May 2026 · Gladia Research Report
Aka Gladia’s “Magique Square” for meeting assistants.
With input from
A map of the meeting assistant market in 2026 that explores how products are differentiating, and where the category is heading. Our framework is built around four “moats” of defensibility, and our POV is informed by product testing, survey data, and conversations with the investors and founders shaping the market.
Any AI-powered platform that (magically) turns a conversation into an artifact. Notepads, notetakers, operating systems, ambient AI… We don’t really care what they call themselves, so long as voice is the input for some sort of useful output.
Primarily founders and product teams operating in (or adjacent to) this space. But honestly? Anyone trying to make sense of the market. That includes investors, enterprise buyers, everyday consumers, and anyone else who cares about how AI is reshaping how work gets done.
I’ve always believed that voice is the most underutilized data source in any organization. Not because it’s inaccessible (we’ve been recording conversations for decades), but because we’ve never had the tools to do anything meaningful with what was captured. But if every conversation could be structured and connected to the systems where work actually happens, the AI workplace assistant — one that genuinely understands what’s happening inside an organization and acts on it — stops being a concept and starts being an inevitability.
Meeting assistants are the wedge into that future.
I use them constantly and it’s remarkable how much has already been solved. Getting a bot into the room. Distilling a messy, hour-long call into three action items someone will actually follow up on. Pushing the right structured data into the right field in Salesforce, or Epic, or Linear. These are genuinely hard engineering problems to solve. But the hardest are still ahead. The teams trying to do this are making consequential bets in a market that’s moving fast. This is especially interesting because what’s unfolding is a classic David and Goliath story.
Zoom has real advantages, like distribution and data. But the startups building in this space have something the legacy players don’t: they were built AI-native from the ground up, and have the speed and flexibility to ship new features faster than any large platform can. As a fellow David — building the infrastructure that many of these tools depend on — we believe some of these products will become more than just useful software.
They’ll become indispensable like Slack, Figma, and Notion. They’ll become tools people can’t imagine doing their job without. And they’ll be powered by voice.
This report is our attempt to understand and celebrate what these builders are doing, the structural bets they’re making, the moats they’re trying to dig, and where this category, which is still very much in its earliest stages, is heading next.
Jean-Louis Quéguiner · Co-Founder & CEO, Gladia
Let’s back up. To understand where this market is today — and where it’s going — we have to trace its evolution. Looking at the key players, there are three primary product origin stories.
Origin 01
Fireflies.ai · Fathom · Granola · Fellow
Origin 02
Otter.ai · Gong · Read AI
Origin 03
Microsoft, Google, Zoom
Product origin shapes what a tool can realistically become, but all of these products (regardless of where they started) are navigating the same evolutionary arc.
At one end: a transcription tool that captures what was said.
At the other: a true AI workplace assistant that understands context, takes initiative, and works like a colleague.
The distance between those two things is where the entire market is competing right now.
Phase 1
The first wave solved a simple problem: stop making humans do the work of recording what happened in a meeting. Reactive and isolated, these tools capture, summarize, deliver. Many products are still essentially here.
Phase 2
The conversation stops being an isolated output and becomes an operational input. CRM fields get updated, tasks assigned, follow-up emails drafted. The AI proposes, the human approves. Most workflow execution stays supervised.
Phase 3A
Conversations don’t just get processed and filed — they get connected. A question today surfaces context from six months ago. A decision in one team becomes visible to another. Over time, it stops feeling like a tool and starts feeling like institutional memory.
Phase 3B
The agent is live on the call with you. It listens in real-time, detects a risk or an opportunity, and surfaces the right context, proactively. After the call, it doesn’t wait for instructions — it acts. No manual input. No handoff. Just execution, on the org’s behalf.
Final Destination
The destination is a collaborative, autonomous tool people can’t imagine working without. One that’s embedded in how teams think, decide, and act.
Each path there solves different parts of the same puzzle, because a true AI workplace assistant requires both. Memory without action is a better filing system. Action without memory is a faster bot.
The products that figure out how to combine them are the ones that will define this category.
The products that will define this category won’t just do one thing well. They’ll become the layer where work actually happens instead of just an input to another tool. They’ll be the place where decisions get made, actions get assigned, and context accumulates over time. And the key to getting there isn’t just integrations or features. It’s collaboration. A tool that one person uses is always replaceable. A tool that holds shared context across an entire team is a different proposition entirely.
