> For the complete documentation index, see [llms.txt](https://docs.slatecx.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.slatecx.com/concepts/prompt-suggestions.md).

# Prompt suggestions

Prompt suggestions are short prospect-facing prompts. Three placements exist, and they are produced differently.

## Placements

| Placement           | Where it appears                  | How the text is produced                                         |
| ------------------- | --------------------------------- | ---------------------------------------------------------------- |
| Starter chips       | On arrival in a Slate             | Entry-point selected Prompt Pill Library text, rendered directly |
| In-chat suggestions | After an assistant answer         | Generated                                                        |
| Wave prompts        | Before the embed widget is opened | Generated                                                        |

Deterministic UI action pills, such as Save Slate, are owned by the chat runtime and are not prompt candidates.

## Prompt Pill Library

The campaign Prompt Pill Library holds reusable, tenant-authored prompt text. It serves two roles: entry points select up to five entries as starter chips, and the library also acts as guidance for generated prompts.

## In-chat suggestions

Generated after an eligible assistant response is persisted. The runtime suppresses generation when the assistant message contains a question, a direct request for input, or an uncertain ending — the prospect already has something to answer.

An eligible turn first reuses a fresh pool for the same assistant message. Failure or zero valid candidates produces no suggestions.

## Wave prompts

Wave readiness builds one sanitized context snapshot, then reads candidates before resolving any provider runtime.

| State                     | Behavior                                                        |
| ------------------------- | --------------------------------------------------------------- |
| Fresh pool exists         | Reuse the whole winning pool in rank order                      |
| Stale winner exists       | Serve it, and schedule one lazy refresh                         |
| Nothing valid ever stored | Return `pending` and schedule generation behind a durable lease |

Generation asks for up to 10 candidates. Reads never expire, invalidate, or bulk-regenerate candidates.

## What generated prompts can be grounded in

Generated prompts are validated to be short, prospect-authored, and grounded only in signals actually present in the final provider request.

Wave prompt context deliberately excludes campaign name and description, entry-point name and greeting, goal and question progress, and campaign cards.

## Telemetry

Prompt lifecycle events are recorded in `prospect_prompt_event`, including reporting-only traffic classification for embedded wave requests. Wave candidates persist a report-safe host page URL when one is available.

Wave performance is reported under **Insights**. See [Review insights](/guides/review-insights.md).

## Privacy

Raw prompts, raw context, and raw provider responses are never exposed through Wallet, Embed, or Admin reporting.

## See also

* [Embed widget](broken://pages/cjAFA8i7EzjwWLpzmaaX)
* [Entry points](/concepts/entry-points.md)