Troubleshooting the qualification bot

A qualification funnel has few moving parts, so when it misbehaves the cause is almost always in one of the five modules — the AI's knowledge, its qualification setup, its automations, the handoff, or the channel. Work through the symptom that matches. Every fix below can be applied two ways — change the setting at Fuely AI → Automations, or tell the AI Co-Worker what you want — so use whichever is faster.

The bot doesn't reply at all

Check these in order — the first hit is your culprit:

1. Is Main bot on? It's the master switch: Main bot off means no AI replies anywhere, whatever the other toggles say. Check the top of the Automations page, or ask the Co-Worker: "Is the AI answering on WhatsApp right now?"
2. Is the source's own automation on? Every source's base automation starts off until someone enables it — so a client whose ad leads get answered can still have silent DMs, simply because nobody ever turned the DM automation on. Find the source the missing reply belongs to and check its toggle.
3. Is another rule catching the message? When custom rules target keywords or specific posts and ads, a priority order decides which automation replies — a content-targeted rule beats a general one, and turning a base automation off doesn't silence the source while an enabled custom rule still matches. If the symptom is "the wrong reply" rather than no reply at all, this is almost always why.
4. Is When AI replies set to "outside working hours" — during working hours? Then the silence is by design: daytime chats are the team's job. If the team isn't actually answering, either staff the inbox or switch the mode to always.
5. Is the channel still connected? Connections occasionally drop. Check the channel's page under Settings — if it shows disconnected, Reconnect fixes it in one click.

The bot replies, but vaguely — or makes things up

A bot that says "I'm not sure", gives generic non-answers, or quotes a price the client never approved has a starving Knowledge Base. The AI can't know what nobody told it — and it should never be guessing prices from thin air.

Fix: find the gap and fill it. Ask the bot the failing question in the Test chat, then add the missing fact — on the Knowledge Base page or by handing it to the Co-Worker: the service and its real price into the catalog, the policy into the FAQ. Here's what a complete Knowledge Base looks like. Tone problems live one page over: how the bot talks is set in the automation's Additional instructions — edit them directly, or tell the Co-Worker what's off ("answers are too long and formal — make them short and friendly") right while you're testing.

Qualified leads never reach the client's team

The bot chats away, leads answer every question — and nothing ever lands in the team inbox. The handoff isn't armed: either the automation's Switch to human setting is off, or no rule says when to switch.

Fix: open the automation and add the rule under Switch to human, or tell the Co-Worker: "Hand the chat to the team once the bot has collected all qualification fields." Then verify end-to-end — play a lead in the Test chat, answer all the questions, and confirm the chat lands in the team inbox. If the client's staff should be picking these up on their phones, make sure they've installed the Chatfuel mobile app: it pushes a notification the moment a qualified lead arrives.

The bot dumps every question into one message

"Which district, what budget, and what's your timeline?" — that's a form, not a conversation, and leads abandon it. Same family of symptom: the bot opens with "we only work with clients who..." or interrogates like a census taker. The conversational rules are missing from the automation's Additional instructions.

Fix: redo the qualification setup — these are the rules to wire in: one question per message, intent first, a "why" before personal questions, WhatsApp-length replies. Write them into Additional instructions yourself, or ask the Co-Worker to redo the setup. Re-test in the Test chat until it reads like chat.

Note: the universal debugging loop works for every symptom — reproduce it in the Test chat, apply the fix in Automations or through the Co-Worker, re-test. The Test chat is a sidebar on each automation that simulates a conversation with exactly that automation's settings — a free, client-invisible sandbox. Nothing you do there touches real leads, so break things freely until the conversation feels right.