If history tells us anything, only a few players will rise to the top of the horizontal enterprise market. While, yes, vertical players face a higher bar — deeper technical challenges, stricter compliance requirements, a more demanding buyer — the reward is real defensibility. Gong already proved that owning a specific function was worth more than being broadly useful, and the same pattern is playing out across industries, functions, and individual workflows too specific for any horizontal tool to serve well.
This market got its distribution through free tiers and low-friction sign-ups. That was smart. But individual users who are easy to acquire are equally easy to lose. The products that matter long-term will be bought by teams and organizations, not expensed by individuals. Tools that stay in consumer mode will be forced to compete for individual loyalty, which means running forever on a treadmill of retention and feature releases. The only way off it is to get bought (and embedded) at the organizational level.
Workforces have gone global thanks to remote work and the globalization of talent. Now, multilingual meetings are the new norm, especially for the youngest generation entering the workforce. But most tools are built and optimized for English, and performance drops big time for other languages. This isn’t a niche problem. For any tool that wants to serve modern teams, multilingual performance is table stakes.
This category of tools is well understood and adopted in Silicon Valley and in tech start-ups. But outside of these circles, awareness drops fast. Most people can’t even name a vendor beyond the one built into their meeting platform. And even where adoption does exist, use cases remain relatively simple: notes, summaries, the occasional action item. Lightweight use cases like these don’t build the habits or shared context that make a tool truly embedded. The agentic future is real and it’s being built right now, but vendors will need to invest as much in driving feature adoption as they do in shipping them.
Not every vendor is playing the same game. These four maps reveal where the market is crowded, where there’s white space, and where the most interesting strategic bets are being made.
Disclaimer: Placement reflects the primary value proposition customers buy, not every capability a product offers.
Moat #1 · Capture
The most governed approach to capture, but also the most limited. These tools record what happens inside meeting platforms. Anything that happens outside a scheduled call gets lost. Microsoft Copilot, Google Meet AI, Zoom AI Companion, Fellow AI, Notion AI Meeting Notes
The hardest quadrant to build in. These tools capture more ambiently by extending beyond meeting platforms, without sacrificing the governance and compliance controls enterprises require. DAX Copilot, Abridge, Nabla, Freed, Suki
Easy to adopt and built for the individual. These tools tend to be botless and invisible, and they operate on the user’s terms, not the organization’s. But that freedom only stretches so far because these products are built primarily for online meetings (for now). Fireflies, Claap, Fathom, BlueDot, Granola
The broadest reach of any quadrant, and the least governed. These tools (mostly on-device or portable) push furthest toward true ambient capture, following users beyond scheduled, online meetings into the rest of their day. Spoiler alert: NOT enterprise-friendly. Plaud, HeyPocket, Wispr Flow, Limitless (acq. Meta)
Strategic Insight
The two axes of this moat are in tension with each other. Broader capture creates more value, but governance is what makes that capture enterprise-safe. So far, only healthcare has been forced to solve both at once. The products that crack this combination in horizontal enterprise contexts will define what this category is ultimately able to become.
“A tool that captures 80% but misses the critical 20% isn’t 80% as valuable. It’s a fundamentally weaker product.”
To dig into what modern capture actually involves at the infrastructure level, we sat down with Amanda Zhu, Co-Founder of Recall.ai, a simple API for recording meetings that’s built for teams who don’t want to maintain their own capture infrastructure.
Moat #2 · Intelligence
These tools win through accumulation. By synthesizing across conversations, documents, and systems over time, they start to map the tribal knowledge that lives inside an organization. This is the most ambitious quadrant in the map, and the hardest to build. Tana, Sana, Read AI, Gong, Avoma, Supernormal, Glean
The only quadrant where timing is the whole point. These tools are built to surface intelligence during the conversation so users can influence outcomes in the moment. But real-time means less context to reason over, and teams must optimize for sub-200 ms latency across their entire stack, which adds significant build complexity. Winn AI, Cluely, Attention, Fireflies Live Assist, Otter Sales Agent
Reliable and easy to love, but limited in their outputs. These tools deliver clean post-meeting summaries and structured notes from a single conversation. What they don’t do is connect the dots across meetings. Each conversation is treated as its own isolated event. Otter.ai, tl;dv, MeetGeek, Microsoft Copilot, Google Meet AI, Zoom AI Companion, Notion AI, Timeless, Circleback
Where the category started. These tools prioritize capturing what was said over understanding what it meant. They’re fast and accurate…but increasingly low-utility and easy to replicate. Tactiq, Voicenotes, Notta
Strategic Insight
When insights arrive is largely a product design choice, and both real-time and post-meeting have genuine use cases. But it’s the accumulation of context and conversation history that creates real lock-in. Products that treat each meeting as an isolated event will remain easy to replace. The ones that build knowledge over time won’t.
