Read AI records, transcribes and scores your meetings, then makes those transcripts searchable alongside your email, chat and documents so you can ask a question instead of scrolling through notes.
Used this tool? Rate it
Used this tool? Rate it
Read.ai is an AI meeting assistant that joins your video calls, transcribes what is said, and turns each conversation into a structured report containing a summary, key discussion points, action items, and a full transcript. That much it shares with a crowded field of note-taking bots. What separates it is the second half of the product: those transcripts do not sit in an archive, they become one more searchable source alongside your email, your chat history, your CRM records and your documents, all queryable from a single prompt.
The company describes the product on its homepage as "Your AI Meeting Agent," with the accompanying line: "Read is your AI copilot—transforming meetings, emails, and messages into summaries, insights, and instant answers on every device, wherever you work." That sentence is a fair map of the product's ambition. Read AI did begin as a meeting tool, but its own positioning statement now claims something larger — the company calls itself "the independent system of record for productivity AI where people attend fewer meetings, send less emails, read fewer messages, and decisions are as simple as swiping right."
The company behind it is Read AI, Inc., founded in 2021 and headquartered in Seattle, Washington. Its three co-founders came from the same team: David Shim, the CEO, previously ran Foursquare as chief executive and before that founded and led Placed, which Snapchat acquired in 2017; Rob Williams, the CTO, was Senior Director of Engineering at both Foursquare and Placed; and Elliott Waldron, VP of Data Science, held the equivalent role at both companies. That shared background in location analytics is worth noting because it explains a design choice that surprises people who expect a pure transcription tool: Read AI is unusually interested in measuring behaviour, not just capturing words.
The funding history tracks the company's expansion beyond meetings. TechCrunch reported in October 2024 that Read AI closed a $50 million Series B led by Smash Capital, bringing total funding to $71 million including a $21 million Series A raised earlier the same year. In that same article, co-founder David Shim said the company had "doubled our sign-ups, active users, and [monthly recurring revenue]" in under six months. SiliconANGLE's coverage of the same round cited more than 100,000 new users added since the prior round, an 81% retention rate among them, and a 720% increase in active users over twelve months — all figures the publication explicitly attributed to the company rather than to independent measurement.
More recent third-party reporting gives a sense of current scale. TechCrunch reported in February 2026 that Read AI had "over 5 million monthly active users" and was seeing "50,000 sign-ups every day." Those numbers, too, originate with the company, but they come via a named reporter at a publication that covers the category closely, which makes them more useful than a marketing page.
One thing to understand before going further: Read AI is a product with genuinely divided opinion, and the division is not random. Its iOS app holds a 4.88 out of 5 rating across 853 ratings on the App Store. Its Trustpilot profile sits at 1.3 out of 5 across 105 reviews. Both numbers are real and both were verified directly for this page. The section on limitations explains why they diverge so sharply, because that explanation is probably the single most useful thing this page can tell you about the product.
The meeting report is the unit of output. Each one contains five components named on the product page: a Meeting Summary covering key points and decisions, Key Discussion Points that let you jump between sections of the conversation, Action Items generated automatically, a full Transcription, and Playback & Highlights — a rewatch capability with roughly a two-minute highlight reel, which is gated to the Enterprise tier and above.
Reports arrive in more than one place. You can open them in the Meeting Report dashboard, receive them by email once a meeting ends, or push them into connected systems including Slack, Zapier, webhooks, HubSpot, Salesforce, Atlassian, Notion and Asana.
This is where Read AI departs from most competitors. Alongside the transcript, the product computes participant-level measurements, and the product page names them explicitly: talk time, bias, and charisma. It also flags "Audience reactions & significant statements" to mark moments of high engagement, and tracks sentiment across speakers.
Treat these metrics for what they are. Talk time is a straightforward measurement. Bias and charisma are inferred scores produced by a model, and the company publishes no methodology, no validation study, and no accuracy figure for any of them. As a prompt for self-reflection — did I dominate that call, did the room disengage twenty minutes in — they can be genuinely useful. As an input to a performance review, they are not defensible, and no one should treat them that way.
Ask Read is the enterprise-search layer, and it is the feature that most justifies the product's broader positioning. The company describes it as helping you "find what you know and discover what you don't," returning context-aware answers with citations and links back to sources.
