How to Search Past Meetings With AI Chat
Stop digging through notes to find what was decided. AI chat lets you search every past meeting with a plain-English question—here's how it works.
"What did we decide last Tuesday?" — the question everyone dreads.
Not because the answer was unimportant, but because finding it means opening a folder of meeting notes, scanning a transcript that's three pages long, and hoping someone labeled the right section. Or messaging three colleagues and waiting to see who replies first. Or accepting that the decision might just be lost and starting the conversation again from scratch.
According to research compiled across several workforce productivity studies, organizations lose an estimated $37 billion annually to unproductive meetings in the United States alone. A significant portion of that loss isn't the meetings themselves — it's the time spent afterward trying to recover what was said, who agreed to what, and what was supposed to happen next.
This is the problem AI chat search is built to solve. And it does it in a way that keyword search never could.
Automate your meeting notes. MinuteKeep records your meeting and uses AI to transcribe, summarize, and extract action items. 9 languages, no subscription, 30 min free.
Why Keyword Search Fails for Meeting Notes
Traditional search — the kind built into most note-taking apps — works by matching exact words. Type "Q3 budget" and it will find any note containing those two strings. That sounds fine until you realize how meetings actually happen.
The finance director said "third-quarter allocation." The VP said "the money we set aside for Q3." A product manager wrote "budget for July–September" in the shared doc. Your own notes say "the number Marcus mentioned." None of these will surface in a keyword search for "Q3 budget."
Meeting language is messy. It's paraphrased, context-dependent, and fragmented across multiple sessions. Keyword search assumes you remember exactly how something was phrased. You rarely do.
That's where AI-powered semantic search changes the equation.
How AI Chat Search Actually Works
The technology behind modern AI meeting search is called Retrieval-Augmented Generation, or RAG. The name sounds technical, but the idea is straightforward: instead of the AI relying only on what it was trained on, it first searches through your actual meeting notes, pulls out the relevant sections, and then generates an answer based on what it found.
Here's the process in plain terms:
Your notes are converted into vectors. When a meeting note is saved, it's processed into a mathematical representation of its meaning — not the exact words, but what those words mean. This is called an embedding.
Your question is also converted into a vector. When you ask "what did we decide about the Q3 budget?", that question gets the same treatment.
The system finds meaning-matches, not word-matches. It compares your question's meaning against the meanings stored from all your meetings, finds the most relevant sections, and retrieves them.
An AI generates a direct answer. It reads those relevant sections and writes a response in plain English — not a list of search results to wade through, but an actual answer.
The result is that you can ask questions the way you'd ask a colleague, and get answers that reflect the full context of what was discussed — even if the words you used don't appear anywhere in your notes.
MinuteKeep's AI Chat: How It Works in Practice
MinuteKeep's AI Chat feature searches across all of your stored meeting notes simultaneously. It uses OpenAI's text-embedding-3-small model to index the semantic meaning of every meeting you've recorded, then uses GPT-4.1-mini to generate conversational responses to your questions.
A few things worth knowing before diving into examples:
- It searches all meetings, not just one. You're not querying a single transcript — you're querying your entire meeting history at once. If a topic came up across five meetings over three months, the chat will draw on all of them.
- No extra cost. The AI Chat feature uses no time credits. It's included with the app and runs on top of notes you've already processed.
- Sync is automatic. When a new note is saved, the embedding process runs in the background. By the time you open the chat screen, your latest meeting is already searchable.
- Chat is session-based. The conversation resets each time you close the app. This is an intentional privacy decision — your questions aren't stored alongside your meeting notes.
Real-World Queries and What to Expect
Here are seven examples of the kinds of questions AI Chat is designed to handle, with an explanation of what kind of response you'd get.
"What did we decide about the marketing budget for Q3?" Returns a summary of the relevant decision, with context about who raised it and what the agreed number or approach was. If the topic appeared across multiple meetings, the chat synthesizes them.
"When did Sarah mention the new vendor?" Identifies the meeting where this came up and provides the relevant excerpt, including surrounding context so you know what the vendor discussion was actually about.
"What action items came out of the product roadmap meeting?" Lists the tasks and owners that were mentioned, drawn from the summary and transcript of that meeting. Useful when someone's accountability needs to be confirmed.
"Has the team discussed remote work policy changes?" A broad question that sweeps across all stored meetings. If the topic came up three months ago in a brief exchange, the chat will surface it — even if you've completely forgotten the conversation.
"What were the main concerns raised about the Q4 launch timeline?" Returns objections and risks that team members flagged, useful when preparing for a follow-up meeting and wanting to make sure previously raised concerns are addressed.
"Did we ever agree on a deadline for the design review?" A decision-retrieval question. The chat checks for any mention of deadlines connected to design review discussions and returns what it finds — or tells you clearly if no decision was recorded.
"Summarize what we talked about with the client last month." A time-bounded summary request. The chat looks across meetings in the relevant time range and produces a consolidated view of client-related discussion topics and outcomes.
These questions work because the AI understands intent, not just keywords. "New vendor" will match notes that say "the supplier Sarah brought up," "that company from the pitch deck," and "the logistics partner we evaluated."
Try MinuteKeep Free
MinuteKeep includes 30 minutes of free transcription on install — no account required. After recording and saving a few meetings, the AI Chat feature is immediately available to search across all of them.
Download MinuteKeep on the App Store
If you're looking at the broader landscape of iPhone meeting apps, our comparison of the best transcription apps for iPhone in 2026 covers the full field with honest pricing and feature breakdowns.
