AI receptionist business brief: what to write down before its first call

Fix what the receptionist will learn from on its own

Your brief starts with the sources a vendor will read without asking, because most of them train on data you already published. Clean those first, or the receptionist will repeat your old promo price to a caller at 9 p.m.

Rosie says it scans your website and Google Business Profile to train, then lets you confirm the data and add FAQs (Rosie). Jobber says its Receptionist is ready on day one from your company profile, services, request forms and booking settings (Jobber). HighLevel Voice AI follows the selected calendar’s availability, buffers, minimum notice and conflict settings (HighLevel Help). Sameday takes special instructions for your processes (Sameday).

Key takeaway: Before any demo, update your website, Google profile and booking settings. For most of these products, those are the brief.

Decide what it may do before it says a word

Set the receptionist’s job in one sentence before you configure anything. For example: “Answer questions, book estimate visits for approved job types, and hand everything else to a person.”

The products give you these controls in different places. Jobber says you control what Receptionist can and can’t do, separately for calls and texts, and by time of day (Jobber). Rosie’s plans differ in how many custom message-taking scenarios you get: 2 on Professional, 5 on Scale, unlimited on Growth (Rosie pricing). Goodcall’s default “operator” workflow transfers to your business line, and you can change it to transfer to a person, take a message, send a link or schedule a callback (Goodcall).

Your business brief worksheet

1. Business name, how to pronounce it, and the callback number callers should hear.
2. Hours by day, time zone, and when the receptionist should answer (after hours, overflow or all calls).
3. Service territory as a ZIP or town list, plus the answer for callers outside it.
4. Services you accept, and the ones you decline or refer out.
5. Job types it may book, their duration, and whether a booking is confirmed or only requested.
6. Fixed fees it may quote word for word, if any, and nothing else.
7. Emergency keywords and the escalation chain: first person, backup, and what happens if nobody answers.
8. How to handle existing customers: what it may confirm and what it must not disclose.
9. Who gets messages and work requests, and within what time they will call back.
10. The owner who approved this version, and the date.

Keep customer records out of the brief. Use fictional contacts when you test, as the demo checklist does.

Area	What to supply	Boundary to write down
Services	Work you accept and decline	No invented services
Prices	Approved fee wording only	No repair estimate without a visit
Booking	Job types, durations, booking source	Requested vs confirmed
Existing customers	How to identify them	No account details over the phone
Escalation	Person, backup, fallback	No promise nobody can keep
Follow-up	Queue owner and hours	No callback promise outside staffed hours

Script the five calls that go wrong

Write the approved answer for the calls that cause complaints, because a confident wrong answer does more harm than “let me get someone”. Read each answer aloud with the office. If a new hire wouldn’t be allowed to say it, the AI shouldn’t either.

Caller says	Approved pattern	Never say
“Can you come right now?”	Take details, escalate if it’s an emergency keyword, otherwise offer the next bookable slot	A promised arrival time
“What will this cost?”	Quote the fixed fee if approved, then book an estimate	A repair price
“I already talked to someone”	Take the name and reason; route to the job owner	Details from another customer’s job
“Are you licensed / insured?”	Use the exact approved sentence	Anything not in the brief
“I’m outside your area”	Your written referral or decline	“We’ll make an exception”

Example: An illustrative plumbing shop writes: “Our diagnostic visit is $89. The technician gives you the repair price on site before any work starts.” Nothing else about price goes in the brief.

Watch out: Jobber documents keyword escalation for words like “burst pipe” or “no heat”, which transfers the call or texts you (Jobber). If your escalation contact is your business number, it won’t work. Jobber’s help center says the escalation number must differ from your business or dedicated number (Jobber Help).

Keep the brief current and carry it into the demo

Treat the brief like a price list: one owner, dated versions, and a retest after each change. When fees, hours, territory or escalation contacts change, update the brief, then call the receptionist to check that it changed too.

Smith.ai’s pricing page lists a “Quality Studio” to test before going live, with 25 real calls a month on its free tier (Smith.ai). Whatever you buy, use a test line like that after every change.

Tip: Save the previous brief next to the current one. When something goes wrong, you can restore the last version that passed your test calls.

This is step one of four. Next, score vendors with the demo checklist, check the integration matrix, and run the 30-day pilot. To pick who to demo, start with AI receptionists for contractors.

What we couldn’t verify: how often each vendor re-reads your website or Google profile after setup, and how a change in your booking settings reaches a receptionist mid-call; we have not configured any of these products ourselves.

Assign an owner to every rule that can change

The deliverable here is an approved input sheet, not a demo score. Add these columns to each business rule: approved value, effective date, office owner, next review and test call that proves the revision. Keep separate rows for service area, diagnostic fee, business hours, appointment capacity and on-call destination. A price or emergency instruction without an owner can stay wrong after the office changes it.

Jobber’s Receptionist help, rechecked October 19, 2026, distinguishes escalation destinations and configured booking behavior. Ask the vendor to show the destination and action your account will actually use; do not copy the advertised number into every forwarding and escalation field. After changing a rule, rerun the corresponding fictional call from the demo checklist before extending coverage. This is a proposed approval process, not an implementation we performed.