Skip to content

Write natural conversation repair prompts when intent remains unclear

Replace robotic 'I didn't understand that' dead ends with smart, narrow clarifying choices that keep misunderstood turns moving forward.

Copy the prompt and paste it into the AI assistant you already use.

Four rules for frictionless conversation repair

  1. 01

    Ban 'I didn't understand that' forever

    Never blame the user or repeat identical static error text; state what you caught and ask specifically about the missing piece.

  2. 02

    Offer two or three concrete hypotheses

    When someone types 'SSO issue,' ask 'Are you setting up SAML for the first time, or is an existing team member locked out?' with tap chips.

  3. 03

    Use current page context to break ties

    Factor in whether the user opened chat from `/billing/invoices` versus `/docs/webhooks` so disambiguation guesses smartly.

  4. 04

    Escalate on strike two, never strike four

    If clarification fails twice in a row, stop guessing and hand the conversation directly to a live specialist or priority ticket form.

Questions and answers

Why do users type single nouns like 'billing' or 'refund' into AI chatbots?

Years of legacy keyword IVRs and rigid button bots conditioned users to type terse caveman keywords expecting a menu first.

How should a text-only chatbot react when a user uploads an image or PDF?

If vision ingestion isn't enabled on that channel, explicitly note you cannot view attachments yet and ask for the error text or escalate to an agent who can.

Should clarifying questions let users type freely in addition to clicking chips?

Always keep the text composer unlocked alongside quick-reply chips in case none of your three guessed options match their actual edge case.

Learn more

Related guides and background articles for this task