A chatbot estimate is not a quote

A homeowner messages a plumbing company's website on a Sunday afternoon. The water heater is leaking and they want a ballpark. The chatbot is friendly and fast: "A standard 40-gallon water heater replacement usually runs about $1,200, installed." The homeowner books a Monday visit and screenshots the answer.

On Monday the technician finds a gas unit in a tight closet, a corroded shutoff valve, and a venting setup that no longer meets code. The real job is closer to $2,600. The homeowner holds up their phone and says, "Your own website told me twelve hundred." Now the owner has a choice between eating the difference and arguing with a customer about something their own software said.

Nobody told the chatbot to quote prices. It found a number on an old blog post about water heaters, saw a question that looked like a price question, and did what language models do: it gave a confident, helpful-sounding answer.

Hive AI builds chatbots, voice agents, and AI integrations for owners who want an assistant that answers from sources they approved, inside limits they chose, with a record of every answer it gives.

Why the chatbot "quoted" anything

A general-purpose chatbot doesn't know the difference between an example price in an article, a number on a seasonal promotion that ended in March, and the rate your office actually uses. To the model they are all text near the words "water heater" and "cost." If the question sounds like a price question, the most helpful-looking reply contains a price.

That isn't a bug you fix with a better prompt. "Never give exact prices" in a system prompt works until a customer asks in a way the prompt didn't anticipate, or until the model decides that "about $1,200" is a range rather than a price. A prompt is an instruction. It isn't a limit.

What an estimate actually is

When your office gives an estimate, it comes from a specific place: a price book, a service menu, a rate sheet, or a technician who has seen the job. It has conditions attached, like "standard install, existing venting, no code upgrades." Someone in your business stands behind it.

An assistant that's allowed to talk about price should work the same way:

• it answers price questions only from a price source you approved and versioned, not from your blog, your old promotions, or the open web
• it states the conditions that come with the number, every time, in the same words your office uses
• it knows which jobs have no phone or chat price at all and says so plainly, then offers a visit or a call instead
• it never turns an estimate into a commitment: no "guaranteed," no "that's the total," no discount it wasn't given

The version that protects you

Here's how the same Sunday conversation goes with an assistant built that way:

"For water heater replacements we give a firm price after a technician sees the unit, because gas, venting, and valve condition change the job a lot. Our published starting price for a standard electric 40-gallon swap is on our service menu, and it assumes existing hookups that meet code. I can book a visit for Monday morning, and the technician will give you a written quote before any work starts."

The homeowner still gets a useful answer, still books Monday, and doesn't arrive with a screenshot of a number your business never set.

Keep the receipt

Every answer the assistant gives about price should leave a record you can read later:

• the question exactly as the customer asked it
• the answer the assistant gave, word for word
• the price source and its version, and the passage the answer came from
• whether it declined to give a number, and why

When a customer says "your website told me," you don't have to guess. You open the receipt, see which version of your price menu it used and what conditions it stated, and decide from facts instead of from a screenshot.

Why this matters before you buy

Most owners worry about a chatbot saying something rude. The more expensive risk is a chatbot saying something specific: a price, a timeline, a warranty term, a "yes, we cover that." Every one of those turns into a promise the moment a customer reads it. The fix isn't a chatbot that says less. It's one that knows which answers belong to you, where they come from, and when to hand the question to a person.

Start with one workflow

Work with Hive AI starts with a five-day Bootcamp: one workflow, your real documents and price sources, and a Friday recommendation with an evidence path. You leave knowing what your assistant may say about price, where every number comes from, and what each answer looks like on your side.

Start a HIVE Bootcamp or talk to Hive AI about a chatbot that answers from your approved sources