Onboarding videos work when each one teaches exactly one concept in about a minute, and the set is ordered so the first video teaches the object every other concept depends on. They fail when they're designed as a sit-back series, because nobody watches onboarding that way — a new user consults video three at minute forty of their trial, stuck on one specific thing, with the product open in the next tab.
That viewing behavior — reference, not campfire — should drive every design decision: length, sequencing, what gets a video at all, and what gets a tooltip instead. This guide covers each, from a production practice of sixty-odd animated software explainers, including full onboarding curricula planned as courses.
Nobody watches onboarding like a campfire
A launch video gets watched once, start to finish, by someone deciding whether to care. Onboarding videos live a different life. They sit in an empty state, a help panel, a docs page, a getting-started email — and they get opened out of order, mid-task, by someone who wants one answer and their cursor back.
Design consequences, each one concrete:
- Every video must stand alone. The published order is a dependency order, but each video self-contains its prerequisites at one-line depth ("a workflow is the thing that runs; this video is about what runs inside it"). One line, then on with it.
- No "welcome back." Our curriculum rule is verbatim: videos are watched non-linearly; only the series opener carries the full welcome and introduction. A greeting at the top of video four is dead air for the person who arrived there from a search.
- Keep them short and honestly titled. The animated concept core of our finished videos runs about 54–124 seconds, typically 60–80. A full onboarding sequence targets 10–15 minutes across four or five short videos — never one twelve-minute tour. A stuck user will spend ninety seconds on "how references work." They will not spend twelve minutes scrubbing a tour for the one part they need, and the title is what lets them pick the right ninety seconds.
- Front-load nothing. The answer the title promises should start arriving in the first sentence. A viewer who came for reference treats a slow open the way you treat a docs page that starts with the company's founding story.
One concept per video — and name what each video refuses to teach
The atomicity rule from our curriculum planning: each video owns exactly one idea and explicitly defers the neighbors. The plan literally annotates ownership — this one owns "what an agent block is"; the rest own "what you hand it." When two videos both half-teach a concept, both fail: the viewer gets two shallow passes and no complete one, and every video gets longer than its title.
The tool that enforces this is a "deliberately not taught" list, written per video before scripting. Each cut is a decision: name it, and name where it lives instead — another video, the docs, a screen recording. Scope isn't what you ran out of time for; it's a list of named exclusions with destinations. This list is the difference between five clean lessons and five overlapping half-lessons.
Atomicity also sets the depth ceiling. The job of an onboarding video is to enrich the viewer's causal model of what the product can do, in the language they already speak — with only enough mechanism to make the behavior predictable. If a video is explaining how the algorithm works under the hood, it's too low; reframe to what it does and when you'd reach for it. Inside each video, the same one-thing discipline applies scene by scene — that's the one idea per scene rule.
Sequence from the gravity center
The most common onboarding order is the tour order: whatever the navigation sidebar lists, top to bottom. The right order is dependency order, and it starts at what we call the gravity center — the one object in your product that every other concept is defined in terms of.
In one product we built a curriculum for, that object was the workflow: tables are read by workflows, logs inspect workflows, deployments expose workflows. Teach workflows first and every later video gets to build on a model the viewer already holds. Teach tables first and you're explaining a filing cabinet for documents the viewer has never seen.
Finding yours is one question: which concept appears in the definition of all the others? Then:
- Video one: the gravity center, at orientation depth.
- Group the rest into tracks by object, and give each track a stated arc — for example: what the object is, then what you do with it, then why it composes with everything else. An arc keeps a track from being a feature list read aloud.
- Keep a plan of record marking each video exists / redo / new. When the product outgrows a video, "redo" is a first-class state — a stale onboarding video teaches a product that no longer exists, which is worse than no video.
And calibrate the opener's ambition. The stated targets for our intro sequences are familiarity, orientation, confidence to click around, basic vocabulary — explicitly not mastery or coverage. The exit state we write down, verbatim: "I know what I'm looking at now. I can start exploring." An opener that tries to teach everything teaches nothing and runs six minutes.
When video beats tooltips — and when it doesn't
Video is not the default answer to onboarding. It's one tool with a specific edge, and the honest routing rule from our production planning is one line: animate the concept, record the live product — and for a good share of onboarding, don't make a video at all.
Video wins when the thing to teach is invisible in the UI. Causality, timing, structure: why a run behaves the way it does, what happens when two branches finish at different times, how your data model fits together. No tooltip can show that the merge step waits for the slow branch — a video can make you feel the wait. Animation exists to stage what a screenshot can't show. This is where a minute of video replaces a support thread.
Tooltips and docs win for procedures and reference. We keep a standing "cut — never animated" list for exactly this: setup flows, editor walkthroughs, permission tables, configuration reference — topics where no diagram earns its keep. "Click here to invite a teammate" as an animated video is expensive, ages with every UI change, and is slower to consume than the two-line tooltip. If the user's question is "where is it," the answer is an arrow, not a film.
Screen recordings take the middle. UI procedures that are too long for a tooltip get a plain screen recording — cheap to make, cheap to remake after a redesign. Save animation for concepts, which age slowly, and let recordings absorb the churn. (Your docs videos live almost entirely in this middle band.)
Two more placement rules from practice:
- Don't autoplay. A video that starts itself inside a product is an interruption, and reference viewers punish interruptions. Show a thumbnail with an honest title and a duration; let the stuck user choose it.
- Put the video where the confusion happens, not on a "tutorials" page nobody visits. The empty state of the feature it explains is worth more than a video library link in the footer.
Ground them like everything else
An onboarding video makes claims to the one audience guaranteed to check them: someone with the product open right now. Every on-screen label, value, and surface should come from a real configuration and a real run — built once by whoever operates the live product, stored, and traced. The grounding method is the same as for any product video; the stakes are just higher, because a new user who spots a screen they can't find doesn't think "artistic license," they think they're lost. That's the exact feeling onboarding exists to remove.
FAQ
How many onboarding videos do we need? Four or five short ones covering the gravity center plus the concepts a new user hits in week one — 10–15 minutes total. Past that, you're building a course, and the extra topics usually belong to docs or screen recordings anyway.
How long should each video be? About a minute of concept content; 60–90 seconds is the working band. If a script keeps wanting three minutes, it's two concepts wearing one title — split it, because reference viewers can only find atomic answers.
Should the founder appear in them? A short on-camera intro works on the series opener — 25–30 seconds, frame the product, hand off. Inside concept videos it costs more than it earns; reference viewers came for the concept, and the intro is the part they scrub past.
How do we keep them current as the UI changes? Route by decay rate. Concepts age slowly — animate those. Procedures age with every redesign — leave those to tooltips, docs, and screen recordings you can cheaply re-record. And keep a plan of record where marking a video "redo" is normal maintenance, not an admission of failure.
If you're deciding what your first onboarding videos should even be, send us your product's URL and judge twenty candidate directions against your real UI before spending anything.