“What else can you do” is a bad question to answer

What it is

“What else can you do?” sounds like a reasonable meta-question. Speech can almost never answer it well. A municipal 311 that reads down the skill table — streetlights, noise, rubbish, permits, appointments — hands the ear a catalogue it cannot hold, and the caller was not actually shopping. The sentence usually arrives when someone is already stuck: they do not know what to say next, so they kick the difficulty upstairs. Answering with a capability list is reading the unscannable menu out loud.

Why it happens

The surface act is “enumerate your coverage.” The real act is often “give me a next step I can use now.” Enumeration in audition hits the serial-list wall: too many items to remember, too few items feels like a brush-off, and order is warped by position effects. A system that answers the coverage table honestly satisfies neither the need for a next step nor the ear’s budget.

Treating the sentence as an ordinary slot to fill is a type error. A good reply does not make a fuller table. It turns the question over: ask what the person is trying to do, or offer only the category next to the current situation, plus one speakable example. The coverage table can live on a screen or a site. Answering “what else can you do” by mouth is agreeing to do a job the channel does badly. The failure is not unpolished copy. The question picked the wrong channel.

Studying it

Collect live “what can you do / what else” requests and hand-code the act: exploration, help after being stuck, or chat. Then two system replies: catalogue dump (read N skills) versus turn-over (“which kind of thing are you trying to do,” or one category plus one example). Dependent measures: whether an executable intent follows, immediate recall of dumped items, barge-in and abandonment. Independent variables: how many items were dumped, whether turn-over carries the current situation.

Yankelovich’s question becomes: after hearing the catalogue, do people actually know what to say next. If a recall test only recovers the first and last items, the dump has already failed. Do not take “they asked, so we must answer at face value” as a design premise — high frequency of the question means many people are stuck, not that a catalogue is the right reply.

Where it stops holding

When there are only two or three capabilities, all bound to the present scene (lights, blinds, temperature in this room), a short face-value answer works. When the user is explicitly comparing products or evaluating a purchase, the catalogue is their goal and should move to a screen or a person. When the law requires a disclosure of range, the disclosure can take another channel; it does not have to be crammed into the spoken answer to this sentence. Turning the question over into “I can do anything, just talk” refuses a boundary; it is not avoiding a dump.

Applying it

• Do not write a spoken master list for “what can you do.” Default to turn-over: one category, or a question about the current goal, plus one speakable example.
• If listing is unavoidable, three categories at most, one example each, then stop and ask “is it one of those.” If three still cannot land, hand off to a screen.
• When the sentence arrives after a miss, help from the scene of the miss, not in exploration mode.
• How to check: code the following turn as an executable intent or another meta-question. A dump that yields another meta-question or silence is spinning. Switch to turn-over and compare the share of executable intents.

Related

• Same group: M2.06.1 A voice interface has no scannable feature list · M2.06.2 Introduce capabilities at the right moment · M2.06.3 A screen is the most effective supplement · M2.06.4 People ask for help after failure · M2.06.5 Capability intros need speakable examples
• Nearby: M3.03 Reading long lists aloud · M1.02 Memory load of screenless interaction · M2.02 Open and closed questions
• Search terms: what-else-can-you-do is a bad question · catalogue dump · meta-request in speech