Novices and experts need different prompt lengths

What it is

The same prompt is scaffolding for a first-time speaker and waiting for someone who already knows the line. A warehouse pick-by-voice headset on day one needs “say aisle, then bin, then quantity.” The same person in week two needs a short tone; the triple is already in the mouth. Play novice length to an expert and the extra words become latency and irritation. Play expert length to a novice and they do not know what the next utterance is supposed to look like. Length is not a copy preference. It is a function of grammar uncertainty.

Why it happens

What novices lack is a speakable template. They have not yet chunked “aisle–bin–quantity”; an example in the prompt externalises that template. After practice, uncertainty collapses and the same example turns from information into dead air — the person is already waiting for endpoint detection so they can ship their line. Prompt length should follow that uncertainty curve: shorten after consecutive successes inside a session; step down across sessions by cumulative successes.

This meets “how many options can be held at once” at one point only: once enumeration exceeds what a listener can hold, even a short prompt cannot carry it. Beyond that, what experts want shortened is not fewer options but the assumption that options are known, leaving only a receipt signal. Reading the full menu to an expert is not teaching. It is taxing the expert at the novice's uncertainty rate.

Studying it

Expertise × length factorial: recruit first-shift users and users with many completed shifts; cross short / medium / long prompts. Dependent measures: latency to speech, task success, rated annoyance, and whether the user's utterance already matches the target grammar. A longitudinal cut tests the same people in week one and week four, one arm on a fixed prompt, one arm tapering with success count.

Do not define expertise as “finished the lab questionnaire.” Cleaner field proxies: historical successes on that skill, gap since last use (a long gap should roll length back), and whether the user has already started barging in. If the expert group's barge-in rate on long prompts sits well above the novice group's, length is no longer allocated by listener.

Where it stops holding

Rare tasks (annual tax filing, a single onboarding) produce almost no experts; keep novice length as the resident form, even if shortening logic exists in the system. Long gaps decay the chunk; yesterday's expert needs a step back today. Shared devices (a corridor intercom on a ward) cannot be tiered per person. The compromise is novice-adequate and interruptible for experts, with the constraint still in a position that barge-in might leave intact. Shrinking length to zero (a tone only) makes a newcomer think the device is broken.

Applying it

• Give each skill two or three prompt tiers: first use with one short speakable example, then action keywords after stable success, then a tone or a minimal receipt.
• Tier on consecutive successes for that account on that skill, not on “days since signup.” One failure, or a gap past a written threshold, steps back a tier.
• Shared devices are not personalised: default to novice length, allow barge-in, keep constraints where they can still be heard.
• How to check: sample first-use and high-frequency sessions separately. If first-use grammar is scattered and follow-up questions pile up, add an example without adding length. If frequent users barge in before the prompt ends and still succeed, the prompt should have shortened and has not.

Related

• Same group: M2.01.1 A prompt must say what you can do in one sentence · M2.01.2 Overlong prompts get cut off or forgotten
• Nearby: M1.12 Opening and closing a dialogue · M2.06 Discoverability and help · M1.02 Memory load of screenless interaction
• Search terms: expertise-adaptive prompt length · tapered prompts · earcon as prompt