Why Does My AI Receptionist Give Customers Wrong Answers?

Niamh runs a locksmith business in Cork with one apprentice. She tried a cheap answering service once — the agent on the other end couldn't tell a euro cylinder from a mortise lock. The caller hung up. The job went elsewhere. When she finally moved to an AI receptionist, it started inventing her opening hours. The technology is brilliant. The setup is where it all goes wrong. Here's the honest version of what causes it, and what actually fixes it.

An AI receptionist that gives customers wrong answers is worse than no receptionist

Here's the uncomfortable truth nobody in the AI hype videos tells you: an AI receptionist can be confidently wrong. Not "I don't know" wrong. Confidently wrong — it will tell a customer you're open until 5pm when you close at 2, quote a price you've never charged, or book a slot you're already working in. The customer doesn't think "the software glitched." They think your business is unprofessional [6].

In the source demo that this post is based on, the AI was asked what time the business opened. Its prompt contained no business details at all — so it invented "8am to 5pm." The real hours were 10am to 2pm [6]. That is the single most common failure mode of DIY AI receptionists: the person setting it up never gave it the facts.

Before: The Night the Answering Service Cost Niamh the Job

Niamh took over her uncle's locksmith van in 2021. The day job is fine — but the money in locksmithing is in emergencies, and emergencies don't ring twice. At 3:12am her phone rang. An elderly pensioner had locked herself out. Niamh was mid-job on the other side of the city, and the cheap answering service she'd hired took the message: "She wants to know how much it costs and how fast you can get there."

The pensioner didn't wait for a callback. She dialled the next locksmith on Google, who answered at 3:15am, quoted a fair price, and sent a van. The €280 job went to a competitor who simply answered [6]. Worse: Niamh had already lost two insurance-firm contracts because their approved-list callers couldn't get through. Insurance companies send work to people who answer — and they blacklist people who don't.

That's the pattern across every trade. Home services businesses miss roughly 35% of all inbound calls, and 78% of those callers go straight to a competitor who picks up [3]. For Niamh, that meant emergency callouts — her highest-ticket work — leaking to whoever picked up the phone first.

After: The AI Receptionist — When It Works, It Works

The fix isn't hiring more people. It's an AI voice receptionist that picks up every call on the first ring — including 3:12am on a Tuesday [6]. The core principle from the source material: "Missed calls means missed revenue. Every unanswered call is a potential lost client. The goal is to answer each call in 20 seconds." [6]

Here's what a properly configured AI receptionist does for a locksmith [6]:

• The call connects in under 3 seconds. The AI greets the caller with your company name — no "press 1 to leave a message."
• It qualifies the emergency. Locked out? Broken key in the lock? Commercial premises? It captures the address, the lock type, and the access details — just like a trained dispatcher.
• It checks your live calendar. The agent is wired to your team calendar through booking tools, so it only offers slots that are actually free [6].
• It books the job on the spot. The appointment appears in your calendar with the customer's name and phone number attached. No callback needed [6].
• It sends you a summary. After each call, you get a text or email with the caller's details, the problem, and the scheduled appointment [3].

The result for Niamh: every one of the 3am emergency calls gets answered. The lockouts get triaged and booked. Nothing goes to voicemail, and nothing leaks to the competitor who picks up at 3:15am.

Why the Same Technology Gives Some Businesses Wrong Answers

Here's the part the AI companies don't put in the brochure. The same platform that books a job perfectly can tell a customer you're open until 5pm — because the person who set it up never gave it the business facts. There are exactly four reasons an AI receptionist gives wrong answers, and every one of them is fixable [6]:

Why Voicemail, Answering Services, and DIY Setups All Fail Differently

Voicemail is a dead end — fewer than 20% of callers leave messages, and the ones who do often hire someone else before you call back [1]. A full-time receptionist costs £2,800-4,200 a month and still goes home at 5pm — exactly when the emergency calls peak [3]. Cheap answering services take messages; they don't know a euro cylinder from a mortise lock [6]. And DIY AI setups work — until they hallucinate your opening hours to a paying customer [6].

The Objections — Answered Honestly

The Behavioural Shift You Can't Ignore

Customer expectations have already changed. Callers now expect an instant answer — and Google surfaces the businesses that respond fastest. The AI customer service market keeps growing through 2026 [4], and every major voice platform now ships conversational agents out of the box [6]. The homeowner with a broken lock at 3am isn't comparing you to a human receptionist. They're comparing you to the locksmith who answered first.

The 12-Month Forecast

Here's where this is going. In 12 months, the trades businesses that answer every call — day and night, correctly — will own the emergency work in their cities. The ones that don't will keep feeding their competitors the 3am calls. In 24-36 months, AI answering will be the default expectation, like websites were in the 2000s. The businesses that adopt it now, configured properly, will be the ones with the reputations — and the ones that tried a half-configured bot and got burned will be the cautionary tales at industry meetups.

The behavioural change is irreversible: homeowners now expect an instant answer, and AI search engines are learning to surface the businesses that respond fastest [4]. An AI receptionist isn't a gadget. It's the difference between being the company that picks up and the company the caller tried first.

There's a Simpler Option Than Building It Yourself

DIY is real — the source material shows a full no-code build with Vapi, n8n, Twilio, and Google Calendar [6]. But DIY means you own the maintenance: the prompt tuning, the model updates, the latency fixes, the edge cases where the AI hallucinates your opening hours. That's a part-time job you didn't hire for.

One service. Zero hardware. Zero plugins. Every call answered, every emergency triaged, every job booked — and none of it hallucinated. The AI receptionist for your trade business — set up in 5 days, configured so it never tells a customer you're open until 5pm when you close at 2.

Sources & Evidence