20cuts

Resources · By use case · 6 min read

Onboarding videos: what users actually watch inside a product

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 as a sit-back series, because nobody watches onboarding as a series.

A new user opens video three at minute forty of a trial, stuck on one specific thing, with your product in the next tab. They arrived from a help panel or an empty state, they want one answer and their cursor back, and they have never seen videos one and two. Users consult onboarding the way they consult docs: mid-task, out of order, in a hurry.

Take an uptime monitoring tool: checks that ping your endpoints, an incident timeline that opens when they fail, and alert rules that decide who gets paged. That tool carries the examples below.

Build for the mid-task viewer

One concept per video, with named exclusions

Each video owns exactly one idea and defers its neighbors, annotated literally in the plan: this video owns what a check is, its neighbors own incidents and alert rules. When two videos half-teach the same concept, both fail — the viewer gets two shallow passes, no complete one, and every video runs longer than its title.

Before scripting each video, write its "deliberately not taught" list. Each cut gets a destination: another video, the docs, a screen recording. Scope becomes a list of named exclusions instead of whatever was left when time ran out.

Keep the depth at behavior, not internals. The alert-routing video shows that a failing check pages the on-call person within a minute; it does not explain the scheduler. A video explaining how the algorithm works should be reframed to what the feature does and when you would reach for it. Scene by scene, the same discipline is the one idea per scene rule.

Start from the gravity center

The common onboarding order is the tour order — whatever the navigation sidebar lists. The right order is dependency order, starting from the gravity center: the one object every other concept is defined in terms of.

For the monitoring tool that object is the check. An incident opens when a check fails, an alert rule decides who a check pages, and the status page publishes what the checks report. Teach checks first and every later video lands on a model the viewer already holds. Teach status pages first and you are explaining a display for data the viewer has never seen. Find yours by looking at which concept appears in the definitions of all the others, then:

  1. Make video one the gravity center, at orientation depth. What the object is, what it does, where it shows up. Nothing more.
  2. Group the rest into tracks by object, each with a stated arc — what the object is, what you do with it, why it composes with everything else. An arc keeps a track from being a feature list read aloud.
  3. Keep a plan of record marking each video exists, redo, or new. A stale onboarding video teaches a product that no longer exists, which is worse than no video.

Aim the opener at orientation, not mastery. The exit state we write down for intro sequences, verbatim: "I know what I'm looking at now. I can start exploring." An opener that aims for full coverage instead runs six minutes and leaves the viewer with none of it.

Route between video, tooltips, and recordings

Video is one tool with a specific edge: animate the concept, record the live product. For a good share of onboarding, the right move is no video at all.

Two placement rules:

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 label, value, and surface on screen comes from a real configuration and a real run (the grounding method). A new user who spots a screen they cannot find does not think "artistic license" — they think they are lost, and lost is the exact feeling onboarding exists to remove.

FAQ

How many onboarding videos do we need? Four or five short ones: the gravity center plus the concepts a new user hits in week one, 10 to 15 minutes total. Topics beyond those usually belong to docs or screen recordings.

How long should each video be? About a minute of concept content; 60 to 90 seconds is the working band (the length math). A script that keeps wanting three minutes is two concepts wearing one title — split it.

Should the founder appear in them? On the series opener, briefly: 25 to 30 seconds, frame the product, hand off. Inside concept videos the intro is the part stuck users 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 recordings you can cheaply re-record. And mark videos "redo" in the plan as normal maintenance.

Your own product settles these calls faster than any checklist. Send the URL and pick from twenty short videos of it.

See the answer for your product instead of the average:

Get my 20 free videos