How do I stop it from summarizing the first 5 minutes of small talk?

I've been evaluating MeetGeek for my team's technical stand-ups and project syncs over the last three weeks, and there's a consistent, glaring flaw in its summary generation that's rendering the output nearly useless for our purposes. The AI seems pathologically incapable of distinguishing substantive discussion from administrative preamble. For every 25-minute meeting, I get a summary where the first third is a detailed breakdown of weather chat, weekend plans, and coffee machine talk.

This isn't a minor annoyance—it's a fundamental failure in segmenting meeting discourse. The value of an automated summary is to distill signals from noise, not to meticulously transcribe the noise itself. My team's meetings follow a loose structure: ~5 minutes of informal check-in, followed by 20 minutes of dense technical discussion on pipeline failures, schema changes, or latency spikes. MeetGeek's output buries the critical technical action items under a pile of irrelevant "how was your weekend" commentary.

I've exhaustively tested the obvious fixes without success:
* Adjusting the "summary length" slider only condenses the entire transcript proportionally, still dedicating 30% of a short summary to the off-topic start.
* The "highlight key points" feature occasionally ignores the preamble, but it's inconsistent and misses crucial technical context from the main discussion.
* I've reviewed the raw transcript, and the tool is simply applying its summary logic uniformly across the entire timeline, with no apparent segmentation or topic detection.

My configuration is straightforward. I'm using the "Detailed" summary format, with speaker labels enabled.

```json
// Hypothetical config alignment - our meetings are on Google Meet
{
"integration": "google_meet",
"summary_format": "detailed",
"speaker_detection": true,
"keyword_highlighting": true
}
```

The core question is this: does MeetGeek provide **any** mechanism—hidden setting, API parameter, or workflow—to define a summary start point, apply a topic sensitivity threshold, or instruct the model to identify and suppress summary of informal segments? I need the tool to recognize that the substantive meeting begins at the first instance of a technical keyword (e.g., "pipeline," "error budget," "PR"), or after a fixed initial time window.

If the platform lacks this functionality, its utility for any structured professional meeting is severely compromised. I'm looking for a concrete, technical solution, not a workaround involving pre-meeting scripts to mute the first five minutes (which is a dysfunctional requirement).

What are other data teams doing? Has anyone reverse-engineered a reliable method to force the AI to skip the preamble, or is this a fundamental architectural limitation of MeetGeek's current model?

—davidr