How to Plan a YouTube Video: From Idea to Script, Step by Step

What you walk away with

• An idea tested as a question a real viewer is asking, before you invest a minute of production
• A working title and thumbnail concept decided before the script, so the video keeps the promise the click was sold
• A research pass that finds the gap in what already exists, instead of remaking the third-best video on the topic
• An outline shaped by how retention actually behaves, not by how essays are structured in school
• A first thirty seconds scripted word for word
• A shoot plan generated from the script: shots, b-roll, graphics, locations
• A publish-day loop that turns the retention graph into next video's plan

The one rule that decides whether it works

Decide the packaging before you write the script.

Packaging, the title and thumbnail, is not decoration applied after the video exists. It is the promise the viewer clicks on, and the video's one job is to keep that promise. When you script first and package later, you end up hunting for a title that fits a video that exists, which produces either an honest title nobody clicks or a clickable title the video does not deliver, and the retention graph punishes the second one at second twenty-one. When the packaging comes first, the script is written to fulfill a specific promise, and the whole plan tightens around it.

This feels backwards to people who came from filmmaking or writing, and it is the single highest-leverage change in this guide.

How AI fits into planning a YouTube video

Genuinely useful: brainstorming title and thumbnail variations once you have the idea, summarizing the top existing videos on your topic so you can find the gap faster, turning your outline into a first-draft script to react against, generating the b-roll and graphics lists from a finished script, and building the shoot checklist. Mechanical expansion and compression around decisions you made.

Where it makes your video worse: generating the idea, the angle, or the opinions. Models average the internet, and the average YouTube video on your topic already exists, several thousand times. Viewers click creators for a specific human take; an AI-generated angle is by construction the take everyone else already has. Same for scripts published as generated: the writing is fluent, the voice is nobody, and watch time knows.

The tooling constraint: planning lives across ideas, research, outline, script, and shoot lists, and the tools that help most are the ones where the AI can see the whole plan, so "generate the b-roll list" works from your actual script rather than from a description of it.

1. Turn the idea into a viewer question and test it

An idea becomes a video plan the moment it is rewritten as a question a specific viewer is asking. "My trip to Iceland" is not a question. "How do you shoot a travel film alone, with one bag?" is one, and it immediately identifies who clicks: solo travelers who film, not everyone, and not nobody.

Three checks before the idea earns production time, fifteen minutes total:

• Who is asking? Name the viewer concretely: what they already know, what they typed into search, which channels they already watch. "Everyone" means the packaging will speak to no one.
• Is there evidence of demand? Search the question. Existing videos with real view counts are good news, proof of demand. Autocomplete, comment sections of adjacent videos, and your own comments are where phrasing comes from, and phrasing becomes titles.
• Why you? One sentence on what your version has that the existing ones do not: access, experience, a contrarian result, a better demonstration. If the sentence will not come, the video is a remake, and remakes underperform their effort.

Keep a running idea board and run this test the moment ideas land on it, not the week you want to film. Ideas that pass wait, validated, until their slot; ideas that fail stop costing you Saturdays.

2. Decide the packaging: title and thumbnail before anything else

Draft five to ten titles for the question, and pick the working title before you outline. A useful title names the viewer or the outcome, creates a specific curiosity, and stays honest, because retention collects on every exaggeration. Alongside it, sketch the thumbnail concept in one line: the image, the three or fewer words on it, the emotion. Title and thumbnail are one unit; they must not repeat each other, and together they are the promise.

Then write the promise explicitly at the top of your plan: "This video promises a solo traveler that they will see exactly how one person shoots a travel film with one bag." Every later decision, outline beats, b-roll, the open, gets tested against that sentence.

Two mechanics that help: say candidate titles out loud, because clickable titles survive speech and clever ones often do not, and check candidates against what is already ranking for the question, because a title identical to the incumbent's is a vote for the incumbent. This is also the one step where AI variation-generation is purely useful: feed it your best title and ask for twenty, keep two, discard the average.

3. Research the topic for the gap, not for completeness

Watch the top three to five existing videos on the question, at speed, with a note surface open. You are not learning the topic, you already chose a topic you can speak on. You are mapping what the viewer who clicks your video has likely already seen, so your video can do something other than repeat it.

Capture three things per video: what it covers well (do not remake this), what it skips or handwaves (candidate gap), and what its comments complain about or ask (the gap, stated by the audience in their own words). Comment sections under popular videos are the highest-density planning resource on the platform and most creators never mine them.

