My bot is not answering: a checklist

Start with these five checks

When the bot goes quiet, or keeps declining questions it should be able to handle, the cause is almost always one of five things. Work through them in order. Each check takes a minute or two.

• Training has not finished, so the bot has nothing to answer from yet.
• Your monthly message quota is used up, which stops the bot everywhere.
• Strict answer style is declining questions that fall outside your trained content.
• The question is not covered by any source you have trained on.
• The AI provider is having an outage and the built-in fallback is answering instead.

1. Check that training has finished

Open your chatbot and go to the Files & Data Sources tab. Every source shows its status here. The bot can only answer from sources that have finished training and show as ready. If a source is still processing, wait for it to finish, then test again in the Playground tab.

One thing to know about website sources: a crawl covers up to 60 pages per source and runs on a time budget of about 20 seconds, so a very large site trains partially and is still marked ready. If answers about deeper pages are missing, add those pages as separate sources or use a sitemap source instead.

2. Check your monthly message quota

The monthly message quota is the one hard limit in Matram. When your workspace reaches it, the bot stops answering on every surface: the widget, connected channels, and the API. Visitors see this message in the widget:

"This assistant has reached its usage limit for now. Please try again later."

If that is what your visitors see, you have three options:

• Upgrade your plan. Basic includes 2,000 messages a month, Standard 5,000, and Pro 20,000.
• Add the extra-messages add-on, which gives you 5,000 more messages a month.
• Wait for the cycle to reset. The reset is anchored to your billing date, not the calendar month.

Upgrades and add-ons live on the Billing & Usage page, which only the workspace owner can use.

3. Check your answer style

Open your chatbot and go to the Response Engine tab. The default answer style is strict: answers stay grounded in your trained content, and the bot declines questions that content does not cover. That is usually what you want, but it can look like the bot refusing to help.

• If the bot declines questions your content should answer, the content is missing. Add it (see the next check).
• If you want the bot to handle small talk and general questions too, switch the style to flexible. Facts about your company are still answered only from your trained content.

4. Check whether the question is covered

Ask the failing question in the Playground tab and look at the citations under the answer. The bot shows up to four source pills, so you can see exactly which pages it drew from. No citations, or citations pointing at the wrong page, mean the answer is not in your trained content.

• Add the missing page as a source on the Website Links tab, then retrain.
• For a fact that is not written down anywhere, add it as a snippet on the Text Snippets tab.
• For a question that needs one exact, fixed answer, add it on the Custom Responses tab. Custom responses match the question directly and skip retrieval and AI entirely.

5. Provider outages

If the AI provider behind your bot has an outage, Matram falls back to its built-in ML model so the chat never goes silent. The fallback handles greetings and basic questions, but its answers are simpler than usual. If your bot suddenly sounds short and basic rather than silent, this is the likely cause. There is nothing to fix on your side; full answers return when the provider recovers.