Why the AI Is Not Replying: 7 Reasons and Their Fixes
The seven most common causes, ranked by how often each one is the culprit
AI not replying almost never means the AI is "broken" in a general sense — it means one specific condition is stopping a reply on that particular conversation. Here are the seven causes, checked in the order they're most likely to be the actual reason.
Reason 1: AI not replying because the toggle is off
The most common cause by far: AI Actions has a per-conversation toggle, and it's easy to switch off during a manual reply and forget to switch back on.
Open the conversation in your inbox and check the AI toggle at the top of the thread. If it's off, the AI won't respond no matter how well everything else is configured — this isn't a bug, it's working as designed to let a human take over a conversation.
Reason 2: Gemini key invalid
WaSMS's AI runs on your own Gemini API key. If that key was revoked, regenerated elsewhere, or simply mistyped during setup, every AI request fails silently from the customer's point of view — no reply, no visible error to them.
Check the AI settings page for a connection status next to your Gemini key. Gemini API free for business covers where to generate and paste the key correctly if you need to redo this from scratch.
Reason 3: quota exceeded
Gemini's free tier includes a generous daily request quota, but it isn't unlimited. A conversation-heavy day, or a sudden spike from a campaign, can hit that ceiling — and once it's hit, replies stop until the quota resets.
Quota exhaustion is easy to miss because everything looks configured correctly — the key is valid, the toggle is on, and the AI still doesn't answer. Checking usage against quota is the one step people skip because nothing on the WaSMS side looks wrong.
Reason 4: guardrail hit
AI Actions includes guardrails that stop it from replying automatically in certain situations — a message containing something that looks like a complaint, a legal threat, or a request outside what the AI is configured to handle. This is intentional: it's meant to route sensitive conversations to a human rather than let the AI guess.
AI safety and guardrails for business explains which situations trigger a pause and how to see when a guardrail, rather than a technical fault, is why a reply didn't go out.
Reason 5: escalation triggered
Related to guardrails but distinct: some conversations are explicitly configured to escalate to a human agent after a certain point — for example, once a customer asks for a refund, or after the AI has replied a set number of times without resolving the issue.
If a conversation was escalated, the AI intentionally stops replying and waits for a human. Check the conversation status; if it shows "escalated" or "assigned," that's the answer. AI escalation: when to hand off to humans covers how to configure or adjust these rules.
Reason 6: no training data
If the AI is live, the key is valid, and there's no guardrail or escalation in play, but replies feel generic or simply don't come for certain topics, the AI may not have enough business-specific context to answer confidently — and it's often configured to stay quiet rather than guess wrong.
Training the AI on your business data walks through feeding it the conversations, FAQs, and product details it needs so it has something to answer from instead of nothing.
Reason 7: outbound approval pending
Some WaSMS setups require a human to approve AI-drafted replies before they send, rather than sending automatically. If that mode is on, the AI has technically already "replied" — it's just sitting in an approval queue waiting for someone to confirm it.
Check the Pending Approvals section of your inbox. This is easy to miss if the team isn't used to checking it regularly, since it looks like nothing happened rather than like something is waiting.