DingTalk (钉钉) Meeting Transcription Without Shanji's Limits

DingTalk (钉钉) is Alibaba’s enterprise super-app — 700 million registered users, more than 23 million organizations on the platform as of 2025, and the default work surface for a large slice of Chinese corporate, government, and education tenants. DingTalk Meeting (钉钉会议) is the video conferencing module embedded in that surface, and it behaves nothing like Zoom or Teams in three ways that matter for transcription: meetings are tightly tied to a DingTalk organization (not a personal account), recordings land in DingPan (钉盘) rather than a standalone cloud, and the AI summary feature — branded 闪记 (Shanji, “Flash Notes”) — is powered by Alibaba’s Tongyi Qianwen LLM and produces a structured Doc within seconds of the meeting ending.

This guide walks through the DingTalk-specific mechanics: edition entitlements across Free, Standard, Professional, and Exclusive (免费版 / 标准版 / 专业版 / 专属版); where local and DingPan recordings actually land; what the built-in real-time AI subtitles and Shanji notes include; and how to layer Atter AI on top for 98.7% accuracy and unified search across Zoom, Teams, Tencent Meeting, and DingTalk in one place.

DingTalk Edition Matrix for Recording and Transcription

Recording and AI features are gated by the organization edition the meeting is hosted under, not by the individual user. The 2026 matrix:

Two consequences. First, unlike Tencent Meeting or Zoom, even free DingTalk organizations get cloud recording — DingPan storage is shared org-wide and free tier is capped at 100 GB total, but the capability is on. Second, Shanji notes are also available on free tier, just rate-limited to three meetings a month per organization. The economic gating on DingTalk is storage volume, not feature availability.

What DingTalk Produces After the Call

A DingTalk Meeting recording bundle depends on whether the host chose local or cloud recording — and crucially, on whether AI 智能助理 was switched on during the call.

Local recording writes to the host machine:

• .mp4 — the recording, default H.264 720p, audio interleaved.
• .txt — the meeting chat log, only if 保存聊天记录 was enabled.

The default path on Windows is Documents\DingTalk\Meeting\ and on macOS is ~/Documents/钉钉会议/. Both are configurable under 设置 → 录制 → 本地录制存储位置.

Cloud (DingPan) recording lands inside the organization’s DingPan space, accessible from the DingTalk desktop client under 工作台 → 钉盘 → 我的录制 and on web at dingtalk.com. You get:

• The MP4 streamable in-browser inside DingTalk.
• An MP4 download via signed URL (valid two hours by default).
• A 闪记 Doc (if AI 智能助理 was on), containing summary, decisions, action items with @-attribution, key timestamps, and a clickable topic outline.
• Optionally a .vtt caption track if real-time subtitles were saved.

Default DingPan retention is permanent as long as the organization’s storage quota holds — there is no 30-day expiry like Zoom or 7-day expiry like Tencent Meeting Member tier. The gotcha is the inverse: when storage fills up, the admin decides which old recordings to purge, not the host. Many users discover this only when a recording vanishes after the IT team did a quarterly DingPan cleanup.

Method 1: DingTalk Real-Time AI Subtitles

DingTalk’s real-time subtitles run on Alibaba’s Tongyi-Qwen speech models and stream during the call.

1. In the meeting, click 更多 → 字幕 → 开启字幕.
2. The subtitle bar appears on every participant’s screen. Spoken language can be set to 中文（普通话）, English, 日本語, or 한국어 as of 2026.
3. Bilingual mode (Chinese + English side-by-side) is available on Standard edition and above.
4. To persist the subtitle stream as a file, the host enables 保存字幕 in the recording panel before clicking 开始录制. The .vtt is then bundled with the recording in DingPan.

A few DingTalk-specific details:

• Real-time subtitles are free even on the free edition, but saving the subtitle file requires the meeting to be recorded (any tier).
• Mandarin accuracy on headset audio sits around 89–93% — Tongyi-Qwen’s speech head is trained primarily on enterprise Mandarin corpora, including a lot of business meeting audio, so it does well on jargon-heavy calls.
• The saved .vtt is generated by a higher-quality offline pass, not the live subtitle stream, so it reads noticeably cleaner than the on-screen captions.
• DingTalk does not support Cantonese in subtitles as of 2026. For Cantonese meetings, Method 3 or Method 5 is the path.

