Skip to main content

Summary Doesn't Match Meeting Content

Check What Was Recorded

To check what was actually recorded, click the share icon at the top of your summary, click "Export as.." and choose Export Audio.

If you notice that your AI-generated summary contains content that has nothing to do with your actual meeting, you may be experiencing a hallucination issue with the speech-to-text processing.

What is Happening​

This is an industry-wide issue affecting all AI speech-to-text systems when they encounter minimal or poor quality audio. When the AI model detects little to no clear speech, it may "hallucinate" and generate content that wasn't actually spoken during your meeting.

Common Causes​

The most frequent causes of this issue include:

  • Phone positioned too far away - The device is placed at a distance where it cannot clearly capture the conversation
  • Headphones plugged in but not in use - Audio input is being routed through headphones that are not being worn or are in another room
  • Very quiet speakers - Participants speaking too softly for the microphone to detect clearly
  • Poor room acoustics - Background noise drowning out speech or echo interfering with audio quality
  • Microphone obstruction - Something blocking or covering the device's microphone

How to Prevent This Issue​

To ensure accurate transcription and summaries:

  1. Position your device properly - Keep your phone or recording device within 3-6 feet of the speakers
  2. Check audio input settings - Ensure audio is being captured through the device's built-in microphone, not external headphones
  3. Test your setup - Before important meetings, do a quick test recording to verify audio quality
  4. Minimize background noise - Choose quiet environments when possible
  5. Speak clearly - Encourage participants to speak at a normal volume and pace

What to Do If This Happens​

If you encounter a hallucinated summary:

  1. Don't rely on the summary - The content is not accurate and should be disregarded
  2. Check your audio setup - Review the common causes above for your next recording
  3. Re-record if possible - If the meeting is still ongoing or can be repeated, restart with proper audio positioning

Technical Background​

This is a known limitation of current AI speech-to-text technology. When AI models receive insufficient audio input, they may attempt to "fill in the gaps" based on their training data, resulting in content that sounds realistic but has no connection to your actual meeting.

We are actively working with our AI providers to improve detection of low-quality audio scenarios and provide better warnings when this might occur.

Still Having Issues?​

If you continue to experience summary hallucinations despite following these guidelines, please contact our support team with details about your recording setup and environment.