is a Shorts machine, a separate channel for long-form is the cleaner experiment: you'll get honest numbers that aren't polluted by the Shorts feed, and you won't train the Shorts audience to ignore your uploads.
Tip
Open YouTube Studio on any channel you have, Analytics, Content, and filter by Shorts versus Videos. Even on a small channel, the traffic source breakdown is different for the two. That's the whole lesson in one screen.
Checkpoint
You can say, in one sentence each, where a long-form view comes from (search, suggested, subscribers) and why retention matters more than views for a new channel.
If it breaksI have 43k followers on Instagram, surely that's a head start? + 1 more
I have 43k followers on Instagram, surely that's a head start?
It's a head start for the promise and the topics, because you know what your audience responds to. It is not a head start for distribution. Instagram followers don't get notified about YouTube uploads. Send them deliberately (story link sticker, a reel that trails the long video) and expect single-digit percentages to cross over.
Should I post Shorts on the new channel to grow it faster?
Not in the first ten videos. Get the long-form retention right first, then use Shorts as trailers that point at specific long videos. Mission 6 covers repurposing properly.
Still stuck? Ask the community
STEP 02 of 06~8 min

Write the channel promise

One sentence that tells a stranger what the next five videos will be about. Everything downstream, from titles to thumbnails, is judged against it.

Now that you know how long-form gets watched, you need a reason for people to keep coming back. That reason is the promise.

A channel without a promise is a playlist of unrelated videos. Nobody subscribes to a playlist. A promise is the reason someone who liked one video expects to like the next one.

The template

  1. Copy this into a file called promise.md and fill in the template:

    promise.md
    [Who it's for] who want [the outcome], shown by [how you do it], without [the thing they hate].
    
    Example (builder, no face):
    Developers and tinkerers who want to know which AI tools are actually worth
    installing, shown by building something real with each one on screen,
    without the sponsored hype.
    
    Test: would someone who watched one video know what the next five are about?

The four slots

Here's what each slot does. Who keeps you from making videos for everyone, which is videos for no one. Outcome is what they get, not what you do.

How is your format and it's where a faceless builder's edge lives: 'by building something real with it on screen' is a promise a talking-head reviewer can't match.

Without is the thing your lane does badly that you refuse to do. For AI tools it's usually hype, sponsored takes, or the 'top 10 tools' listicle that never installs any of them.

The test

Imagine someone just watched your video about one tool. Could they guess what the other nine videos on the plan are? If the promise is 'AI stuff' they can't.

If it's the example above, they can: more tools, each one actually used, each one honestly judged. That predictability is what a subscription is.

Start from your short-form lane

Your short-form page already has a promise, even if you never wrote it down. If you did the pick a lane mission, start from that lane and narrow it: the long-form promise should be a subset, the part that needs ten minutes to deliver.

Tip
Say the promise out loud. If it takes a breath in the middle, it's two promises. Cut one.
Checkpoint
A promise.md file exists with one sentence in the template shape, and it passes the 'could they guess the next five' test.
If it breaksMy promise sounds like three other channels + 1 more
My promise sounds like three other channels
Good, it means the lane has demand. Sharpen the 'how' and 'without' slots. Those two are where you differ. 'Shown by building something real' versus 'shown by reading the landing page' is a different channel.
I want to cover AI tools and also robotics and also self-hosting
Pick the one you can make ten videos about without running out. Write the others down for a second channel in a year. Channels that span three topics teach the algorithm nothing about who to show them to.
Still stuck? Ask the community
STEP 03 of 06~12 min

The three video types that work without a face

Faceless long-form fails when it tries to be a talking head without the head. These three formats don't need one.

A face on camera buys trust and holds attention through the slow parts. Without one you need a format where the screen itself is the thing worth watching.

These three are the ones that work for builders, and your ten video plan will use all of them.

The screen-led build-along

'Build a local AI agent that reads your files, from a blank folder.' The viewer watches you go from nothing to a working thing. The screen is the star: terminal, editor, browser, and the moments where it breaks and you fix it. Length: 10 to 20 minutes.

What makes it work: the result is shown in the first 20 seconds, the steps are chaptered so a viewer can jump to the part they need, and every step ends with something visibly running.