Method 2: DingTalk Shanji AI Minutes (闪记)

闪记 is DingTalk’s LLM-based meeting summary feature, separate from real-time subtitles and arguably the most polished native AI feature across Chinese meeting platforms. It produces a structured DingDoc (钉钉文档) with:

• A 5-bullet executive summary at the top.
• Decisions, each linked to the timestamp where it was made.
• Action items with @-attribution to specific DingTalk users in the org, automatically pushed as tasks to their 待办 (To-Do) list.
• A clickable topic outline that jumps to the relevant point in the MP4.

To turn it on:

1. Host or co-host clicks AI 智能助理 → 开启闪记 during the call (or sets it as default in 个人设置).
2. After the meeting ends, the 闪记 Doc is generated within 60 seconds for a one-hour call — much faster than Tencent’s智能纪要 — and appears at 工作台 → 闪记.
3. The Doc can be shared via a DingTalk link (org-internal by default), exported to DingDoc, or forwarded as a chat card.

闪记 is bundled on Standard, Professional, and Exclusive editions. Free tier is capped at three Shanji per month per organization. The summary supports Chinese and English; English summaries are usable but visibly weaker than Mandarin output.

Method 3: Connecting Atter AI to DingTalk Cloud Recordings

For teams that want consistently higher accuracy across all DingTalk recordings — including Cantonese, code-switched Mandarin-English, dialect-heavy calls, and PSTN dial-in audio — the cleanest path is the DingPan plus Atter AI integration.

1. In Atter AI, open Integrations → DingTalk and click Authorize.
2. The OAuth flow opens at open.dingtalk.com. The user signing in must be a DingTalk org admin or be granted explicit auth by an admin — DingTalk does not allow individual-user-scoped recording access on most editions.
3. Grant the 会议录制 (meeting.recording) and 钉盘 (drive.read) scopes.
4. Exclusive edition tenants may need their IT admin to pre-approve the Atter AI app inside the org’s DingTalk admin console (管理后台 → 第三方应用 → 应用列表) before users can install.

After a DingPan recording finishes processing, Atter AI fetches the MP4 via the signed download URL the DingTalk Open Platform issues, runs transcription with diarization tuned for DingTalk’s audio profile, and posts the result to your dashboard alongside Zoom, Teams, Webex, and Tencent Meeting transcripts.

The integration handles the signed-URL refresh automatically — DingTalk’s signed URLs expire after two hours by default, which is shorter than most platforms.

Method 4: Atter AI Notetaker Bot in a DingTalk Meeting

When you can’t use cloud recording — meeting hosted by a partner org, a guest joining from outside DingTalk, or a call your admin won’t enable the integration on — invite the Atter AI bot as a participant:

1. Paste the DingTalk Meeting URL (meeting.dingtalk.com/j/<id>) into Atter AI’s Add Meeting field, or connect your DingTalk calendar and let it auto-join.
2. The bot joins as “Atter AI Notetaker.” If the host has 入会需主持人允许 on, it lands in the waiting room.
3. Audio is captured live and the transcript is ready within seconds of meeting end.

A few DingTalk-specific behaviors worth knowing:

• DingTalk Meeting allows external (non-org) guests but treats them differently — guests cannot start recording themselves, so the bot must be present from the host’s org or invited explicitly.
• If the meeting uses 加密会议 (the E2EE mode introduced in 2024 for high-security calls), no third-party notetaker can attend, by design.
• DingTalk enforces an admin policy called 允许外部成员入会: when an org has this off, the Atter AI bot cannot join meetings from that org’s hosts. Ask the admin to whitelist Atter AI under 管理后台 → 会议 → 入会安全策略.

Method 5: Uploading a Local DingTalk Meeting Recording

For ad-hoc transcription or archived calls:

1. Open the DingTalk recording folder (Documents\DingTalk\Meeting\ on Windows or ~/Documents/钉钉会议/ on macOS) and find the meeting folder named <date>_<MeetingTitle>/.
2. Upload the .mp4 to Atter AI. DingTalk’s local recording does not produce a separate audio file, so you upload the MP4 directly — Atter AI extracts the audio track on the fly.
3. Pick the meeting language. For a Mandarin-English bilingual meeting, use auto-detect so Atter AI switches models per segment.

A 60-minute DingTalk MP4 typically transcribes in 2–4 minutes.

