# Should a chatbot answer your leads, or should you approve each reply?

> A chatbot that answers leads on its own, or an AI-drafted reply a person approves first: speed, what each can promise, what happens when it's wrong.

If a lead's question has one right answer that's on your website, a chatbot can answer it on its own, at any hour. If the reply could quote a price, promise a date or set the tone with a new client, have AI draft it and a person approve it first. A chatbot's wrong answer can still count as yours.

You know when a lead writes in while you're with a client, or after you've gone home? If the reply waits until morning, they look at others in the market. Two kinds of AI tool promise to close that gap. A chatbot on your website answers the lead by itself. A drafted reply goes to a person first, and goes out when they approve it.

I built the second kind for my own business, so read its column with that in mind. On 2026-10-05 I sent my live inquiry desk a test lead from a Gmail address outside SignalNine. It drafted a reply, I approved the card on my phone, and the reply went out. I approve there because it's my business.

*Six questions to ask before AI answers a lead*

| Question | Chatbot | Drafted reply you approve |
| --- | --- | --- |
| How fast the lead hears back | Straight away, at any hour, while the lead is still on your website. | As fast as the approver gets to the card. Minutes when they're near their phone, the next morning when they're asleep. |
| What it can promise | Whatever it says. The lead reads it as the business speaking, and in one 2024 decision a tribunal held an airline responsible for what its chatbot told a customer. | Only what the approver lets through. The draft can also be checked for a made-up price or date before it reaches them. |
| What happens when it's wrong | The lead reads the wrong answer before anyone at the business sees it. | The approver catches it on the card and edits or rejects it. An email the AI wrongly sorts out as a sales pitch never reaches a card, so the approval can't catch that one. |
| Who sees the lead | Nobody at the business, until someone opens the chat history or the bot hands the lead on. | A person reads every lead it drafts for, with the draft beside it, before anything goes back. |
| Effort per lead | None once it's set up. The work moves to keeping its answers current as your services and prices change. | A read and a tap per lead, more when a draft needs changing. At high volume that becomes a job of its own. |
| Questions it suits | Ones with a single right answer: where you work, what you offer, how to book a call. | Ones that turn on scope, price, timing or fit, where a wrong answer is hard to take back. |

## Where each one fails

### A chatbot

A chatbot fails on the questions that matter most to a service business. A lead asks what a project like theirs would cost, or whether you could start next month, and the bot answers with the same confidence it uses for your opening hours. Once the lead has read it, it's hard to walk back.

British Columbia's Civil Resolution Tribunal ruled on that in a small claims dispute decided on 2024-02-14. An airline's website chatbot told a customer they could apply for a reduced fare after they had travelled, and the airline's own policy page said that wasn't allowed. The airline argued it couldn't be held liable for what its chatbot said. The tribunal found the airline responsible for all the information on its website, whether it comes from a static page or a chatbot, and ordered it to pay damages. That's one decision in one province, and the point carries anyway: a lead reads the bot as you.

A chatbot also goes stale. Set up with last year's services, it keeps describing them until someone changes what it reads, and nobody at the business sees the answers it gives in the meantime.

### A drafted reply you approve

A drafted reply fails on speed when the approver is away. A card that waits overnight is a reply that waits overnight, which is the delay you set out to fix. It also fails at volume. With many leads a day, approving each one becomes a job, and a tired approver starts pressing approve without reading.

The approval also covers only what reaches the card. In my inquiry desk, an AI step sorts real leads from sales pitches before anything is drafted. I tested that step alone on sample emails on 2026-09-12, with 60 runs per setting. With the old wording on the smaller model, it threw out an inquiry from someone describing how their own team works as a sales pitch in 8 of 60 runs. After I changed the wording and moved that step to the larger model, it was 0 of 60. An email thrown out at that step gets no card, so no approver ever sees it.

## What I'd pick, per situation

- If most leads ask the same few questions, like where you work or how to book a call, pick a chatbot for those. Give it only answers you'd sign off on, and have it hand anything about price, dates or scope to a person.
- If you get more leads a day than one person can read with care, pick a chatbot for the first answer and keep a drafted reply for the ones that turn into a real conversation.
- If each lead could become weeks of paid work, pick a drafted reply you approve. A short wait costs less than a promise you can't keep.
- If nobody can check their phone during the hours leads arrive, a drafted reply will sit. Name a second approver, or let a chatbot take the simple questions until someone is free.
- If you already answer every lead yourself within the hour and there aren't many, keep doing that. Adding AI there adds a step without saving one.

