Teach an AI Receptionist to Answer FAQs Correctly
Find the questions callers really ask, write voice-friendly answers, handle edge cases, and measure how many calls your AI receptionist resolves.
To teach an AI receptionist your FAQs, collect the questions callers actually ask (from voicemails, texts, emails and staff), write a short, spoken-style answer for each with any conditions spelled out, add a rule to defer anything not covered, and then review real calls to find and add missing questions. Ten to twenty well-written answers usually cover most routine calls.
Step 1: Find the real questions
Your guess about what customers ask is usually incomplete. Gather from:
- Voicemails and their transcripts from the last month
- Inbound texts and emails
- Staff memory: ask whoever answers the phone to list the top ten questions
- Website contact forms
- Online reviews that mention confusion ("I didn't know they closed at 4")
Group similar questions. "When do you close?", "Are you open Saturday?" and "What time can I come by?" are one topic: hours.
Step 2: Rank by frequency and value
| Topic | How often | What happens if unanswered |
|---|---|---|
| Hours | Very often | Caller may go elsewhere |
| Service area | Often | Wasted callback if you don't serve them |
| Services offered | Often | Lost lead |
| Pricing basics | Often | Lost lead or wasted visit |
| Booking process | Often | Lost booking |
| Payment methods | Sometimes | Minor |
| Parking/directions | Sometimes | Minor |
| Warranty/guarantee | Sometimes | Trust |
Write answers for the top of the list first.
Step 3: Write answers for the ear
Each answer should be one or two sentences, in full words, with no symbols or URLs:
Hours: "We're open Monday through Friday, 8 a.m. to 5 p.m., and Saturday 9 a.m. to noon. We're closed Sundays and major holidays."
Service area: "We serve Dayton, Kettering, Oakwood and Centerville. We don't currently travel to Springfield."
Pricing: "A standard drain cleaning is $149 for one drain. Bigger jobs are quoted after we see them, and the estimate visit is free."
Booking: "I can take your details and preferred time, and the office will call or text to confirm the appointment."
Payment: "We accept all major cards, checks and cash. We don't offer financing."
See AI agent business facts sheet for the full template.
Step 4: Spell out conditions and exceptions
Most wrong answers come from answers that are true only sometimes. Add the condition to the answer:
| Vague | Specific |
|---|---|
| "We offer free estimates." | "Estimates are free for replacement jobs. Repair visits have an $89 diagnostic fee, credited if you go ahead." |
| "We're open Saturdays." | "We're open Saturdays 9 to noon, except holiday weekends." |
| "We guarantee our work." | "Repairs carry a one-year labor warranty. Parts follow the manufacturer's warranty." |
Spell it out. The agent can't know about the exception you didn't mention.
Step 5: Add a defer rule
In the instructions, add: "If a question isn't answered in the business facts, say the team will follow up, and take a message." This one rule prevents most invented answers. See stop an AI agent from making things up.
Step 6: Handle "it depends" questions
Some questions can't be answered without details: "How much to fix my AC?" Give the agent a pattern:
- Say it depends and why ("It depends on what's wrong")
- Give whatever fixed piece you can ("The diagnostic visit is $89")
- Offer the next step ("I can get you on the schedule")
That turns a non-answer into a booking.
Step 7: Measure resolution
After two weeks, read conversations and sort each call:
- Resolved: caller got the answer and needed nothing else
- Resolved + action: answer given, then booking or message
- Deferred: agent didn't know; message taken
- Wrong: agent gave incorrect information
- Transferred: sent to a person
Every "deferred" call on a question you could have answered is a fact to add. Every "wrong" answer is urgent to fix. See AI agent performance metrics.
Questions an AI receptionist shouldn't answer
Even with facts, keep these with your team:
- Specific availability ("Can someone come at 3 today?"), unless the agent can see your live schedule
- Account-specific details it can't access (balances, order details in another system)
- Advice in regulated areas (medical, legal, financial, tax)
- Exceptions and discounts
- Complaints that need judgment
Write a rule for each so the agent knows to take a message or transfer.
Keeping answers current
FAQs go stale quietly. Set reminders:
- When anything changes (hours, prices, staff, services), update the same day
- Before holidays, add closure dates
- Monthly, read the full list aloud and check every line
- After each call review, add any new questions
A wrong FAQ answer repeated to every caller does more damage than a missing one.
Accuracy and advertising claims
What your AI says on the phone is your business making a claim. If an answer overstates a warranty or promises results, you own that statement. Keep answers factual and match what's in your written policies. For outbound calls, federal telemarketing rules also prohibit misrepresentations about goods, services and terms, so phone answers should be as accurate as anything you'd put in writing.
FAQ setup checklist
- Collected questions from at least four sources
- Grouped into topics and ranked
- Wrote spoken-style answers for the top 10 to 20
- Added conditions and exceptions
- Added a defer rule in instructions
- Wrote "it depends" patterns for pricing
- Listed topics the agent must never answer
- Scheduled a two-week review
FAQs for different roles
The same business may need different FAQ sets per agent:
- Receptionist: hours, location, services, booking process, who to talk to about what
- Sales rep: services in detail, price ranges, service area, how quotes work, timelines
- Support rep: troubleshooting steps, warranty terms, how to request service
- Appointment setter: appointment types, durations, preparation instructions, cancellation policy
Keep shared facts (hours, address) identical across agents. If you update one, update all.
Using FAQs in Callata
In Callata, your FAQ answers go in an agent's "Business facts the agent can share" field. Callata agents are told to state only facts from that field and to say the team will follow up on anything else. For callers who are already in your contacts, the agent can look them up to greet them by name and see notes and past call summaries, which helps with "I called last week about..." questions.
When the agent can't answer, it takes a message for your team or schedules a callback request. If texting is enabled, it can text a caller details such as your address. AI minutes are $0.25 per minute; Callata Office is $99 per month with five users included. Add your FAQs to a Callata agent.
Frequently asked questions
How many FAQs should an AI receptionist know?
Start with the 10 to 20 questions that make up most of your calls. Add more as call reviews reveal gaps. Very long FAQ lists slow responses without improving accuracy much.
What if a caller asks a question that's close to an FAQ but different?
The agent may stretch the closest answer to fit. Prevent this by writing conditions into answers, such as 'for residential customers' or 'except holidays', and adding a rule to defer when unsure.
Can the AI receptionist look up a customer's specific information?
Only if the system connects it to that data. In Callata, agents can look up a caller in your contacts to see their name, company, notes and past call summaries, but they don't access external systems like your billing software.