The opening should hint what this session can do

What it is

The floor is caught; the user has not yet handed over an intent. “I’m here” only reports availability. “Music, weather, or a timer?” offers a range of activity for this session. A capability-hinting opening treats the system’s first utterance (or first on-screen chips) as an activity offer — not a manual, and not the wake itself. Voice UIs have no menu that stays in view. The opening is one of the few moments that can broadcast a thin slice of capability before anyone has committed to an intent.

Why it happens

After wake, the first command has to be pulled from memory. The opening line can shrink that search to a few things that actually pay in the current situation: navigation and messages in the car; weather and news on a morning speaker. Too generic (“how can I help”) dumps the whole search back on the user. Too long (a list of ten skills) occupies the floor; experts barge, novices forget the list they just heard. The grain is “what this session can roughly do,” not “everything the product can do” — that larger problem is long-horizon discoverability.

The hint also sets expectation. An opening that named music and weather, followed by a leap into whole-home lighting, feels like the wrong room. Two or three situated handles beat a universal help-desk greeting at keeping the first intent inside coverage. When a screen exists, chips can carry the broadcast; the mouth does not have to read a menu.

Studying it

Compare opening variants: silent earcon, generic “I can help,” two situated skills, an eight-item list. Dependent measures: first-turn out-of-coverage, “what can you do” immediately after open, and first-turn success — new and returning users split. Do not measure the wake word here. Under barge-in, also count how many people speak before the hint finishes; that group is treating the opening as noise.

Slice field traffic by situation. If the same generic opening yields very different “what can you do” rates in the kitchen and in the car, the opening is not eating context.

Where it stops holding

Experts who fire the same short command every day experience any spoken hint as a speed bump; barge-in must cancel the opening. When hands and eyes are both busy, people should not have to hear a menu before they can act. A single-purpose timer barely needs a hint; a TV that is also a household butler needs a stronger range claim. An opening that overstates capability makes later refusals sting more, because the expectation was raised by the opening itself.

Applying it

• Default opening: a short availability signal plus at most two handles tied to the current situation. If a screen exists, put chips there and speak even less.
• The opening must be barge-able. After barge-in, do not treat the unheard menu as user intent.
• Watch two numbers: first-turn out-of-coverage, and “what can you do” right after open. When both rise, help copy is not the first leak — the opening failed to hint what is doable now.
• Write a different opening per situation. Ban one product-wide “how can I help.” Sample morning-peak and in-car sessions to see whether first intents land in the class the opening named.

Related

• Same group: M1.12.1 Wake is an explicit handoff of attention · M1.12.3 Session end needs a perceptible close · M1.12.4 False-wake exit cost decides whether people dare to use it
• Nearby: M2.06 Discoverability and Help · M2.01 Information Density of Prompts · C7.01 Wake Words
• Search terms: capability-hinting opening · session-initial prompt · activity offer