A suggested prompt can help someone understand an AI assistant. Its wording also shapes what that person expects to happen when they click it. A team introducing suggested choices should be clear about whether a choice starts a conversation or invokes a command.
In a November 18 announcement, Google described starter prompts for Chat apps and agents. A text prompt sends a message; a command prompt runs an existing slash or quick command. Developers configure the choices. Google lists rollout starting November 16 over one to three days, while support for agent sessions is separately described as developer preview. These are vendor descriptions, not evidence that a suggested choice bypasses permissions or performs any particular business action.
Describe the result of choosing an option
The following is our original design and operating analysis, illustrated with a hypothetical internal assistant. It does not describe a tested Google Chat app or the behaviour of a particular customer’s system.
Imagine a team offering an assistant to help organise office supplies. Its first screen includes a choice labelled “Arrange supplies.” One employee expects to discuss what is needed. Another expects the choice to start a prepared request. The label leaves both interpretations open.
Before choosing the wording, decide what the option actually does. If it opens a conversation about a request, say that. If it invokes an existing command, explain the immediate result in terms the employee can understand. The description should match the configured behaviour.
A broad phrase can be attractive because it sounds simple. It becomes less helpful when it hides the distinction between asking a question and starting a defined operation.
Show where the user enters the process
In our invented assistant, suppose the team has separate stages for collecting a request, reviewing it and passing it to the person responsible for supplies. A starter choice should make clear which stage it begins.
Starting a request would not necessarily mean that the request has been approved or completed. The team’s explanation should preserve those distinctions. Otherwise, an employee could interpret a successful first step as evidence that the whole job is underway.
The same question applies to the reply after a choice is used. A useful response would tell the person what happened and what remains to be done. It should avoid using a general success message for several different outcomes.
Try the wording with someone new to the tool
Ask a colleague who did not build the hypothetical assistant what they expect each choice to do. Have them describe the expected result before they use it, then compare that account with what the configured workflow actually produces.
The purpose is to find a misunderstanding the builder may have stopped noticing. A familiar term inside one team can mean something different to an employee arriving from another department. Discovering that difference early gives the team a concrete wording or workflow question to resolve.
This kind of review would be evidence about the examples examined, not proof that every future user will understand the interface. Keep the findings specific: which choice was unclear, what the person expected and what the team changed.
Keep the entry point aligned with later changes
If the operation behind a choice changes, review its label and description too. In the hypothetical supplies workflow, moving a review stage could change what “start a request” means to the person using the assistant.
Give someone responsibility for that alignment. The person editing a workflow may need a colleague who understands its users to check whether the introductory choices still tell the right story. Treat that as part of maintaining the tool’s meaning, rather than a separate cosmetic task.
A good starter choice helps people enter the right process with a reasonable expectation of what happens next. As AI assistants become easier to discover, clear descriptions of those first actions can make the surrounding business workflow easier to understand as well.
AI-generated conceptual photograph of a fictional workplace choice; not an actual Google Chat interface or customer deployment.