[Collaborative Docs]Tencent Docs & Feishu: AI Features You Already Have

Summary: No new software to install — AI fields in Tencent Docs smart tables, field shortcuts in Feishu Base, Feishu Minutes, and Knowledge Q&A. Get a handle on these four built-in AI features and three jobs — auto-tagging, meeting minutes, team Q&A — are all lit up within 30 minutes.

🧭 Where this fits: this article is the series' companion piece on collaborative docs, complementing the B1 Excel/WPS track — B1 covers AI in a desktop spreadsheet; this one covers AI in the online documents your whole team edits together. Zero background required; you don't need to read any other article first.

1. The Pain Point: Sound Familiar?

Mention "AI for office work," and many people's first reaction is:

  • "Doesn't that mean installing some overseas AI suite? Don't I need a paid subscription?"
  • "My company laptop is locked down — no installing software at will, and the Microsoft bundle is out of the question."
  • "Every tutorial online is about foreign tools. For the Chinese software I use every single day, I can't find a single guide."

But the Tencent Docs and Feishu you already open every day have a full set of AI built in: smart tables let you add AI fields, Feishu Base has AI field shortcuts, meetings can be transcribed and summarized automatically by Feishu Minutes, and you can ask questions directly against your team's documents.

None of this needs a new install. Most of it works on the free tier; a few pieces (the full Knowledge Q&A feature set, for example) may require a team or business plan — varies by version/plan, check the official page. The catch: nobody tells you where these features live or how to use them. This article fills that gap — no concepts to memorize, just follow along and light up the AI inside the collaborative docs you use every day, one feature at a time.

2. What You'll Walk Away With (Goal)

  1. A self-tagging collaboration table (Tencent Docs smart tables): drop customer feedback in, and the AI field — that is, a column that runs a prompt on every row automatically — classifies it, extracts keywords, and translates it into English. Colleagues open the sheet and can work by category right away;
  2. Meetings that produce their own minutes (Feishu Minutes): Feishu Minutes (妙记), the built-in meeting transcription & summary assistant — it turns meeting audio into a transcript, then writes a summary on its own. Leave it running during the meeting, and when you walk out you already have the transcript, the summary, and the action-item list. No more hand-writing minutes;
  3. A Q&A entry point over your team documents (Feishu Knowledge Q&A): Knowledge Q&A is an AI question-answering flow that first searches within a document scope you define, then composes the answer. A new hire asks "what's the reimbursement standard?" and it finds the answer inside the docs you scoped — with the source cited.

3. Before You Start

  • A Tencent Docs account you can log into (WeChat or QQ both work) and a Feishu account. The personal free tier covers most of what's in this article (subject to the current version); a company team plan unlocks more;
  • Work in a desktop browser — creating fields and writing prompts are far more comfortable at a keyboard; checking results on your phone is fine;
  • One important caveat up front: both products move their AI entry points, button labels, and free quotas around from version to version. This article teaches the approach and the step skeleton; go by what you actually see on screen. Can't find a feature? Look first in the top toolbar or the "+" new-item menu for anything labeled "AI" or "Smart";
  • 🌍 A note for readers outside mainland China: Feishu's international edition is Lark (larksuite.com), and Tencent Docs has an international entry point too. But feature lists and AI capabilities differ substantially between the China and international editions — this article follows the China editions. If your team runs on Lark, the workflows here still transfer; just expect different menu names and a different feature set;
  • General-purpose AI assistants (ChatGPT / Claude / Gemini / Grok) are supporting actors in this piece — use them to polish your prompts. The stars of the show are the platforms' built-in AI.

4. Walkthrough 1: The Tencent Docs Track — Smart Tables + AI Fields

First, one sentence to memorize: an AI field is just a column that runs a prompt on every row, automatically. You write the requirements in that column's settings; the AI executes them on each row's content — classifying, extracting, translating, all on its own — and the results land straight in the cells.

Step 1: Create a smart table

  1. Open Tencent Docs and choose "smart table" when you create a new document. The difference from an ordinary online spreadsheet: every column is a typed "field" — besides text and numbers, a field can hold options, attachments, and people. It can hold AI, too;
  2. Create a "Customer Feedback Tracker" and fill in 3 rows of test data by hand:
Feedback Submitted by
导出 PDF 时页眉总是错位，急用 ("PDF export keeps misaligning the header — urgent!") Wang Qian
The mobile app crashes when uploading photos Kevin
希望增加深色模式 ("Please add dark mode") Liu Yang
  1. Keep the data small for now — why only 3 rows? See Pitfall 1.