If I install one for your business, no reply the AI drafts for a lead goes out until someone at your business approves it from their phone. You choose who at setup, and SignalNine is never the approver. Everything else in the system acts on its own, like logging the lead or creating a task.

*Diagram: The drafted reply's path: the email is screened, the draft waits on a card in chat, and a person's approval sends it once. A chatbot has no card and no person between the lead and its answer.*

## Questions

**Can a chatbot quote prices to leads?**

It can, and a lead will read the quote as yours. In one 2024 decision, British Columbia's Civil Resolution Tribunal held an airline responsible for what its chatbot told a customer, the same as for a page on its website. If your prices depend on scope, have the bot pass those questions to a person.

**Is approving every reply too slow?**

It's slower than a chatbot by however long the card waits. When the approver is near their phone, that's minutes. When they're asleep, it's the next morning, so pick an approver who keeps their phone close during the hours your leads come in, or name two.

**Who approves the replies the AI drafts?**

Someone at your business, chosen at setup: you or someone on your team. No reply the AI drafts for a lead goes out until that person approves it from their phone, and SignalNine is never the approver.

**Can I use a chatbot and approved replies together?**

Yes, and it's the split I'd start with. Let the chatbot answer the questions with one right answer, and send everything else to a drafted reply that a person approves. Write down which questions the bot may answer, and keep that list short.

**What does a drafted reply need to know to be any good?**

What you sell, how you price, what you won't promise, and the earlier emails with that lead. Without those, it writes a polite reply that says little, and the approver ends up rewriting it.

[BOOK A FREE CONSULTATION](https://cal.com/mashrur-signalnine/30min)

## Sources

- [Civil Resolution Tribunal (British Columbia), Moffatt v. Air Canada, 2024 BCCRT 149, decided 2024-02-14: the airline's website chatbot suggested a reduced bereavement fare could be applied for after travel, while the airline's own policy page said it did not apply after travel; the airline argued it could not be held liable for information from its chatbot; the tribunal found it responsible for all the information on its website, whether from a static page or a chatbot, and ordered it to pay damages](https://decisions.civilresolutionbc.ca/crt/crtd/en/item/525448/index.do), read 2026-10-05

## Related

- [A drafted reply to each lead, approved from my phone before it sends](https://signalnine.io/md/systems/inquiry-desk): AI drafts replies to lead emails that pass a junk screen, from my business facts. Risky asks get flagged, and nothing sends until I approve it on my phone.
- [How do I approve email replies from Slack before they go out?](https://signalnine.io/md/integrations/gmail-to-chat-approval-to-reply): Approving email replies from Slack, Telegram or Google Chat: what the built-in apps do, where an approval loop breaks, the send-once rule, and what I'd pick.
- [Human-in-the-loop approval](https://signalnine.io/md/glossary/human-in-the-loop-approval): What human-in-the-loop approval means in an automation, which steps should wait for a person, and how an approval card makes sure an email sends once.
- [Send-once](https://signalnine.io/md/glossary/send-once): Send-once means one approval sends one email, however many times the button gets pressed. Why duplicates happen and how a system keeps the rule.
- [Speed to lead](https://signalnine.io/md/glossary/speed-to-lead): Speed to lead is the time between a lead reaching you and a real reply going back. Why it matters, an example, and how a system shortens it.
- [The lead-response bottleneck: how fast to answer a new lead](https://signalnine.io/md/bottlenecks/lead-response): How fast to answer a new lead, why a slow reply hands the decision to someone else, and a system that drafts from your rules for a person to approve.
- [Do you need an AI agent, or a workflow with one AI step?](https://signalnine.io/md/compare/ai-agent-or-workflow): When a business needs an AI agent that picks its own steps, and when a fixed workflow with one AI step is cheaper to run, easier to test and simpler to check.

---
Mashrur Rahman, Founder, SignalNine. Published 2026-10-05, updated 2026-10-05.
Id: C09
Canonical: https://signalnine.io/compare/chatbot-or-approved-reply