AI meeting software: the 2026 comparison

AI meeting software joins your video calls, captures audio on the phone or in the room, transcribes the conversation and writes the minutes: decisions, tasks, deadlines. Some then send them into your CRM.

The six tools compared here (Leexi, Fireflies, Otter, Fathom, Granola and Microsoft Copilot) all transcribe French and produce a summary. Three criteria genuinely separate them: where your data is stored, whether your meetings are used to train AI models, and what is covered beyond video calls (phone, in-person, CRM).

Transparency: this comparison is published by Leexi, which is one of the tools compared. We applied the same criteria to all six. All the information comes from official sources, verified on 16 September 2026, including where Leexi comes up short.

Where your data goes: the three hosting situations

Before the tables, a way to read them. The six tools fall into three situations, and that is the first question your DPO will ask.

Data stored and processed in the United States. That is the case for Otter (AWS, no documented European residency), for Fathom ("All Fathom data is stored in the United States") and for Granola, which states it plainly: European residency "isn't available yet". Fireflies stores and processes in the United States by default; a European storage option exists, with processing that stays American, and private storage is reserved for the Enterprise plan.

European data boundary of an American provider. Microsoft commits to storing and processing European customer data within the European Union (EU Data Boundary), with one exception written into its documentation: models provided by Anthropic are excluded to date.

Storage and processing in France. Leexi hosts on AWS Paris region, offers a Scaleway option (a French infrastructure provider) and runs its language model through Azure France, in a closed loop. The honest limit: AWS and Azure remain companies under US law; for a requirement of immunity to extraterritorial legislation, it is the Scaleway option that answers the need.

Data, AI training and certifications

Scope, languages, CRM and pricing

Minutes features: summary, tasks, speakers

All six produce a summary and detect tasks. The real differences are in the limits: plan quotas, quality of speaker identification, scope of the question assistant.

Free plans, applications and technical access

Worth noting on MCP servers (how your AI assistants reach your meetings): the five dedicated tools now offer one, Otter's is included from the free plan, Leexi's is read-only with OAuth.

Fireflies: the broadest of the American tools

Fireflies covers video calls, file uploads and in-person meetings through its mobile app, transcribes more than 100 languages and connects to all three CRMs from the Pro plan. Its privacy policy is precise: no use of personal data to train models, and a contractual ban on third-party providers doing so.

Two points deserve attention. Data is stored and processed in the United States by default; the European storage option leaves processing in the United States, and private storage is reserved for the Enterprise plan. And the free plan's "Unlimited transcription" comes with conditions: it requires auto-join to be enabled, applies only to accounts created after 1 September 2026, and AI summaries remain limited by the plan's credits.

Otter: the long-standing transcription tool, trained on de-identified data

Otter is one of the oldest tools in the category and its paid plan is the cheapest in this comparison ($8.33/user/month annually). French transcription is supported, with one rule to know: one language per meeting, chosen before recording.

Two elements can weigh on the decision. Data is hosted in the United States, with no documented European residency. And Otter trains its in-house models on de-identified data; its privacy policy states that transcripts used for training "may contain personal information". Its third-party AI providers, on the other hand, are not allowed to use your data to train their models.

Fathom: the most generous free plan, video calls only

Fathom offers unlimited recording and transcription on its free plan, with summaries translated into French. SOC 2 Type II and HIPAA are displayed, and third-party providers are not allowed to train their models on your data.

Worth knowing before choosing: all data is stored in the United States, with no European option. In-person capture and the mobile app are announced as "coming soon", so capture is currently limited to Zoom, Meet and Teams. On the free plan, advanced summaries stop at 5 meetings per month; unlimited applies to recording and transcription.

Granola: notes without a bot, hosting without a European option

Granola captures the computer's audio: no bot joins the meeting and no audio recording is kept. The app covers macOS, Windows, iOS and Android, and exposes a documented MCP server, usable from Claude or ChatGPT.

Two elements are worth checking with your DPO. Notes are stored in the United States and European residency is not yet available, which the documentation states explicitly. And Granola trains its models on anonymised data by default, with an opt-out in the settings; off-by-default is reserved for the Enterprise plan, which qualifies the "never used for training" promise on the Enterprise page.

Microsoft Copilot: the built-in recap, for Teams meetings

If your organisation lives in Teams, Copilot is the default option: intelligent recap, questions put to the AI during and after the meeting, mobile included, and the EU Data Boundary commitment for European customers. Prompts and responses are not used to train the foundation LLMs.

Scope is the real limit: Teams meetings only (no documented Zoom or Meet coverage), in-person through equipped and licensed Teams Rooms, and no documented sending of minutes to HubSpot or Pipedrive; Salesforce goes through the Sales agent, a separate licence. The documentation also notes that Anthropic models are excluded from the EU Data Boundary to date.