What kills it: typing in silence, long installs shown in real time, pretending nothing went wrong. Search-friendly because the title is a thing people type.

The deep tool teardown

'I used [tool] for a week of real work, here's what it's actually like.' One tool, properly used, honestly judged. Length: 8 to 14 minutes.

Structure: what it claims, what you built with it, three things it did well with the screen to prove it, three things that annoyed you with the screen to prove it, who should install it, who shouldn't.

What makes it work: specificity. Not 'the UI is nice' but 'it took four clicks to do the thing the landing page says is one click, watch'.

What kills it: reading the feature list off the website. Search-friendly: tool name plus 'review', 'vs', 'worth it'.

The 'I tested X for a week' report

'I ran five open source alternatives to [paid product] for a week. One of them replaced it.' Several tools, one job, a clear winner. Length: 10 to 16 minutes.

Structure: the job, the contenders in one screen, a test you ran on all of them, the elimination in order of worst to best, the winner with the proof, the catch.

What makes it work: the same test applied to every tool, so the viewer trusts the ranking. This is your tier-list reel grown up.

What kills it: ten tools at a minute each, which is a listicle. Five tools, properly, beats ten tools, skimmed. Browse-friendly: the thumbnail writes itself (five logos, one circled).

What all three share

  • All three put the result first. Faceless viewers have nothing else to trust in the first ten seconds, so show them the thing working before you explain anything.
  • All three are chaptered. YouTube shows chapters in the progress bar and in search results, and a skipping viewer who finds their chapter is a viewer who stays.
  • All three can be voiced after recording. You record the screen, then write the script to what happened, then record the voice. That's the secret to clean faceless videos and it's how mission 3 is set up.
Warning
Avoid the fourth type you'll be tempted by: the news round-up ('this week in AI'). It's a race against channels with staff, it has a shelf life of three days, and it teaches viewers to come to you for news instead of for your judgement. One a month at most, only if it drives the other three.
Checkpoint
You can name a topic from your own lane for each of the three formats, and say which traffic source (search or browse) each one is built for.
If it breaksMy lane is more opinion than building, can I just talk over slides? + 1 more
My lane is more opinion than building, can I just talk over slides?
Only with a very strong voice and tight editing, and it's the hardest faceless format to retain on. Slides are a wall. If you must, make the slides move (one idea per screen, animated in, see HTML to video) and keep it under 8 minutes.
Do I need to be an expert to do a teardown?
You need to have actually used the thing for the time you claim. The viewer wants an honest user, not a maintainer. Say what you don't know on screen; it reads as trust, not weakness.
Still stuck? Ask the community
STEP 04 of 06~12 min

Find topics by reading outliers, not trends

A video that does ten times a channel's median tells you what that audience wants. Find those on five channels in your lane.

You have a promise and three formats. Now you need topics, and a video that did far better than usual on someone else's channel is a good place to look for them.

Why outliers

Trend lists tell you what's crowded. Outliers tell you what's working. On any channel, most videos sit near a median and a few are multiples of it.

The multiples are where the demand is. You're going to measure that on the channels in your lane instead of eyeballing it.

The script

  1. Install yt-dlp if you haven't (pip install yt-dlp), then save this as outliers.py.

    It reads public listing data only (title, views, duration) and skips anything under four minutes so Shorts don't skew the median.

    outliers.py
    # outliers.py: which videos on a channel beat that channel's own median?
    # usage: python outliers.py "https://www.youtube.com/@SomeChannel/videos"
    import json, subprocess, sys, statistics
    
    url = sys.argv[1]
    cmd = ["yt-dlp", "--flat-playlist", "--playlist-end", "60", "-J", url]
    data = json.loads(subprocess.run(cmd, capture_output=True, text=True, check=True).stdout)
    
    vids = []
    for e in data.get("entries", []):
        views = e.get("view_count")
        dur = e.get("duration") or 0
        if views is None or dur < 240:        # skip Shorts and anything under 4 minutes
            continue
        vids.append((views, dur, e.get("title", "")))
    
    if not vids:
        sys.exit("no long-form videos found")
    
    median = statistics.median(v for v, _, _ in vids)
    print(f"{len(vids)} long videos, median {median:,.0f} views\n")
    for views, dur, title in sorted(vids, reverse=True)[:12]:
        print(f"{views / median:5.1f}x  {views:>10,}  {dur // 60:>3}m  {title[:70]}")
  2. Point it at a channel's videos tab:

    Terminal
    python outliers.py "https://www.youtube.com/@SomeChannel/videos"

    You get the number of long videos, the channel's median, and the top videos as multiples of that median. Here's what the output looks like:

    What the output looks like (illustrative numbers)
    58 long videos, median 41,200 views
    
     11.4x     469,000   14m  I replaced VS Code with this open source editor for a week
      7.9x     326,000   11m  Ollama vs LM Studio: which one should you actually install
      5.2x     214,000   17m  Build a local AI agent that reads your files (full setup)
      1.9x      78,000    9m  ...

Run it on five channels

  1. Repeat that for five channels in your lane, including two that are smaller than you'd like to be and two you'd like to become.

    Smaller channels' outliers matter more: a 10x video on a 5,000 subscriber channel is pure topic demand, not brand.

  2. Paste the top five from each into a research.md file.

Read the titles as a group

Three patterns usually appear: a comparison ('X vs Y'), a replacement ('I replaced X with Y'), and a full setup ('from scratch', 'complete guide').

Note which tools recur. Those are the tools people are typing into search right now. Your ten videos should hit the same demand with your angle, not copy the titles.

Add your own data

Cross this with what you already know. Your own Shorts data is the best research you have: the reels that hit tell you which tools and which framings your audience responds to.

A reel about a tool that did 1.8 million views is a 10-minute video about that tool that already has its title tested. Write those next to the outliers.

Tip
Also type your lane's main tool names into YouTube search and read the autocomplete. Those suggestions are real queries ranked by volume. 'ollama' followed by 'vs lm studio', 'on windows', 'with gpu' is three video ideas with proven search intent.
Warning
This is reading public listing pages, which yt-dlp does by fetching them like a browser would. Don't hammer it: five channels, once, is fine. Don't download anyone's videos for reuse; that's their content, not your research.
Checkpoint
research.md has at least 15 outlier titles from five channels, plus your own top five reels, and you've marked the three patterns and the recurring tools.
If it breaksyt-dlp errors with 'Sign in to confirm you're not a bot' + 2 more
yt-dlp errors with 'Sign in to confirm you're not a bot'
YouTube occasionally challenges datacentre IPs and heavy use. Run it from your normal home connection, once. If it persists, add --cookies-from-browser firefox (or chrome) so it uses your logged-in session. Don't loop it.
The channel URL gives me zero long videos
Make sure the URL ends in /videos, not /shorts or the channel root. Some channels put long content on a second channel; check the About tab.
view_count is None for everything
Flat playlist extraction sometimes misses it on certain channel layouts. Drop --flat-playlist and lower --playlist-end to 30; it's slower but reads each video's page.
Still stuck? Ask the community
STEP 05 of 06~10 min

Write the ten video plan

Ten titles, three formats, weighted towards search, each one tied to evidence. This is the document you'll execute for the next ten weeks.

You have a promise, three formats and your research. This step turns them into one document you can work from.

Ten is the number because it's enough to learn from and short enough to finish. Fewer and you quit before the search traffic arrives. More and you're planning instead of making.

  1. Save this as plan.md and fill it in from research.md:

    plan.md
    # Ten video plan: [channel name]
    
    Promise: [paste your one-sentence promise]
    
    | # | Type       | Working title                                            | Search or browse | Shorts that proved it |
    |---|------------|----------------------------------------------------------|------------------|-----------------------|
    | 1 | Teardown   | [Tool] honestly reviewed after a week of real use         | search           | reel 14, 220k         |
    | 2 | Build      | Build [thing] with [tool] from a blank folder             | search           | reel 9, 1.8M          |
    | 3 | Report     | I tested 5 open source [category] tools, one is worth it  | browse           | tier list reels       |
    | 4 | Build      | ...                                                        |                  |                       |
    | 5 | Teardown   | ...                                                        |                  |                       |
    | 6 | Report     | ...                                                        |                  |                       |
    | 7 | Build      | ...                                                        |                  |                       |
    | 8 | Teardown   | ...                                                        |                  |                       |
    | 9 | Report     | ...                                                        |                  |                       |
    |10 | Build      | ...                                                        |                  |                       |
    
    Rules: 4 builds, 3 teardowns, 3 reports. At least 6 with a search term in the title.
    Every title names the tool or the outcome. No "episode 1".