Add your own material: the demonstration only you can do, the numbers from your own experience, the mistake you made that the polished versions do not admit to. The plan's angle is the intersection: demanded by comments, skipped by incumbents, and available to you specifically. Write the angle as one line under the promise. AI is good here as a summarizer of the incumbent videos; it cannot tell you which gap is yours.

4. Outline for retention, not for essay logic

School taught a structure, intro, background, three points, conclusion, that retention graphs eat alive. Viewers do not owe you the background section, and they leave during it. Outline instead in beats, each earning the next thirty seconds of attention:

• The open (0:00-0:30). Restate the promise visually and verbally, prove the destination exists (show the result, the finished thing, the surprising number), and hand over the first piece of value fast. No channel intro, no "before we start", no logos.
• The spine: three to six beats. Each beat is one step or one revelation, opens with a mini-promise ("here is the mistake that costs most people the shot"), pays it off, and hands to the next beat with an open loop. Under each beat, bullet what must be said and what must be shown, two columns, because the second column becomes your shot list.
• Planned pattern breaks. Every ninety seconds or so, something changes: location, graphic, demonstration, story. Mark these in the outline rather than hoping the edit finds them.
• The payoff and the handoff. The moment the promise is visibly kept, then one next action. One. A video that asks for like, subscribe, comment, and two other videos gets none of them; the end screen should hand the finishing viewer the single most likely next video.

Order beats by value, strongest early, because the audience at minute two is triple the audience at minute ten, and the classic mistake is saving the best for an ending most viewers never attend.

5. Script the open word for word, then script to your format

The first thirty seconds get written out completely, every sentence, and rewritten until nothing in them is throat-clearing. This is where my twenty-one-second corpse taught its lesson: I had opened with context. Nobody wants context; they want evidence they were right to click. Draft the open, cut its first sentence, and read what remains; the open usually improves, and often it improves again when you do it twice.

Past the open, script to the format you can perform:

• Full script for essays, and for anyone whose talking-head takes wander. Write for speech: short sentences, contractions, read it aloud and cut everything your mouth stumbled on.
• Beat script for tutorials and demonstrations: the open and each beat's first and last lines written out, the middles bulleted, so explanations stay live but the structure cannot drift.
• Either way, write the b-column. Next to every scripted section, what is on screen: the demo, the b-roll, the graphic. A script without a visual column is a podcast plan.

AI's honest role: turning your outline into a first-draft script that you then rewrite into your own voice is a real accelerant, the draft exists to be argued with. Publishing its draft unedited is how channels converge on the same fluent nobody.

6. Plan the shoot from the script

Now, and only now, does production planning start, and it is mostly extraction rather than invention:

• Shot list from the b-column. Every "what is on screen" note becomes a line: the setup, the demo close-up, each b-roll idea with the beat it serves. Group by location and setup so you shoot in rig order, not script order.
• Graphics and asset list. Every chart, title card, and screen recording, listed with owner and deadline, even when every owner is you. Assets discovered during the edit cost a day each.
• The boring logistics, one line each: location and light for the date, gear and batteries, wardrobe if it matters for continuity across shooting days.
• Time the plan honestly. A talking-head beat shoots at maybe three times its runtime; demos at five to ten. If the arithmetic says nine hours and the day has six, cut beats now, at planning price, instead of on camera at day-rate price.

For a solo creator this whole step is an hour on top of the script, and it is the hour that makes the difference between a shoot day that executes and one that improvises. Keeping the script, shot list, and asset list on one surface matters here, because they are one document in three views, and they drift apart the moment they live in three tools.

7. Publish, then close the loop with the retention graph

The plan's last step happens after upload, and skipping it is why creators replan the same failures annually.

Within a few days of publishing, put the retention graph next to the outline and read it against the beats: where the open lost people (compare against your promise: too slow, or overpromised?), which beat held or bled, whether the pattern breaks show as recovery bumps, where the payoff landed versus where viewers gave up waiting for it. Write three lines onto the video's plan: what the packaging promised versus delivered, which beat structure held attention, and one thing to repeat or kill next time. Then the click-through side: if CTR was weak, the packaging step gets the attention next video; if CTR was fine and retention was weak, the outline and open do.

This loop is the actual moat. Any tutorial can teach the structure above; the creator who annotates twenty of their own retention graphs against twenty of their own plans owns knowledge about their audience that no generic advice, and no AI, has.

