An empty box conveys no boundary; the first sentence is a guess

What it is

The caret blinks in an empty field: no examples, no prior thread, no greyed-out menu items. The first sentence is not a pick from known operations. It is a live sample of “what does this even accept.” First-utterance guesswork means the blank entry turns the first submit into a hypothesis test — a decent answer confirms the guess; a refusal, a tangent, or a fluent wrong answer revises it, and the next sentence is another guess.

Capability range can be patched in later by other means. This entry is that one cold-start beat. An empty box is not a neutral wait. It forces a bet.

Why it happens

A channel with no boundary triggers information foraging: people probe to build a mental model. Web search at least returns last-round results as an echo of the boundary. A blank generate box has no echo, and the probe costs a full generation. A first success is taken as a map; a first failure as “you cannot ask that” — sample size one, and the map is already inked.

Dialogue adds a politeness pressure: the first turn should not be long or sound like an order, so people hand over the shortest, vaguest sentence. The shortest sentence carries the fewest constraints; the model fills the most from its prior; the first result drifts the farthest. The probe is sabotaged by its own brevity.

Studying it

Look only at turn zero: from empty state to first submit. Record drafts, deletions, pauses, and how specific the submitted sentence is (object, format, success criterion). Independent variables: empty box, one domain placeholder, three clickable examples (the example content is not the object of study, only the condition “was blankness broken”). Dependent variables: length and slot completeness of the first sentence, first-round task success, whether the second sentence revises the hypothesis rather than advancing the task.

Operationalize “the first sentence is a probe” as think-aloud tokens like “let’s try” and “not sure this will work.” If you first drop in a filled draft and ask people to edit, the probe should vanish — that is a control, not a way to count template effects here.

Where it stops holding

Returning users open a thread that already has history; the first sentence is not a cold start. Entries beside a document selection, a reply-to-mail box, or a spreadsheet cell carry an object, and the guess space is cut by that object. A forced registration quiz or mode picker before the field moves the real first utterance earlier. This entry does not argue whether examples are good materials for showing capability, nor whether help docs can salvage discoverability.

Applying it

• Do not hand a true empty state to a generate entry. At least start with an object: “about this article,” “about the three selected rows,” so the first sentence continues something rather than creating the world.
• If a cold start is unavoidable, make round one a cheap probe: short choices, reversible, with the result well marked as “this is the assumption taken from your sentence.”
• After a first-round miss, ask which class of thing they wanted. Do not send them back to guess another sentence with a generic refusal.
• Check: new account, wiped state, time to first submit, score the first sentence for constraint completeness. If most first sentences are an intent verb with no object and no success criterion, and the product treats that sentence as a real task, the empty box is still forcing a guess.

Related

• Same group: L2.01.1 Open input does not cue the range of capability · L2.01.2 Not knowing how to say it is the main barrier · L2.01.3 Differences in wording produce differences in results · L2.01.5 Open input steers failure attribution toward “I said it badly” · L2.01.6 Synonymous phrasings yield different results, so users invent phrases to memorize · L2.01.7 Openness makes functions non-enumerable; the product can no longer show a complete capability list · L2.01.8 Error messages for open input stay vague because the system does not know what the user meant to do
• Nearby: L2.02 Discoverability of What Can Be Said · L2.14 Carrying and Clearing Context · L6.03 Cold Start
• Search terms: first-utterance guesswork · cold-start prompting · empty-state probe