When demo GIFs and explainer videos go stale
A screenshot takes a minute to retake. A narrated demo takes a script, a recording session and an edit, so it is the last thing anyone updates. This page covers a 2024 lab study of what happens when a tutorial video no longer matches the app, and a rule for deciding which interface changes mean a re-record.
Why video is the last thing updated
Text can be fixed one sentence at a time. A screenshot can be swapped for a new file. A video cannot be patched that way. The narration is timed to the picture, and one moved button can mean recording a whole section again.
Cloudflare's documentation team says this directly in the public part of its style guide. Its video maintenance page says videos are "inherently harder to edit than text", and that small changes often mean re-recording, re-timing or redoing the visuals. Its page on when to use video lists "frequently changing UI" as a reason not to make one at all.
So the cost decides the outcome. When an update costs an afternoon, it waits for a quiet week, and the video stays wrong in the meantime. The same thing happens to screenshots, more slowly. The evidence for those is in what the research says about product screenshots.
The study
Raphaël Perraud, Aurélien Tabard and Sylvain Malacria wrote "Tutorial mismatches: investigating the frictions due to interface differences when following software video tutorials". It was published at ACM DIS 2024, the Designing Interactive Systems conference, held in Copenhagen in July 2024 (pages 1942–1955). It is peer-reviewed.
Its starting point is the gap this page is about. A learner's interface can differ from the one in the video, either because the software has a newer version or because someone customised it. The authors asked how those differences get in the way when a learner tries to copy what the video shows.
They did it in two steps.
- A catalogue of differences. Through what they call a morphological analysis, they identified 13 types of "interface differences". The types vary along three dimensions: whether a feature is available, how it is reached, and where it sits on screen.
- A lab study. 26 participants each followed 13 short video tutorials in Affinity Designer 2, a vector graphics app. There was one tutorial per type of difference and a 5-minute limit per task. In most tasks the interface on their screen differed from the one in the video. Two tasks were identical controls.
The authors report that participants compared their own screen with the one in the video, and that they used several different approaches to find features visually.
When the feature they needed was still there somewhere, participants finished 89.8% of those 11 tasks on average. Three kinds of difference hurt most: needing to rearrange the workspace first (61.5% finished), and a feature that had moved into a floating window (80.8% and 73%). When the feature was gone or disabled, only 25% finished, because the video showed no other way to do it.
What the study does not say
We read the authors' copy of the paper on HAL, and checked its title, authors and venue against ACM's DOI record. One thing you might want is not in it:
- It studied 13 short tasks, one per type of difference, in one vector graphics app, with 26 people in a lab. That is a careful study of how people cope with a mismatch. It is not a measure of how often tutorials in the wild are out of date.
What it does establish is useful on its own. A mismatch between video and app is a recognised problem, it comes in distinct kinds, and viewers deal with it by comparing two screens. Our reading is that the comparing is work the video was supposed to save them. For still images the trade-offs are different. See screenshots in instructions.
Which changes force a re-record
The study's three dimensions make a workable test. Ask them of each step the video shows, not of the app as a whole. A change in a part of the screen the video never uses costs the viewer nothing.
| Question for each step in the video | If the answer changed | Example |
|---|---|---|
| Is the feature still there? | Re-record that step. The viewer cannot do it. | An option was removed, or now needs a different plan |
| Is it reached the same way? | Re-record that step. Following the video now fails. | A toolbar button moved into a menu; a step was added before it |
| Is it in the same place? | Re-record if the video points at it or the narration says where it is | A panel moved from the left side to the right |
| Does the narration name it? | Re-record the audio, or fix the caption, if the label changed | "Click Publish" when the button now says "Deploy" |
| Did only the look change? | Usually leave it, and note the version in the description | New colours, a new icon set, a font change |
Re-record when a viewer who copies the video exactly would fail or get lost. Leave it when they would only notice that it looks different.
Marketing demos need a stricter version of the last row. Visitors don't follow a homepage GIF step by step. They compare it with the product they sign up for. A demo in last year's colours tells them the page is old, even if every click still works. The same applies to stale homepage and deck screenshots.
Making the next update cheaper
If a re-record is the expensive part, the fix is to make re-recording cost less. Two published sources point the same way, and both describe practice, not measured results. The third step below is our own.
- Short segments. TechSmith, which makes Camtasia, suggests recording training in chunked segments, so a changed process means replacing one segment rather than the whole module. It also suggests simplified drawings of the interface instead of real screenshots. A simplified drawing has fewer details to go out of date, at the cost of looking less like the real app.
- Periodic checks. Cloudflare checks videos against its docs and dashboard from time to time, and relies on product managers and subject experts to flag UI changes. It says it is building an internal tool for code-driven screen demos that would update footage when the product changes. Cloudflare describes that tool as in development, not shipped.
- Render from a script. If the demo is produced by a script that drives the app, a re-record is a re-run. That is the same move that lets a rebrand without retaking a screenshot work for still images.
How Screenery handles clips
Screenery does not record or render video. screenery demo publishes a demo you already rendered, a GIF and a video under one name, and prints the embed code. See Demo clips for the folder layout and flags.
- Stable URLs. The GIF and the MP4 each get an @latest URL. In a project bound to your repository through Screenery's GitHub App and not set to manual promotion, a job on your default branch that runs screenery demo changes the bytes behind those URLs. A pull-request build goes to @pr-{n} and leaves @latest alone. A README or page that embeds them shows each new render without anyone editing it.
- GIF for READMEs, MP4 for pages. GitHub plays a GIF in a README. On your own pages, use the MP4 in a <video> tag, with the poster image.
- Screenshots in the same project. On CLI 0.4.0 or later, a demo push adds the demo and removes none of your screenshots. The next full screenshot push does remove the demo from the channels it moves, so put demos in a project of their own or set the project's channel mode to additive on the console's Access screen.
One limit matters more than the rest. A URL only helps where the URL is embedded. A copy uploaded to YouTube, attached to a social post or pasted into a slide deck is a separate file. It does not follow @latest, and it goes stale like any other upload. Keep a list of those copies, and run the table above on them when the app changes.
Sources
We opened every page below on 15 November 2026 and checked each claim on this page against it.
- Perraud, Tabard and Malacria, "Tutorial mismatches: investigating the frictions due to interface differences when following software video tutorials", DIS '24, pages 1942–1955, July 2024 (authors' copy: HAL)
- Cloudflare docs style guide, "Maintenance" (how we video)
- Cloudflare docs style guide, "Why and when we use videos"
- TechSmith, "The Case for Designing Training Content for Reuse", November 2026 (vendor)