Suggestions that appear only after failure arrive past the point of giving up

What it is

Conversational products often put “you could try…” after an error, an empty result, or a model apology. The suggestion is logically right and temporally late. Post-failure suggestion shows up once the person has already judged “I cannot use this” and is preparing to leave. The abandonment point precedes the suggestion point. A good chip on a tab that is already closing was never given.

This is not bad copy, and not a debate about whether examples should exist. It is the order of suggestion relative to giving up. Right content, late moment, still a miss.

Why it happens

The first failure of an open field is read as self-attribution: “I did not say it clearly.” Self-attribution burns retry willingness fast, because the next attempt still has no new wording to try. In search and form-filling, people leave after consecutive failures; they do not stay for the moment the UI finally offers help. Help that is bound to the failure state appears under exactly the condition in which people are ready to go.

Failure states also steal attention. Errors, red, apology copy get read first. Suggestion chips, as a second visual object, often are not scanned at all. Even if the page stays open, the suggestion loses to the failure signal.

Studying it

Build an abandonment curve: origin at first submit, proportion still in the session against time or against failure count. Mark the moment the first suggestion appears on the same curve. If the median suggestion time sits after the steep drop, timing is past the abandonment point. Independent variables: suggestion trigger (first screen / while typing / after first failure / after second failure), whether the suggestion interrupts the failure copy. Dependent variables: suggestion exposure (was the person still there when it rendered), click-through, continuation of the session after the suggestion.

A lab that requires people to finish the task delays abandonment and will score late suggestions as effective. Use a leaveable task, or operationalise abandonment in the product as close, navigate away, or a long idle.

Where it stops holding

Irreversible, high-cost failures (an outbound send, a payment) should get recovery suggestions after they fail; that is repair, not discoverability. This entry is about early failures while “what can I say” is still unformed. Experts stay and read the error; late suggestions help them. Advancing suggestions to the first screen is not success by itself — if they are unrelated to the current draft, people treat them as unseen, and the problem becomes relevance rather than order.

Applying it

• Bind “how you can say it” to the act of input, not to the failure state: show phrasings related to the current draft on empty input, on pause, before submit.
• On first failure, do not only apologise. If a suggestion must follow failure, make it the primary visual, shrink the failure line to one sentence, and keep the suggestion in the same viewport with no scroll.
• Detect rapid consecutive failures (two submits close together). On the second, do not swap in a longer explanation; offer one concrete next step that can be clicked, or replace the field with a structured fill-in.
• Check: align in logs “time first suggestion rendered” and “time session ended.” If median end is earlier than median render, the suggestion is playing to an empty room. Also check whether focus had already left the field when it rendered — focus gone means the suggestion did not appear.

Related

• Same group: L2.09.1 Natural-language interfaces have nothing to scan, so discoverability is worse than in graphical ones · L2.09.2 Examples draw the range of what is possible, so diversity matters more than count · L2.09.3 People treat seen examples as a ceiling; a narrow set shrinks real use · L2.09.5 Capability hints must move with the dialogue; a one-shot start-screen hint does not cover later turns
• Nearby: L1.06 Graceful Degradation of AI Failure · L2.01 Openness of Natural-Language Commands and Its Cost · L2.03 Examples and Template Guidance
• Search terms: post-failure suggestion timing · abandonment point · just-too-late help