“The most interesting company would surface information when you need it. Actually, when you don’t even know that you need it. We’re not there yet. It needs to proactively surface things. Today it’s very, very reactive.”
Read the full Q&A
Moat #3 · Integrations
Wide reach, shallow roots. These tools connect to the apps people already use, but mostly in one direction. They’re also adopted by individuals, not deployed across teams. Fireflies, Otter.ai, tl;dv, Read AI, Circleback, Fyxer, Grain, Granola, Timeless
The hardest quadrant to displace. These tools combine deep, bi-directional integrations with wide team adoption. The product becomes the place where outputs live and where teams build shared context. Hard for product teams to build, hard for buyers to replace. Gong, Microsoft Copilot, Google Meet AI, Zoom AI Companion, Fellow AI, Avoma, Notion
Built for the individual, not the organization. These tools do one thing well — capturing and exporting meeting content — but they don’t attempt to deeply embed themselves in broader workflows or connect the people around them. Jamie, Tactiq, HappyScribe, Krisp, Notta AI, Plaud, Wispr Flow, BlueDot
Narrow by design, but deeply embedded within their niche. These tools integrate tightly with the systems of record that matter most in their domain — CRM, EHR — but the data they handle is sensitive by nature. Used by many, shared by few. Steno, Abridge, DAX Copilot, Claap, Chorus, Momentum, Attention, Winn AI, Clari Copilot, Sybill
Strategic Insight
Integrations aren’t just a feature. They’re what determine whether a product stays a tool or becomes a platform. That’s true on two dimensions: how deeply a product connects to the tech a business runs on, and how many people are connected through it. Get both right, and you become something a business can’t easily remove.
“For these meeting assistant companies to become really big, they need to crack collaboration. That’s the next stage where they can really become systems of work. For that same reason, the top player will be significantly larger than number two and number three. Network effects from collaboration features are strong.”
Moat #4 · Market Strategy
Specialized tools sold to enterprise buyers within a specific vertical. The technical complexity is high, user expectations are exacting, and that’s reflected in the price tag. These products grow by going deeper. More workflows, more integrations, more of the stack owned within the niche they already dominate. Abridge, DAX Copilot, Nabla, Steno, Winn AI, Momentum, Sybill, Claap
The biggest contracts and the broadest deployments. These tools typically arrive through existing platform relationships or dominant distribution (for now). Lower price per seat than their vertical counterparts, but what they lack in depth they make up for in scale. Gong, Microsoft Copilot, Clari, Google Meet AI, Zoom AI Companion, Chorus, Glean
These tools are built for specific, smaller audiences (investors, financial advisors, recruiters), and note-taking is often just one feature of the overall product. The total addressable market is limited by definition, which raises a real strategic question: do you go deeper into the niche like a Vertical Giant, or broader like a PLG Mover? Freed, Metaview, Zocks, Saturn
Where the energy is right now. Low friction and often free to start. But not all PLG tools are heading in the same direction. Some are actively moving upmarket: expanding into team deployments, adding governance, chasing enterprise contracts. Others are settling into an App Store tier. PLG Movers (Fireflies, Otter.ai, Granola, Read AI, Fellow AI, tl;dv) to App Store Tier (Fathom, Voicenotes, Krisp, Tactiq).
Strategic Insight
Some of today’s most popular tools are adopted by individuals, but the products with the most room to grow are those that can move up: from individual users to organizational deployment. That’s true for both horizontal and vertical players, but vertical has an edge: there are more niches to own, less competition, and higher contract values up for grabs.