How Competitors Approach Meeting Search
MinuteKeep isn't the only tool trying to solve this problem. But the implementations vary significantly — and so do the trade-offs.
Otter.ai
Otter.ai includes an "AI Chat" feature that lets you ask questions about your transcripts. It can handle queries like "What did the team decide about the project timeline?" and retrieve relevant sections. The experience works well for recent meetings in English. Limitations surface with older archives, non-English content, and the fact that Otter's per-seat subscription model means search only works well if your whole team is paying for the same tier. The search index is tied to your account's stored transcripts, so if you've been using a different tool until now, your history doesn't transfer.
Fireflies.ai
Fireflies offers "AskFred," an AI assistant that answers questions from your meeting library. It has broader organizational capabilities — global search across team meetings, filtering by speaker, date, keywords, and action items. For teams sharing a Fireflies workspace, this is genuinely powerful. The constraint is the AI credits system: every query to AskFred consumes credits, and Pro plan users get 30 per month. Heavy use means hitting limits. For an individual user searching their own personal meeting history, the per-query cost model creates friction that accumulates.
Fathom
Fathom's "Ask Fathom" feature allows searching across your meeting library and is included in the free tier for individual users. It's well-reviewed for English-language meetings and integrates tightly with Zoom and Google Meet. The limitation is platform dependency: Fathom joins your meetings via a bot, which means it only captures meetings you've scheduled through a supported platform. Spontaneous in-person meetings, client calls on mobile, or discussions that don't go through a calendar event aren't captured.
What MinuteKeep does differently
The core difference isn't the underlying AI — RAG-based chat is increasingly common. The difference is scope and cost.
MinuteKeep records meetings where they actually happen: in person, on-the-go, across languages, without requiring a bot to join a calendar-connected call. Because all recordings happen on-device and notes are stored locally, the AI Chat searches across the full picture of your meeting history — not just the calls that ran through a specific conferencing platform. And because there are no AI credit limits or per-query charges, you can ask as many questions as you need without watching a counter.
For remote workers dealing with distributed team meetings across time zones, and for team leads who need cross-meeting intelligence rather than just per-meeting summaries, that combination matters.
Tips for Getting Better Results
AI chat is good. It's not perfect. A few practices improve the quality of what it returns.
Record and save consistently. The chat can only search meetings that exist in your library. The more complete your history, the more useful the cross-meeting search becomes. A month of consistent recording is dramatically more valuable than scattered notes from a few sessions.
Let the summary process complete. MinuteKeep generates AI summaries after transcription. The chat searches both the raw transcript and the summary. Meetings processed with summaries return richer results than transcription-only notes.
Ask specific questions when you can. "What was the decision on vendor pricing?" returns more targeted results than "what happened with vendors." Broad questions work, but specific ones work better.
Use the custom dictionary for proper nouns. If your product has a specific name, or a key stakeholder has an unusual name, add it to the dictionary. Better transcription quality means better embedding quality, which means better search results.
Ask follow-up questions. If the first answer is close but not quite what you needed, ask the chat to narrow it down. "Which meeting was that from?" or "What else was said about that decision?" are valid follow-up questions.
Don't expect the chat to catch things that weren't recorded. If a decision was made in a conversation that wasn't transcribed, it won't appear. The system surfaces what's in your notes — it can't fill in gaps.
Frequently Asked Questions
Does AI Chat search only the meeting I have open, or all meetings?
All meetings. MinuteKeep's AI Chat searches across your entire note library simultaneously. This cross-meeting capability is the point — it's designed to surface information that you wouldn't know which single meeting to look in.
Does using AI Chat cost extra or consume time credits?
No. The AI Chat feature is included with the app and uses none of your transcription time credits. Time credits only apply to recording and transcribing new audio.
What happens to my questions when I close the app?
Chat conversations are session-based and are not persisted when the app closes. Your meeting notes remain stored, and search will work the same way when you reopen the app — but the chat history from your previous session won't be there. This is intentional: your questions about sensitive meeting content don't stick around.
Can I search in languages other than English?
MinuteKeep supports 9 languages for transcription and summarization. AI Chat searches the stored notes regardless of language, so if you recorded meetings in Japanese, German, or Spanish, those notes are part of the searchable library. Response quality in non-English queries continues to improve as the underlying models improve.
How many meetings do I need before AI Chat becomes useful?
It's functional from the first meeting, but the value compounds with volume. With five or more meetings saved, you'll start to see the cross-meeting patterns — topics that recurred, decisions that evolved, action items that were revisited. At twenty or more meetings, the chat starts to behave less like a search tool and more like an institutional memory.
Key Takeaways
- Keyword search fails at meeting retrieval because meeting language is inconsistent and paraphrased across sessions. Semantic AI search understands intent, not just words.
- Retrieval-Augmented Generation (RAG) is the technology that makes AI chat search possible: your notes are indexed by meaning, and your questions are matched against that meaning rather than literal strings.
- MinuteKeep's AI Chat searches all stored meetings simultaneously, is included at no extra cost, and doesn't require a meeting bot or specific conferencing platform.
- Competitors like Otter.ai, Fireflies.ai, and Fathom all offer similar capabilities, but with trade-offs around credit limits, platform dependency, and language support.
- Search quality improves with consistent recording habits, complete summaries, specific questions, and a well-maintained custom dictionary.
- AI Chat won't recover decisions from meetings that weren't recorded. The system is only as complete as your note library.
MinuteKeep supports five different summary formats to help structure your meeting notes for different use cases — which also improves the quality of what AI Chat can retrieve from each meeting.