Its reach is wide. Ask Read indexes meeting transcripts, email, chat, documents and knowledge bases, and cloud storage, connecting to Zoom, Google Meet and Microsoft Teams for meetings; Gmail and Outlook for email; Slack and Microsoft Teams for chat; Notion and Confluence for knowledge; Google Drive and Microsoft OneDrive for storage; Salesforce and HubSpot for CRM; and ChatGPT and Claude as LLM endpoints.
Crucially, it respects existing permissions rather than creating a new access surface. The company states: "Only data that is accessible to you via the apps you connect will be searchable, and nothing is shared automatically with others." Unlimited searches are included on both free and paid plans, which is an unusually generous boundary — the free tier restricts how many meetings you can transcribe, not how much you can search.
Read AI's most ambitious feature is a digital twin named Ada, which operates through email. The product page lists three core actions: Ada Answers pulls insights from your knowledge base, Ada Schedules books meetings without back-and-forth negotiation, and Ada Covers For You helps you catch up after time away. Around those sit post-meeting scheduling, calendar conflict resolution, intelligent CRM updates and scheduled updates.
TechCrunch's coverage of the launch fills in the mechanics: users start by emailing ada@read.ai with "Get me started," after which Ada can reply with availability, handle rescheduling negotiations, and answer questions using "a company's knowledge base, topics discussed in your prior meetings, and public internet searches."
The company is emphatic about keeping a human in the loop. Ada "only takes actions with your explicit consent and never shares information without your approval," and users remain in control by "reviewing drafts before anything is sent." That is the right design for an agent that speaks on your behalf, and it is worth verifying the setting holds in your own account before letting it near an external thread.
The homepage names a wider module list than most users will touch: Meeting Reports, Assistant, Real-Time Meeting Notes, Playback, Digital Twin, Workspaces, Ask Read, Email Summaries, Meeting Assistant, Speaker Coach, For You, File Uploads, Recommendations and Inbox Insights. Real-time notes surface during the call rather than after it; File Uploads lets you feed documents into the same index that Ask Read queries, metered by monthly upload credits that vary by plan.
Catching up on a meeting you missed. The most common and most defensible use. Rather than reading a full transcript, you open the summary, scan the action items, and jump to the key discussion point that concerns you. Ada extends this to returning from leave, where the volume of missed context is the actual problem.
Sales and customer-facing teams. With Salesforce and HubSpot connected, meeting outcomes flow into CRM records without someone typing them up afterwards. The intelligent CRM update capability targets exactly the gap where deal notes get written three days late or not at all.
Finding a decision nobody documented. This is Ask Read's strongest case. The answer to "what did we decide about the pricing change" often exists — in a call from six weeks ago, a Slack thread, and a half-finished Notion doc — but no one can find it. Searching all three at once with citations back to the source solves a problem that transcription alone does not.
Improving how you run meetings. The speaker coaching output suits individuals who want to see their own talk-time ratio and engagement patterns. Used privately and voluntarily, it is a reasonable feedback loop.
Cross-language teams. The iOS app is localised into 19 languages and the free plan advertises support for 20+ languages, which makes the product viable for distributed teams that do not work in English.
Research and interview work. Anyone who conducts repeated conversations — user researchers, recruiters, journalists — accumulates a corpus that becomes valuable only when searchable. File uploads plus meeting transcripts in one index serves that pattern well.
Two commands matter, and anyone who attends meetings — not only Read AI customers — should know them. Typing "Read stop" in the meeting chat removes Read and generates a report covering the conversation up to that point. Typing "opt out" removes Read and deletes all meeting data. The privacy policy confirms that any participant, host or not, can use the opt-out command: "you (as a host or meeting participant) can remove Read from the meeting by typing 'opt out' into the meeting chat."
Start in manual join mode and stay there for a month. The single most common complaint about this product is not that it works badly but that it shows up where it was not wanted. Manual mode eliminates that failure entirely while you learn the tool.
Audit the meetings it attends weekly. Recurring internal standups are fine. One-to-ones about compensation, legal calls, therapy-adjacent conversations, and interviews with external candidates are not. The bot does not know the difference; you do.
Verify action items against your own memory before forwarding. Automatic extraction is good, not perfect, and the failure mode is silent — an action item attributed to the wrong person reads exactly as plausible as a correct one.
Treat charisma and bias scores as private. They are model-inferred with no published methodology. Sharing them across a team invites comparison that the underlying measurement cannot support.