DingTalk Transcription Gotchas

Storage quota means old recordings can vanish. DingPan doesn’t have a default retention expiry, but it has a hard storage cap (100 GB free, 1 TB on Standard, more on higher tiers). When the cap is hit, the org admin chooses which recordings to delete. If a recording matters, download the MP4 the same week.

Admin-only recording on some Exclusive deployments. Many large enterprises on Exclusive edition disable user-initiated recording entirely and route everything through IT — your host may not be able to start the recording even if they’re chairing the meeting. Confirm before the call.

External guests sometimes can’t see the 闪记 Doc. Shanji is org-scoped by default; if a partner from outside your org needs the summary, the host has to explicitly share the DingDoc externally (and the org admin must allow external doc sharing).

Cantonese is not supported in DingTalk’s subtitles. Unlike Tencent Meeting, which supports Cantonese subtitles at 75–82% accuracy, DingTalk’s real-time captions do not support Cantonese at all as of 2026. Cantonese-speaking teams should use Method 3 or Method 5.

The DingTalk app’s mobile recording is audio-only. A meeting recorded from the DingTalk mobile app produces an .m4a, not an .mp4. Atter AI accepts both, but be aware you won’t have a video track to scrub through.

Shanji has a quiet 90-minute single-meeting cap. Meetings longer than 90 minutes are still summarized, but the action-item extraction quality drops noticeably. For all-day workshops, break the recording into 60-minute chunks before uploading.

DingTalk Native vs Atter AI

How DingTalk compares to its peers: unlike Tencent Meeting, where cloud recording is gated by edition (Member, Business, Enterprise) and capped by time, DingTalk gates by storage volume and keeps recordings indefinitely until the org runs out of space. Unlike Zoom, which has a blanket 30-day cloud retention before recordings auto-delete, DingTalk hands retention authority to the org admin. And unlike Microsoft Teams, where recordings flow into OneDrive and inherit Microsoft 365 retention, DingTalk recordings live in DingPan and inherit Alibaba Cloud + org IT policy.

DingTalk Transcription FAQ

Does the free edition of DingTalk allow recording?

Yes. Unlike Zoom or Tencent Meeting, even the free DingTalk edition allows both local and DingPan cloud recording. The constraint is total storage (100 GB org-wide) and a three-per-month cap on Shanji AI minutes. Free users who need a full transcript should still consider uploading to Atter AI for 98.7% accuracy and unlimited Shanji-equivalent summaries.

Where exactly are DingTalk local recordings stored?

On Windows in Documents\DingTalk\Meeting\, on macOS in ~/Documents/钉钉会议/. The path is configurable in the desktop client under 设置 → 录制 → 本地录制存储位置. Each meeting gets its own folder named with the date and meeting title.

How long does DingTalk keep cloud recordings?

Indefinitely, until the org’s DingPan storage quota is filled. There is no automatic time-based expiry. However, when storage runs out, the org admin decides which recordings to delete, so individual hosts should not assume their recordings are safe forever. Download anything important.

Does DingTalk transcribe Cantonese accurately?

No — DingTalk’s real-time AI subtitles do not support Cantonese as of 2026. The only supported languages are Mandarin, English, Japanese, and Korean. For Cantonese transcription, upload the recording to Atter AI, which uses Cantonese-specific acoustic models with 95%+ accuracy.

Can the Atter AI notetaker join a DingTalk Meeting hosted by another organization?

Yes, as long as the host org has 允许外部成员入会 enabled. If they don’t, the bot is rejected at the entry policy layer — ask the host’s admin to whitelist Atter AI or use Method 5 (upload after the fact).

Is Shanji (闪记) the same as real-time AI subtitles?

No. Real-time subtitles are the live caption stream on screen, optionally saved as .vtt. Shanji is a separate post-meeting DingDoc with summary, decisions, action items pushed to To-Do lists, and a clickable topic outline. Subtitles and Shanji can run independently — you can have one without the other.

Can I transcribe a meeting recorded by someone else on DingTalk?

Yes — they share the MP4 with you via DingTalk chat or a DingPan share link, and you upload it to Atter AI via Method 5. Your Atter AI account does not need to be connected to their DingTalk organization.

Does Atter AI integrate with DingTalk’s To-Do list for action items?

Yes, through the DingTalk Open Platform’s task API. Once authorized (Method 3), action items extracted by Atter AI can be pushed to the assignee’s DingTalk 待办 in addition to (or instead of) the native Shanji push, so cross-platform teams using both DingTalk and Slack or Notion get a consistent task surface.