Aimogen ProDocumentation
Product page Support

Chatbot troubleshooting

Symptom-first fixes for Aimogen Pro chatbot problems, from a bot that does not appear to one that invents answers or costs too much.

The chatbot does not appear#

Symptoms. The shortcode renders nothing, or the global bubble is missing.

Likely causes

  1. No API key. Most chatbot screens refuse to render without one. Check Settings › API Keys and use Test API key.
  2. The plugin main switch is off. Settings › General Settings.
  3. A global condition excludes this view. URL, role, device, browser, language, IP or time. See Global and conditional chatbots. Excluding administrator is a common cause of "it works for visitors but not for me".
  4. The shortcode is not executing. Check for a typo, and that the theme renders shortcodes in that template.
  5. Page caching. Clear the cache.

How to diagnose. View source and search for the chatbot container. Present but invisible is a CSS problem; absent entirely is one of the causes above.

The chatbot appears but does not respond#

Symptoms. Messages send, nothing comes back, or an error appears.

Likely causes

  1. The API call is failing. Check System & Logs › Activity Logs for the provider error, then API and provider errors.
  2. A usage limit was reached. The visitor sees the limit message. Check System & Logs › AI Usage Limits.
  3. Rate limited. Limit User Messages was exceeded.
  4. admin-ajax.php is blocked. A security plugin or WAF. See Plugin and server conflicts.
  5. The AJAX response is cached. Exclude admin-ajax.php from page cache.

How to diagnose. Open the browser developer console. A failing request to admin-ajax.php with a 403 points at a firewall; a 200 with an error payload points at the API.

Replies are cut off#

Raise Chatbot Max Token Count. If it is already high, the model output limit or PHP timeout is the constraint — see Tokens, context and output length.

The bot invents answers#

Likely causes and fixes

  1. Bot Context does not forbid it. Add an explicit instruction to say "I do not know". See Bot configuration.
  2. No retrieval configured. The bot only has the model training data. See Knowledge and context.
  3. Retrieval is configured but not enabled for the chatbot. Tick Chatbot Shortcodes under Settings › Embeddings › Enable Embeddings For. This is the single most common cause.
  4. Temperature too high. Lower it to 0.3–0.7 for factual bots.

The bot forgets the conversation#

Max Chat Messages To Send As API Context is too low, or Max Chat Context Size (Characters) is truncating. For memory across sessions, enable persistent chat.

Extensions never fire#

  1. The model does not support function calling. Check the capability list in Selecting models.
  2. Extensions are not enabled for this placement. Preview, global and per-shortcode are separate switches. See Chatbot Extensions.
  3. A provider Responses API toggle is on. Several beta implementations do not support tool calls. Turn it off under Settings › Advanced AI Settings.

Streaming does not work#

  1. The model does not support streaming.
  2. A Responses API toggle is on.
  3. Server buffering. Some hosts and reverse proxies buffer the response, defeating streaming.
  4. Disable Streaming is set on the placement.

Note that retries do not work in streaming mode — see Errors, retries and timeouts.

Voice does not work#

  1. The page is not HTTPS. Browsers require a secure context for microphone access.
  2. Permission denied. Check the browser site settings.
  3. No voices in the dropdown. The speech provider key is missing or invalid.
  4. Azure avatar blank. Free tier, wrong region, or an ad blocker. See Voice, speech and talking avatars.

Styling looks wrong#

  1. Theme CSS conflict. Use Custom CSS to raise specificity.
  2. The design changed the relevant option. Some styling settings apply to specific designs only.
  3. A shortcode attribute overrides the global setting. Attributes win.

The chatbot costs more than expected#

Work through, in order:

  1. Context size. Every message resends the history. Reduce Max Chat Messages To Send As API Context.
  2. Retrieval chunks. Every retrieved chunk is input tokens on every message. Reduce the top-K.
  3. Internet access. A search plus page fetches per affected reply. Turn it off unless needed.
  4. Model. A flagship model on a busy chatbot is expensive. A fast mid-tier model is usually better here anyway.
  5. Speech. Billed per character. Realtime is billed per minute of audio.
  6. No limits. Set usage limits.

Someone abused the chatbot#

Immediate steps:

  1. Set Max Guest Credits to zero, making the chatbot login-only.
  2. Add the offending IP to Restrict User IPs List From Using The AI.
  3. Set a rate limit under AI Chatbot › Limitations.
  4. Check System & Logs › Usage Logs for the extent.
  5. Set a spend cap in the provider dashboard.

Then re-open with limits in place. See Moderation, GDPR and limits.

Getting help#

Collect before opening a ticket:

  1. The error from System & Logs › Activity Logs
  2. The relevant Usage Logs rows
  3. The output of System & Logs › System Info
  4. Your shortcode, with any API keys removed
  5. Browser console output from a failing message

Then open a support ticket.

Still stuck? Open a support ticket and include the diagnostics from Aimogen Pro › System & Logs › System Info.