Connect Ask Read sources incrementally. Add one, test the quality of retrieval, then add another. Connecting nine sources at once makes it impossible to tell which is producing bad answers.
Use the citations, always. Ask Read returns links to sources. An answer without a followed citation is a claim, not a fact, and the cost of checking is one click.
Check your plan against the free tier limit first. Five meeting transcripts per month is genuinely usable for a light meeting load. If you have four calls a week, you will hit the ceiling in the first week and should plan for that rather than discovering it mid-month.
Set data retention deliberately on Enterprise+. Custom data retention is available at that tier. Configuring it is a five-minute task that materially reduces how much sensitive conversation the platform holds two years from now.
A good fit for: individual professionals with heavy internal meeting loads; sales teams that live in HubSpot or Salesforce and lose notes between calls; distributed teams working across languages; managers who want a private read on their own meeting habits; and organisations whose real pain is not transcription but retrieval — the ones where decisions exist somewhere and nobody can find them.
A poor fit for: anyone whose meetings are predominantly with external parties who have not agreed to recording; teams in two-party-consent jurisdictions without a clear consent process, since the terms of service place that obligation on you rather than on Read AI; regulated environments needing HIPAA compliance, SAML or SCIM at a modest price point, since those sit behind the most expensive tier; and anyone who needs a documented, auditable transcription accuracy figure, because the public product pages do not provide one.
Requires a deliberate decision from: IT and security teams. This product asks for calendar access, mail access, chat access, document access and CRM access. Each connection is individually reasonable and the permission model is sound — search respects the access you already have. The aggregate, though, is a single vendor holding a remarkably complete picture of an organisation's internal communication, retained for up to two years. That is a decision to make consciously rather than one to arrive at through a series of individual approvals.
Read AI distributes through four client channels linked directly from its homepage: a Windows desktop installer, an Android app on Google Play, an iOS app on the App Store, and a Chrome extension on the Chrome Web Store. Dedicated landing pages exist for Zoom, Google and Microsoft, reflecting the three meeting platforms it targets.
The iOS app, verified through Apple's official lookup API, is published by Read AI, Inc. under the name "Read AI: Transcripts and Notes." It is free to download, rated 4.88 out of 5 from 853 ratings, carries a 4+ content rating, and at version 2.11.0 released on 5 August 2026 requires iOS 18.0 or later. It is localised into 19 languages: Catalan, Danish, Dutch, English, Finnish, French, German, Hindi, Italian, Japanese, Korean, Norwegian Bokmål, Polish, Portuguese, Russian, Simplified Chinese, Spanish, Swedish and Ukrainian.
Beyond first-party clients, Read AI publishes an MCP Server and an API Reference in its support centre, and lists an entry in the ChatGPT app directory — meaning its data can be reached from external AI clients rather than only through its own interface. The support-centre documentation was not directly retrievable during research, so this page describes the existence of those interfaces without asserting anything about endpoints, authentication or rate limits.
The website itself is published in ten interface languages: English, Spanish, Italian, Portuguese, French, Japanese, Hindi, Russian, German and Chinese.
Four tiers are listed on the official pricing page. Annual billing carries a 25% discount against month-to-month across all paid tiers.
Free — $0. No credit card required. Limited to 5 meeting transcripts per month, which is the binding constraint. Includes unlimited enterprise search, summaries for meetings, email and messaging, the personalised meeting coach, basic integrations, topic readouts, 20+ languages, and the desktop and mobile apps.
Pro — $15/month annually, $19.75/month monthly. Unlimited transcripts, 100 file upload credits per month, priority report processing, premium integrations including Notion, Salesforce, HubSpot, Jira, Confluence, Zapier and webhooks, plus unlimited storage and workspace access.
Enterprise — $22.50/month annually, $29.75/month monthly. Everything in Pro plus audio and video playback, video highlights, 200 file upload credits per month, and premium support. Requires 5 or more licences, so it is not available to individuals.
Enterprise+ — $29.75/month annually, $39.75/month monthly. Everything in Enterprise plus 300 file upload credits per month, HIPAA compliance, SAML and SCIM, domain capture, custom data retention and workspace onboarding.
Two points deserve emphasis. First, the free plan's generosity is asymmetric: search is unlimited while transcription is capped at five meetings, which tells you where the company sees its cost and where it sees its hook. Second, the terms of service state plainly: "Read AI does not offer refunds for Services." Combined with annual billing being the discounted default, that makes the monthly plan the sensible way to trial a paid tier despite the higher headline rate.
