CALLER-FIRST CONVERSATION STARTER Free planning resource for Burki. Checked October 10, 2026. Scope: caller-first entry with two intents and unknown clarification inside one opening Conversation stage. This is a worksheet and prompt starter, not an importable graph, saved workflow or completed voice test. Actual model, voice, platform and telephone usage can incur costs. 1. IDENTIFY THE INTENDED CONFIGURATION Business name: Public facts owner / approved date: Public hours / location / timezone: Approved staff contact process: Assistant name / identifier: Voice configuration / provider prerequisites checked: Workflow name / version or unsaved draft state: Assistant publication state: Telephone or browser route, if a test is authorized: Funding and account limits reviewed: Data retention and review permissions: Use an existing Guided assistant's Open entry to inspect Workflow. Do not rely on a Flows sidebar or guessed URL. Creating a workflow is a persisted action, not a preview. 2. CHECK THE FIRST-TURN CONTROLS Shared instructions -> Across the conversation: - Who speaks first? Caller - Begin delay (ms): 0 GPT Live is caller-first only in the checked configuration. Assistant is unavailable. Nonzero delayed flow start is not supported by the checked backend policy even though the delay input is visible. This starter does not establish a silence timeout or fallback greeting. Select the existing Start here stage: - Stage type: Conversation - Stage key: preserve the existing key unless there is a separate reason to change it. - Stage instructions: adapt section 4 below. - Skip response: off; unavailable for GPT Live. 3. SHARED INSTRUCTIONS TO ADAPT You are [BUSINESS]'s reception assistant. Use only the supplied public business facts. Distinguish public opening-hours questions from service enquiries for staff review. Ask a short clarification question when the caller's meaning is unclear. Do not invent availability, appointments, completed messages or private account information. A caller's requested visit is not an accepted appointment. Follow the approved staff contact process when the request is outside this scope. APPROVED PUBLIC FACTS Hours: Timezone: Location: Staff contact process: Known exclusions: Replace all placeholders and resolve contradictions with the content owner before saving. Keep provider keys and private customer information out. 4. OPENING CONVERSATION STAGE INSTRUCTIONS TO ADAPT Respond to the caller's opening message before asking a generic intake question. If they clearly ask when the public desk is open, answer with the approved hours and timezone. Do not infer staff availability from those hours. If they clearly want to discuss a service, ask which service they mean if needed, then explain the approved staff contact process. If their words could mean either an hours question or a service request, ask which they mean before answering. If neither intent fits, state the limit and offer the approved staff contact route. Accept corrections and use the caller's latest stated request. Never say an appointment is booked, a message was sent or staff accepted a request without a separate verified result. This prompt describes conversational choices in ONE stage. It does not create graph branches, a deterministic classifier, external actions, a structured business record or automatic contact with staff. Path editing is not required or established by this guide. 5. ILLUSTRATIVE AMBIGUOUS OPENING Caller: Can I come in on Friday? Expected: Clarify public hours versus a requested service visit. Caller: I mean when reception closes. Expected: Answer approved public hours with timezone, without claiming that a staff member is available. Caller: Actually, I want to discuss a service. Expected: Use the corrected intent and explain the approved staff process. Do not claim a visit was accepted or a message was sent. 6. EXPECTED VERSUS OBSERVED REVIEW Write expectations before an authorized test. No test is required to fill in the expected column. Do not fabricate observed results. Case A: Clear hours question Expected: Approved hours / correct timezone / no staff-availability promise. Observed: Evidence reference: Disposition / next owner: Case B: Clear service enquiry Expected: Relevant clarification / approved contact process / no fake action. Observed: Evidence reference: Disposition / next owner: Case C: Ambiguous Friday visit Expected: Clarification before an hours answer or request assumption. Observed: Evidence reference: Disposition / next owner: Case D: Caller corrects intent Expected: Latest stated intent replaces the earlier interpretation. Observed: Evidence reference: Disposition / next owner: Case E: Private order or unsupported question Expected: Honest information limit and approved staff route. Observed: Evidence reference: Disposition / next owner: Case F: Silence Expected: No behavior established by this starter. Review the actual configured session behavior separately; do not invent an idle policy. Observed duration / response / ending / cost evidence: Disposition / next owner: 7. STATE THE RESULT YOU ACTUALLY HAVE [ ] Prompt and facts reviewed only [ ] Draft saved (which workflow / date): [ ] Workflow published (which version / date): [ ] Voice configuration admitted (evidence): [ ] Authorized browser voice test completed (evidence): [ ] Authorized telephone test completed (evidence): [ ] Caller-heard behavior reviewed (evidence): Save draft, Publish and Test this version are different operations. Publishing is not proof of a passed call. Record the actual workflow, assistant configuration, route and observations used for acceptance. Article: https://burki.dev/blog/caller-first-voice-agent-workflow Commercial context: https://burki.dev/use-cases/ai-receptionist Current cost review: https://burki.dev/pricing