Too many follow-ups turn the exchange into a questionnaire and cancel the efficiency natural language was for

What it is

The user says “help me reply to this mail.” The system returns: “1. Accept or decline? 2. Formal or casual? 3. Mention a next time? 4. Chinese or English?” After four questions, opening the mail and hitting Reply is cheaper. Natural language is efficient because intent can be said once and obvious slots left unsaid. Asking those slots back turns the exchange into a questionnaire in dialogue clothing. The advantage cancels itself.

The questionnaire feel does not depend on how polite the questions are. It depends on whether the person must finish the system’s list before seeing any result.

Why it happens

Dialogue systems often gate action on slot completeness: missing one slot, ask one question. The slot table is the designer’s task model, not the user’s decision at this moment. A single sentence already declared how much information they were willing to give; asking every remaining slot forces them to work at the system’s completeness. Each ask also has a switch cost: from composing intent to answering a multiple-choice. More asks, more shattered switches.

Over-asking also changes attribution. People start to feel “this thing cannot read speech, it only does forms,” and they answer with the shortest option numbers, no longer offering rich constraints. Slots look fuller in the data; the language channel has been closed by the user.

Studying it

Run the same intent on three surfaces: a result in one shot, at most one question, the full slot table. Dependent variables: time to first usable result, voluntary abandonment, whether later turns still use whole sentences or switch to “pick 2.” Plot ask count against questionnaire-feel ratings and find the ask at which feel jumps — that is this product’s over-ask point, not the theoretical slot count.

A lab that forbids “I’ll just reply to the mail myself” will under-count abandonment. Escape to an external tool must be allowed; escape is itself a primary endpoint.

Where it stops holding

Mandatory asks required by regulation or safety (is this going outbound, does it involve personal data) are not questionnaire items; ask them once, separately, and say why this turn cannot skip them. In specialist software people expect to be asked everything (tax, prescribing); the questionnaire is the work. On voice every extra turn is expensive, so over-asking hurts more than on glass; on glass, collecting several slots in one short form is sometimes less like a questionnaire than a chain of spoken asks.

Applying it

• Default to at most one ask, on the dimension that actually forks, and fill the rest with assumptions and start work. Ban a “finish the form then generate” flow on an entry that sells natural language.
• If several slots must be collected, use one short form with skippable fields, not four dialogue turns. Skipped fields get assumptions and can be changed after.
• When two consecutive turns are answered with option numbers, treat the language channel as closed: stop asking new slots and put down an editable draft instead.
• Check: on “reply to this mail,” count system questions before the first draft. More than one, and no draft yet, is a questionnaire incident. Then look at whether answers are whole sentences or numbers — a rising share of numbers means people are already using the product as a form, and the natural-language advantage is gone in behaviour.

Related

• Same group: L2.15.1 Assuming one reading and running with it is faster, and it also enlarges the cost of being wrong · L2.15.2 A follow-up is worth asking only when the ambiguity would lead to clearly different outcomes · L2.15.4 Leading with a result and stating the assumption is a third path between asking and running · L2.15.5 Weigh the cost of clarifying against the cost of undo; if undo is cheap, do not ask
• Nearby: L2.03 Examples and Template Guidance · L2.01 Openness of Natural-Language Commands and Its Cost · L4.07 Pre-action Confirmation
• Search terms: over-clarification as questionnaire · slot-filling in dialogue · clarification fatigue