Otter.ai is the incumbent in standalone meeting transcription with a long track record and a large user base; it is stronger as a pure transcription and note product and weaker on cross-source search.
Fireflies.ai competes directly on meeting capture with deep CRM integration, and is frequently evaluated alongside Read AI by sales organisations.
Fathom focuses on free, fast meeting summaries with a notably simple product surface, and appeals to users who find broader platforms overbuilt.
Granola, named by TechCrunch in its coverage of this category, takes a different architectural approach by augmenting the user's own notes rather than deploying a bot into the call — which sidesteps the consent friction that dominates complaints about bot-based tools.
Microsoft Teams Premium and Google Gemini for Workspace deserve consideration for organisations already committed to one of those suites, since intelligent recap arrives without adding a vendor or a third-party bot to the call.
tl;dv targets similar ground with a free tier and CRM integrations.
The competitive context matters here. TechCrunch's 2024 reporting quoted investor Brad Twohig observing that "services that transcribe meetings are becoming a commodity" — an unusually candid assessment from someone who had just invested in the round. If transcription is the only thing you need, price and consent model should drive your choice more than feature lists. Read AI's differentiation rests on cross-source search and the digital twin, not on the transcript itself.
The rating divergence is the headline, and it is explicable. Read AI's iOS app is rated 4.88 out of 5 across 853 ratings. Its Trustpilot profile is rated 1.3 out of 5 across 105 reviews. Both were verified directly. These populations are different in a way that explains almost everything: App Store ratings come from people who chose to install the app, while Trustpilot reviews come substantially from people who encountered Read AI's bot in someone else's meeting and went looking for somewhere to complain. Note also that the Trustpilot profile is unclaimed, and the platform states the company "hasn't invited their customers, so reviews may not be representative." That caveat cuts both ways: the sample is not representative of the customer base, but it is also free of the positive selection bias that vendor-solicited reviews introduce.
The dominant complaint is persistence, not quality. Recurring Trustpilot themes concern difficulty getting rid of the product rather than poor transcription. One reviewer in July 2026 wrote: "I have deleted my account more than three times without ever even activating the account. I have removed the integration from my google workspace and ye…" Another in June 2026 called it "Truly obnoxious, borderline spam/malware, probably GDPR non-compliant" — a user's opinion, not a regulatory finding, and it should be read as such.
The company's official position directly contradicts these accounts, and both deserve to be stated. Read AI's account and privacy centre says: "Read will never join a meeting on its own. It only joins when someone with a Read account invites it." The company attributes the "it's still in my meetings" experience to another participant using Read. Both things can be true simultaneously — the bot is invited by someone, just not by the person objecting to it — and that is precisely the design problem. Multiple universities and public bodies, including Cornell and UW-Madison, have published internal guidance to staff about AI notetaker bots auto-joining meetings via calendar integrations, which suggests the friction is structural to this product category rather than unique to one vendor.
Non-users have no independent way to block the bot. This is the sharpest gap. The privacy centre advises checking who invited Read and contacting that person; it offers no mechanism for someone without a Read account to pre-emptively prevent the bot from attending their meetings. The per-meeting "opt out" command works, but it is reactive — you must already be in the call, notice the bot, and know the command.
Recording-consent responsibility falls on you. The terms of service do not state who must obtain legal consent to record. Instead they require that "you provide end users with notice of our Privacy Policy, and you agree to maintain and provide a process to facilitate the end users' exercise of their rights under applicable privacy law." In two-party-consent jurisdictions this means the compliance obligation sits with the deploying organisation, not with Read AI.
Liability is capped at $100. The terms state: "The total liability of Read AI and the other Read AI Parties for any claim arising out of or relating to these Terms or our Services, regardless of the form of the action, is limited to $100 USD." For a product that holds recordings of confidential business conversations, that ceiling is worth reading twice before deployment.
You indemnify them. Users must "indemnify, defend, and hold harmless Read AI and our officers, directors, agents, partners, and employees" against losses arising from their use of the services, their content, or their breach of the terms.
Your data trains their models, with one carve-out. The privacy policy describes "using information to train and improve our models within our Services." A specific exception exists for Google Workspace: "We do not, and do not permit third party AI tools to, use user data collected via Google Workspace APIs to develop, improve, or train generalized/non-personalized AI and/or ML models." Note the scope precisely — the carve-out covers data obtained through Google Workspace APIs and generalized model training, not all data and not all training.