Common mistakes

• Planning the production and skipping the video. Shoot dates and gear lists, but no packaging decision and no scripted open. The video dies at second twenty-one.
• Scripting first, packaging later. Produces honest titles nobody clicks or clicked titles the video betrays. Packaging first; script to the promise.
• "This video is for everyone." Then the title speaks to no one. Name the viewer.
• Remaking the incumbent. Research is for finding the gap, and the gap is written in the comments of the top videos.
• The essay outline. Background sections and saved-for-last payoffs assume an audience that stays. It does not.
• An unscripted open. The first thirty seconds are the whole video's leverage; they are worth thirty scripted minutes.
• No visual column. A script that only says words is a podcast with a camera pointed at it.
• Publishing AI's draft. Fluent, voiceless, average by construction. AI drafts, you rewrite, always.
• Never reading the retention graph against the plan. The one habit that compounds, skipped because publish day feels like the finish line.

How this went for a one-bag travel filmmaking video

The idea arrived as "my Iceland trip", and step 1 rewrote it into "how do you shoot a travel film alone, with one bag?", which named the viewer (solo travelers who film) and survived the demand check: the search results were full, and their comment sections were full of the same complaint, that every "minimal travel filmmaking" video still assumed a drone, a gimbal, and a patient girlfriend holding a reflector.

That complaint became the angle, and the packaging came before a single scene was planned: working title "One Bag, No Crew: How I Shot a Travel Film Alone", thumbnail concept of the single open bag with everything visible, promise line written at the top of the board.

The outline came out at five beats, ordered by value: the bag itself first, because that was the click's curiosity, then shooting yourself without a crew, then the three shots that make any location cinematic, then sound, then the edit-on-the-road workflow. Each beat had its said/shown columns, and the pattern breaks were the location changes the trip supplied for free.

The open was scripted word for word and cut twice; it ended up opening on the finished film's best five seconds, then the bag hitting a hostel bed, then the promise sentence. The shot list extracted from the b-column ran 34 lines grouped by location, and the honest arithmetic cut one beat, sound got folded into the shooting beat, before the flight rather than in the field.

The retention graph on that one held 68% at thirty seconds and the biggest dip was in beat four's workflow section, which the notes recorded as "too much screen, not enough Iceland", a sentence that has improved every tutorial-shaped video since. That is the loop working: the plan taught the next plan.

The tools you will actually use

The plan is five artifacts: idea board, packaging, research notes, outline-plus-script, and shoot lists. The tool question is the usual one: how many surfaces do they fragment across?

• Docs and spreadsheets hold scripts and shot lists fine for a solo creator, at the cost of the visual column, the research, and the shot list living in three files that stop agreeing the week you are busiest.
• Notion is the popular pipeline home: databases for the idea backlog and per-video pages. Strong for tracking many videos; the single video's plan, outline, script, references, and shot list as one visual thing, is where it flattens into linked lists.
• Dedicated YouTube tools like VidIQ or TubeBuddy earn their keep in step 1 and 2, demand signals, title testing, competitor stats, and stop there: they are research instruments, not planning homes.
• A visual canvas as the video's plan is the structural fit for this process, and it is what Storyflow is built for: the idea board, the packaging candidates, the incumbent-video notes, the beat outline with said/shown columns, the script, and the extracted shot list on one surface per video. The AI works on the actual plan, generating title variations from your idea, first-draft scripts from your outline, and the b-roll list from your script, and can grab frames from YouTube references straight onto the board for the packaging and research steps. Free to start with unlimited boards and collaborators, paid plans from 7.99 dollars a month billed annually, per account rather than per seat, which matters the day an editor joins.
• The honest summary: a solo creator with one video a month loses nothing running this in a doc, and VidIQ knows the demand data a canvas does not. The case for one surface is the moment the pipeline is real, weekly videos, an editor, a backlog, because that is when the plan fragmenting across five tools starts costing shoot days.

You are ready

The chain is short and the order is the whole trick: a question a real viewer is asking, a promise decided before the script, a gap the incumbents left you, beats ordered by value, thirty seconds scripted like they carry the video, because they do, and a shoot plan extracted instead of improvised.

Plan the video, not just the production. Read the graph afterward like it is grading the plan, because it is.

Do that for ten videos and you will own something no tutorial and no model can hand you: a map of exactly what makes your audience stay.

Author

Justkay is a documentary filmmaker and the founder of Storyflow. He has planned videos that died at second twenty-one and videos that held two-thirds of their audience past the open, and built Storyflow so the plan, the script, and the shot list could stop living in five tools.