AI Assistant: Coaching Users on How to Ask
Last updated: May 4, 2026
Even with a well-structured library, the way a user phrases a question shapes the answer they get. Share the following principles with your end users so they get the most out of the Sidekick.
1. Question Framing vs. Topic Framing
How a prompt is phrased signals to the AI what kind of answer the user wants. A direct question invites personal guidance. A topic phrase invites a general overview.
First-person questions invite coaching: Phrasing such as “How do I handle a customer complaint?” signals that the user wants actionable, personalized guidance — the response tends to read like instruction or advice.
Topic phrases invite explanation: Phrasing such as “handling customer complaints” signals the user wants a broader overview of the subject — the response tends to read more encyclopedically.
Choose framing that matches intent: If the user is about to perform a task, a first-person question will produce more useful output. If they are researching a concept, a topic phrase is often better.
2. Point of View Matters
The word “I” is a strong signal that the user is the one performing the action.
Use “I” for personal guidance: Including “I” tends to produce second-person, actionable responses.
Drop “I” for descriptive answers: Removing the first person shifts the response toward a more neutral, descriptive tone.
Match pronoun to need: Encourage users to consciously choose between “How do I…” and “How to…” based on whether they want coaching or a summary.
3. Add Context the AI Cannot Infer
The Sidekick does not automatically know which store, brand, or role a user belongs to. Including that context in the prompt sharpens the answer significantly.
State the role: Mention whether the user is an associate, keyholder, manager, or trainer when it matters.
State the setting: Include brand, banner, region, or store format when rules vary across them.
State the moment: Mention whether the question is about onboarding, a live customer interaction, an end-of-day task, etc.
4. Ask One Question at a Time
Compound questions produce diluted or partial answers. Tightly scoped questions produce cleaner, more accurate responses.
Split compound prompts: Encourage users to ask follow-up questions rather than stacking multiple topics into one prompt.
Be specific about what is needed: A list, a script, a checklist, and a policy summary are all different outputs — asking for the specific format yields the specific format.
5. Iterate and Refine
The first answer is a starting point, not always the final answer. Treat prompting as a short conversation.
Ask for more detail or a rewrite: If an answer is too general, ask for specifics; if it is too long, ask for a summary.
Provide correction: If the Sidekick misses the mark, users can tell it what was wrong and ask again — it will adjust.
Flag gaps back to the admin team: If a question consistently fails, that is a signal the underlying content may need to be added or improved.