Hi everyone! 👋 I just started a trial of Read AI for my project team. So far, I'm really impressed with how it summarizes meetings, but I keep running into the same issue.
We use a lot of specific acronyms and internal project names (like "Project Bluebird" or "the CRF process"). Read AI sometimes misinterprets these or writes them out in a weird way that doesn't match how we talk. I feel like the summaries would be way more useful for my team if it understood our lingo.
I've looked around the settings but I'm a bit lost. Is there a way to "teach" Read AI our specific terminology? Do I need to upload a glossary document somewhere, or is it more about correcting it over time?
Also, if it *can* learn, does that training apply to all meetings it processes for my workspace, or just the ones I'm in? My team uses Asana and Slack heavily, so I'm wondering if integration with those tools helps at all.
Any guidance from people who've figured this out would be amazing!
> "Do I need to upload a glossary document somewhere, or is it more about correcting it over time?"
I've been using Read AI for about three months on a product team with a similar problem (acronyms like "PDLC" and "QBR" plus internal codenames). The short answer is: there is no explicit glossary upload function. The learning mechanism is entirely implicit based on corrections you make in the summary editing interface. You correct a misinterpretation, and over time the model should adjust. But "should" is doing a lot of work here.
In my experience, the correction signal is weak and slow. I've corrected "Project Bluebird" being rendered as "Project Blue Bird" at least six times across different meetings, and it still occasionally flips back. The model seems to learn per-meeting more than per-workspace. I've seen no evidence that correcting one meeting's summary propagates to all future meetings in the workspace. So if you're hoping for a single "teach it once" workflow, you'll be disappointed.
Regarding your Asana and Slack question: integration does help, but indirectly. Read AI can pull in messages and task references from those tools, which gives it more context for names and project mentions. However, I haven't seen it explicitly use those to update its internal term dictionary. The integration is more about providing meeting context (e.g., "This meeting was about Asana task #142") than about training terminology.
One thing I've found useful: before each meeting, I paste a short list of key terms and their correct spelling into the meeting notes field in Read AI's calendar integration. That seems to prime the model for that specific session. It's a kludge, but it works better than pure post-hoc correction.
If you want a more systematic approach, you might also try using the "custom vocabulary" feature in the transcription settings if you're using a third-party transcription service alongside Read AI. Read AI itself doesn't have that, but some users combine it with Otter.ai or Fireflies which do accept glossary uploads. Not ideal, but it's a workaround.
Has anyone else found the learning model to be more persistent than I have?
Data > opinions
The idea that you can "teach" a hosted AI model your proprietary jargon is mostly marketing. These services don't have a training feedback loop you can trust, they have a post-processing step that might apply your corrections to future transcripts, maybe. It's opaque and unreliable.
You'll spend more time correcting "Project Bluebird" than you saved by using the tool. If you need this level of control, you need a system you can actually train, not one that treats your corrections as a vague suggestion for a model you don't own.
Your question about integrations is on point. No, linking Asana or Slack won't help. It's separate data ingestion pipelines, not a unified learning system. The model stays generic.
Just my 2 cents