Why the rules matter

The rules at the bottom are not decoration. Four builds, three teardowns, three reports gives you a balanced test of which format your audience wants, which you'll only know after all ten.

Six with a search term in the title means that even if suggested traffic never arrives, six of these will keep finding viewers for years. Every title names the tool or the outcome because a new channel has no brand to trade on; the title has to do all the work.

Use your Shorts as evidence

The 'Shorts that proved it' column is your unfair advantage over every other new long-form channel. You already know which tools and framings your audience responds to.

A reel that did 220k on a tool is evidence that a long video on the same tool has an audience. Fill that column for as many rows as you can. Rows with no evidence are bets; keep them to three or four.

Order

The sequence matters a little. Put a build-along first (it's the format most likely to retain, and retention on the first video sets the tone), a teardown of your most-proven tool second, and a report third. After that, alternate.

Tip
Each row should be a video you could record this week with what you already know. If a row needs a month of learning first, it's a video for plan two. Swap it.
Checkpoint
plan.md has ten rows, the format counts are 4/3/3, at least six titles contain a search term, and at least half the rows cite a reel or an outlier as evidence.
If it breaksAll my ideas are teardowns + 1 more
All my ideas are teardowns
Teardowns are the easiest to plan and the hardest to retain on, because the viewer has to trust your judgement without seeing you. Turn two of them into builds ('I built X with it') and one into a report by adding two competitors.
I can't find search terms for my titles
Type the tool name into YouTube search and take the autocomplete suggestions. If a tool has no autocomplete at all, it's too early for search; make that video a browse bet and give it a strong thumbnail idea instead.
Still stuck? Ask the community
STEP 06 of 06~3 min

Choose the first video and write down why

Commit to one. The next five missions turn it into a published video.

A plan only matters if you start it. This last step is a single decision.

  1. Pick row one. It should be a build-along on a tool you've already used, with a reel behind it if you have one.

  2. Write under the table: 'First video: [title]. Because: [the evidence].'

    That sentence stops you second-guessing for the next week.

Set expectations

Be honest with yourself about what to expect. The first video on a new channel will typically get a small number of views from search and almost nothing from suggested.

You'll judge it on two numbers: how much of it the average viewer watched, and whether the retention graph has a cliff in the first 30 seconds. Those tell you whether to change the script or the packaging. Views come after the tenth video, not the first.

Where next

Next: write a script that holds attention for 12 minutes. It's the structure that makes retention happen on purpose instead of by luck.

Checkpoint
plan.md ends with 'First video:' and 'Because:' lines, both filled in.
Stuck? Ask the community

Your first wins

It works. Now do something with it.

Pressure-test the promise

Give the model your promise and let it find the vagueness:

Prompt
Here is my YouTube channel promise: [paste]. My lane is [lane] and I never show my face. Tell me: (1) what the next five videos would obviously be about if you only had this sentence, (2) which of the four slots (who, outcome, how, without) is weakest and a sharper version of it, (3) two existing channels this would be confused with and the one change that separates it. Be blunt.

Turn research into titles

Paste research.md and get ten candidates per format:

Prompt
Here are outlier video titles from five channels in my lane, plus my own top Shorts: [paste research.md]. My channel promise is: [paste]. Give me ten build-along titles, ten teardown titles and ten week-long test report titles for MY channel, each naming a specific tool or outcome, each with a search term a real person would type. Don't copy any title in the list; use the demand it shows. Mark your five strongest with a reason.

Undo it cleanly

How to remove everything this mission installed.

Nothing installed except yt-dlp (pip uninstall yt-dlp removes it). promise.md, research.md and plan.md are yours. The channel can be deleted from YouTube Studio settings if you change your mind.

Next mission · Long-form YouTube for buildersWrite a script that holds attention for 12 minutes without a face on screen
→