Leexi: data in France, AI not trained on your meetings, video, phone and in-person

Leexi stores and processes data in France (AWS Paris region), with a Scaleway option for organisations that require a French infrastructure provider. The language model runs through Azure France, in a closed loop, and non-training applies to all accounts, by default, with an attestation provided on request. ISO 27001 certification is held, and the DPA and the list of sub-processors are public. Scope covers video, phone and in-person meetings (iOS and Android app, recording up to 4 hours offline), and the minutes go automatically into HubSpot, Salesforce or Pipedrive. Around 10,000 companies use it across Europe.

Where Leexi comes up short: no SOC 2 report displayed, where Fireflies and Fathom display one. The default infrastructure providers (AWS, Azure) remain companies under US law: immunity to extraterritorial legislation goes through the Scaleway option, charged separately. And Leexi is not SecNumCloud qualified to date.

Which AI meeting software to choose for your need

Three needs dominate the conversations with prospects and customers. The count on Leexi's internal corpora (1,944 demonstration transcripts, 1,191 customer success calls and 1,141 support tickets, January to June 2026, anonymised) puts in-person meetings first (1,225 documents, including 473 where the customer asks for it), speaker identification second (678 documents, 310 in the customer's own words) and capturing decisions and tasks third (774 documents, 222 in the customer's own words). The table crosses these three needs with what each tool documents.

For the other situations:

Your meetings contain sensitive data and your DPO decides. Look at Leexi (storage and processing in France, non-training by default) and Copilot if the EU Data Boundary is enough for your risk analysis.

Your organisation lives in Teams and the Microsoft licences are already paid for. Copilot first: the recap is included with a Copilot or Teams Premium licence. A dedicated tool is justified if you need Zoom, Meet, phone, in-person or CRM.

You want to test on your own, free of charge. Fathom (unlimited recording, advanced AI on 5 meetings per month) or Fireflies, reading the free plan conditions.

Your meetings are mostly in English and budget comes first. Otter, at $8.33/user/month annually.

You do not want a visible bot in the meeting. Granola, knowing about the American hosting and the default training setting.

FAQ

How much does AI meeting software cost in 2026?

Entry paid plans recorded on 16 September 2026 range from $8.33/user/month (Otter Pro, annual) to $15 (Fathom Team), with Fireflies at $10, Leexi at $12 and Granola at $14. Free plans exist everywhere except at Leexi (7-day trial) and Copilot (licence on top of Microsoft 365), with limits worth reading: 300 minutes per month at Otter, 5 meetings with advanced AI per month at Fathom, notes readable for 30 days at Granola.

Which tools train their AI models on your meetings?

Otter and Fathom train their in-house models on de-identified data; Granola does so by default on anonymised data, with an opt-out, off by default only on Enterprise. Fireflies, Leexi and Microsoft Copilot commit not to. All six prohibit their third-party AI providers from training on your data.

Can meeting data stay in Europe?

With Leexi, yes by default: storage and processing in France (AWS Paris region, Scaleway option). With Copilot, yes for European customers through the EU Data Boundary, excluding Anthropic models. With Fireflies, storage can be European but processing stays in the United States. Otter, Fathom and Granola store in the United States, with no documented European option.

We already have Copilot: do we need a dedicated tool as well?

Copilot covers Teams meetings and equipped Teams Rooms. A dedicated tool is justified in four cases: your meetings also run on Zoom, Meet or the phone; you record in person without an equipped room; you want the minutes sent automatically into HubSpot or Pipedrive; or not all of your teams have a Copilot licence.

Which software covers in-person meetings?

Leexi (iOS and Android app, recording up to 4 hours offline) and Fireflies (One-Touch Record on mobile) document it. Otter records through the device microphone. Fathom announces it as "coming soon". Copilot requires an equipped and licensed Teams Room.

Is it legal to record a meeting with AI?

Yes, under three conditions set by the GDPR: a legal basis (legitimate interest internally, explicit consent with external participants), prior information of the participants, and a controlled retention period. The detail by meeting type is documented in our compliance section, meeting by meeting (business meeting, recruitment interview, management committee, sales call).

Which tool sends the minutes into the CRM automatically?

Leexi and Fireflies cover HubSpot, Salesforce and Pipedrive natively. Otter covers HubSpot and Salesforce from the Pro plan, without native Pipedrive. Fathom covers HubSpot and Salesforce, with field synchronisation from the Business plan. Granola covers HubSpot natively, the rest through Zapier.

Sources

Data verified on 16 September 2026 on the official pages of the six vendors: fireflies.ai (pricing, security, privacy, guide.fireflies.ai, docs.fireflies.ai), otter.ai (pricing, privacy-policy, privacy-security, help.otter.ai), fathom.ai (pricing, privacy, overview, help.fathom.video, developers.fathom.ai), granola.ai (pricing, security, enterprise, docs.granola.ai), learn.microsoft.com and support.microsoft.com (Copilot, intelligent recap, EU Data Boundary), leexi.ai (pricing, security, compliance and sovereignty FAQ, DPA, non-training, transcription-diarisation, task-manager, leexi-studio). Need volumes: Leexi internal corpora, January to June 2026, anonymised. Prices change: the verification date is what counts. Anything a vendor does not document publicly is stated as such, never estimated. Try Leexi for free