Retention runs up to two years. Audio and video, and information derived from it, is stored "in no case for longer than 2 years." Custom retention control is an Enterprise+ feature.
No published accuracy figure. Neither the transcription page nor the meeting reports page states any transcription accuracy percentage, and no export format list is published. For a product whose core function is producing an accurate record, the absence of a verifiable quality metric is a real gap.
Report deletion is restricted to the owner. Only the report owner can delete a report — so if a colleague recorded a meeting you attended, deleting that record is not within your control.
Compliance features sit behind the top tier. HIPAA compliance, SAML and SCIM are Enterprise+ only, which means regulated organisations do not get a mid-priced option.
No refunds. Stated without qualification in the terms of service.
Ada's channel coverage was limited at launch. TechCrunch reported the digital twin was email-only at release, with Slack and Teams support planned rather than shipped.
Age requirement. Users must be at least 18 years old.
Yes, there is a genuinely usable free plan at $0 with no credit card required, but it is capped at 5 meeting transcripts per month. Notably, enterprise search is unlimited even on the free tier, as are summaries for meetings, email and messaging, the meeting coach, basic integrations, 20+ languages and the desktop and mobile apps. If you attend more than about one meeting a week that you want transcribed, you will hit the transcript ceiling.
Four tiers, with annual billing discounted 25% against monthly. Pro is $15/month annually or $19.75 monthly. Enterprise is $22.50/month annually or $29.75 monthly, and requires at least 5 licences. Enterprise+ is $29.75/month annually or $39.75 monthly. Because the terms of service state that Read AI does not offer refunds, trialling on monthly billing before committing annually is the prudent path.
The company's official position is no: "Read will never join a meeting on its own. It only joins when someone with a Read account invites it." In practice, many complaints come from people who never signed up and still found the bot in their calls — which happens when another participant has calendar auto-join enabled. So the bot is always invited by someone, but not necessarily by the person objecting to its presence. If you want to eliminate the risk entirely on your side, use manual join mode.
Type "Read stop" in the meeting chat to remove it and generate a report up to that point, or type "opt out" to remove it and delete all meeting data. The privacy policy confirms any participant can do this, not just the host. There is no documented way for a non-user to block the bot from meetings in advance.
Partly. The privacy policy describes "using information to train and improve our models within our Services." There is one explicit carve-out: data collected via Google Workspace APIs is not used to develop, improve or train generalized or non-personalized AI/ML models, and third-party AI tools are not permitted to do so either. Read that scope carefully — it is narrower than a blanket no.
Audio and video information, including anything derived from it, is stored for no longer than 2 years. Custom data retention settings are available only on the Enterprise+ plan.
Effectively, you are. The terms of service do not assign recording-consent responsibility explicitly, but they require you to give end users notice of Read AI's privacy policy and to maintain a process for those users to exercise their privacy rights. In jurisdictions requiring all-party consent, that obligation rests with the organisation deploying the tool.
Meeting capture works with Zoom, Google Meet and Microsoft Teams. Clients include a Windows desktop app, iOS and Android apps, and a Chrome extension. The free plan advertises 20+ languages, and the iOS app is localised into 19 languages including Japanese, Korean, Simplified Chinese, Spanish, French, German, Russian, Portuguese, Italian, Hindi and Ukrainian. The website itself runs in ten interface languages.
Ada is Read AI's email-based agent. It answers questions from your knowledge base, books meetings and handles rescheduling, updates CRM records, and helps you catch up after time away. Users enable it by emailing ada@read.ai with "Get me started." The company states Ada only acts with explicit consent and lets you review drafts before anything is sent. At launch it was email-only, with Slack and Teams support planned.
Because the two populations are different. The iOS app is rated 4.88 out of 5 from 853 ratings — people who chose to install it. Trustpilot sits at 1.3 out of 5 from 105 reviews — a sample weighted toward people who met the bot in someone else's meeting. The Trustpilot profile is unclaimed and the company has never solicited reviews there, so the sample is not representative of paying customers, but it is also not inflated by vendor-invited feedback. Read both numbers as measuring different questions: the App Store rating tells you whether users like the tool, and the Trustpilot rating tells you how the tool is experienced by people who did not choose it.