“Many of [these products] have captured that core behavioral loop: join a meeting, get notes, retain users. But now it’s becoming a question of who’s going to build out the rest of the features. And that’s still all to play for.”
Read the full Q&A
The decision to go vertical is one of the most consequential a founder can make. To understand what that choice actually looks like in practice, we spoke with Delphine Groll, Co-Founder and COO of Nabla — an ambient AI assistant purpose-built for clinical workflows.
We couldn’t map this market without hearing from the people actually using these tools. What do they prioritize? How much do they trust what gets produced? Are they getting value beyond basic note-taking? We surveyed 2,100 meeting assistant users across the US, UK, France, Germany, Spain, and Italy to find out. Here’s what we learned.
The market is flooded with new entrants, but brand awareness remains a real challenge. Case in point: Only 1 in 7 respondents had heard of specialized new entrants like Granola and Fathom. Fewer still have actually used them. When you compare that to platform players like Teams, used by more than 50%, it’s clear that most people simply use the note-taker that’s built into their default work suite.
The vast majority of users rely on their meeting assistant for the most basic use case: summaries/notes. Fewer than half use on-call prompts or workflow triggers, the features that start to make these tools genuinely embedded in how work gets done. That gap narrows with age, though: younger workers are meaningfully more likely to use the “advanced” stuff.
97 % Trust Notes
For AI meeting assistants to fulfil their potential, users need to trust what they produce. Overwhelmingly, they do. 97% of respondents trust AI-generated meeting notes to at least some extent. 10% trust them completely, while less than 1% say they don’t trust them at all.
Despite an influx of free, consumer-grade tools marketed directly to individuals, only 1 in 3 people chose their meeting assistant themselves. The majority arrived at their tool through an employer, IT team, or a bundle, meaning the real buying decision is still overwhelmingly an organizational one.
Accuracy tops the list of what users want, which isn’t a surprise. What is a surprise? Data privacy ranks third, ahead of speed, integrations, and multi-speaker transcription. In a category where the product is always in the room, users have clearly decided that trust in the tool extends beyond the output to the infrastructure behind it.
Accurate summaries — 39%
Accuracy of key details — 36%
Strong data privacy — 33%
Fast output — 29%
Speaker identification — 26%
Integrations — 26%
Multilingual transcription — 22%
Industry/company terminology — 22%
Action item extraction — 19%
Botless assistants — 16%
Bot fatigue has spurred a new generation of invisible capture tools. And if a true ambient workplace assistant is the end state, local capture without a visible bot is likely how we get there. But appetite in our dataset appears limited: only 1 in 6 users cite botless capability as a top priority today.
The oldest respondents are twice as likely to cite accurate summaries as a top priority compared to the youngest. Younger workers are also less concerned about privacy. So, what do they care about? Accurate transcription in multiple languages. Our take: this is unsurprising for a cohort that entered the workforce during remote work, collaborates globally by default, and has never known a world without ambient data collection.
| Feature | 18–24 | 55–64 |
|---|---|---|
| 1. Accurate summaries of discussions | 26% | 43% |
| 2. Botless assistants | 24% | 12% |
| 3. Multilingual transcription | 30% | 24% |
On the whole, users are satisfied with product performance today. But there are some gaps that have big implications. Domain terminology and non-English transcription score lowest, and in both cases, the stakes are high. For vertical tools, accurate domain language isn’t a nice-to-have, it’s the entire value proposition.
48% participate in multilingual meetings frequently or at every meeting
In a category where a new tool launches every week, loyalty is low. 59% say they’ll switch in the next 12 months. For vendors, the imperative is clear: get embedded, get integrated, become indispensable.
This report was informed by a survey of 2,100 meeting assistant users across the US, UK, France, Germany, Italy, and Spain via OnePoll and interviews with investors and founders.
Our research started with 100+ vendors, selected based on market relevance and product differentiation across the category.
Once the axes and criteria for each moat were established, vendors were shortlisted to ~50 against rubrics defined per moat. Placements are based on product testing and publicly available information gathered between February and April 2026. Not every vendor appears in every moat; inclusion reflects standout positioning rather than exhaustive coverage.
We’re not analysts. Placements reflect a point-in-time view based solely on what was publicly available. There were no vendor briefings, and this space moves fast. That means some of this will date quickly.
