First-run teaching decays almost immediately

What it is

Even when people watch first-run guidance to the end, the sentences barely stick. Rapid forgetting means the lesson is encoded as a one-shot episode in session one, and is gone by the time the action is actually needed. This is not the same failure as skipping a whole tour—there, nobody looked. Here they looked, and the product still treats them as taught. First-run guidance is a single exposure: no spaced rehearsal, no retrieval practice on the target action.

Why it happens

Session one floods working memory with new labels, layout, and goals. Tour copy arrives as speech and is rarely used at once to produce a successful action, so encoding stops at “heard” rather than “done.” Retrieval fails on a scale of minutes to a day: people remember that an intro existed, not where export lives. A layer down, first-run guidance also lacks a retrieval cue—the sentence lived on a mask or welcome page, and later the UI has no matching place, so even recognition is hard. Spacing needs repeats at separated times; the first-run dumps every sentence into the same minute and spends the rehearsal budget at once. “Finished the tour” is finished playback, not finished learning. Treating playback as mastery means no later cue, and the failure shows up as “undiscovered features” when the memory has simply dropped.

Studying it

Have people finish a first-run explanation, then do the task immediately, after fifteen minutes, and the next day, with no repeat of the sentence.

Independent variables: delay between instruction and test, whether they had to perform the taught action while watching, whether the later UI still has a recognizable copy of the sentence. Dependent variables: free recall of the steps, recognition of the control, share who “remember seeing it” but cannot do it.

Immediate lab post-tests make teaching look successful; delayed tests carry the information. Do not use tour completion as a learning metric. Being able to say “there is an export” while clicking the wrong control is recognition failure, not success.

Where it stops holding

Actions that must be done on the spot and then end (consent, language, connecting an account) do not need long-term retention. High-frequency work (an inbox opened daily) gets recoded in use, so the first sentence can vanish and the UI still teaches. Infrequent, high-cost actions (first export, first share permission) drop hardest: learning and reuse are far apart. Experts do not rely on first-run sentences; measuring forgetting on them underestimates true decay.

Applying it

• Do not write “guidance completed” as “user has mastered it”; completion means playback, mastery means the action later succeeds unassisted.
• Move the key sentence off a one-shot mask and onto the place the action will happen (empty state, beside the button, first open of that panel) so it can be recognized later.
• After teaching, require a real result from that action, not a “next” that starts another sentence.
• Verify the next day: same people, no prompts, taught actions only. Steps they cannot do were never stored; make those sentences just-in-time or persistent, rather than lengthening the first-run tour.

Related

• Within the group: H2.01.1 Show first-use guidance at the moment of need, not as a dump up front · H2.01.2 Multi-screen intro tours are skipped as a set
• Adjacent: H2.03 Progressive onboarding · H2.05 Feature Discovery · E4.13 Tooltips
• Search terms: onboarding forgetting · one-shot instruction · retrieval cue