Step 2: Add AI fields (auto-classify / extract / translate)

  1. Click "+" at the far right of the table to create a new field, and look for an AI-related type in the field-type list (the name and location differ slightly between versions — go by your current version);
  2. Write a prompt for the field, for example:
text
Read the "Feedback" cell in this row and output a category: Bug / Feature Request / UX Improvement.
Output only one of these three options — nothing else.
  1. Confirm, and the AI runs through the rows one by one, filling the column. Any row you add later gets processed automatically;
  2. Following the same pattern, add two more columns:
    • A "Keywords" column — prompt: "Extract 2-3 keywords from the feedback in this row, separated by commas";
    • An "English translation" column — prompt: "Translate the feedback in this row into English. Output only the translation."

One detail worth noticing: nail the output format down inside the prompt (output only one of the options; separate with commas) — that's what keeps the whole column tidy. This is the same prompting method taught in A3, and we'll come back to it in Section 10.

Step 3: @ the AI for formulas and polishing

  • Want to count entries per category but can't remember the formula syntax? Find the AI assistant entry where you edit formulas (location varies by version), and say it in plain words: "count the rows where the Category column equals Bug." Let it generate the formula, give it one careful look, then confirm — don't press Enter with your eyes closed;
  • Text documents in the same team space (say, a "Feedback Handling Policy") can also use the in-document AI assistant to polish paragraphs, draft outlines, and adjust tone and form of address.

That's the Tencent Docs track done. Now over to Feishu — three more wins in a single pass.

5. Walkthrough 2: The Feishu Track — Base + Minutes + Knowledge Q&A

Step 1: Field shortcuts in Base

Feishu Base (multidimensional tables) and Tencent Docs smart tables are the same species — both are built on the "table + typed fields" logic — and their AI works the same way. After creating a Base, choose "field shortcuts" when adding a field: a library of pre-installed field capabilities that includes AI-powered shortcuts.

  • The classic play: create a "Meeting Topics Intake" table, add an AI shortcut field, and let the AI auto-triage each topic into "FYI Only / Discuss This Week / On Hold" based on its content;
  • The list of available shortcuts varies by version/plan — check the official page — but the underlying idea is exactly the same as the previous section: run the prompt once per row.

If your company is Feishu-only, read the Tencent Docs section as pure methodology: the two table AIs share one logic. Learn either one and you can find your way around the other.

Step 2: Feishu Minutes — transcribe live, or upload the recording afterwards

Minutes, remember, is the assistant that turns meeting audio into a transcript and then writes the summary for you. Two ways to use it:

  1. Live transcription: in a Feishu video meeting, the organizer switches on Minutes from the meeting interface (entry point varies by version). It transcribes as it listens; who said what appears on screen in real time;
  2. Upload transcription: for in-person meetings, or meetings run on other conferencing software, upload the audio/video file to Minutes after the meeting and let it transcribe.

If the meeting involves anything confidential, stop right there — Pitfall 2 is required reading before you go any further.

Step 3: AI summary and action-item extraction

Once transcription finishes, Minutes usually hands you three things:

  • A full transcript (with speaker labels, so you can see who said which line);
  • An AI summary (what the whole meeting covered, distilled into a few paragraphs);
  • An action-item list (who, by roughly when, doing what).

Recommended order of operations: after the meeting, skim the AI summary first to confirm the big picture wasn't misread; then go through the action-item list line by line — AI-extracted actions can miss items or merge the wrong people, and a human must pass over it before it goes into the group chat. That is the baseline discipline when AI writes your minutes. Once checked, forward it to the project channel or convert items into tasks.

Step 4: Knowledge Q&A — ask questions against your team documents

Last piece: the Q&A entry point. Feishu's smart companion / Knowledge Q&A family of features lets you define a "knowledge scope" — say, a few shared documents or a knowledge base — and then ask questions directly against it:

  1. Find the Knowledge Q&A entry (usually in the workbench or inside the AI assistant; location varies by version);
  2. Define the scope: add the expense policy, the product manual, and the new-hire guide to the knowledge scope;
  3. Test it: "What is the reimbursement standard for business-trip lodging?" — and check whether it answers only from the documents you scoped;
  4. Always switch on citation tracing: every answer should note which document and which section it came from. An answer with no cited source does not get taken at face value.

And with that, all three goal outputs from Section 2 are in hand.

6. The Prompts (One-Click Copy Version)

Every AI-field prompt from this article is collected here. Copy one, swap the column names for your own, and it just works. All of them follow the "task + output format + constraints" structure (to learn the method properly, see A3).

① Auto-classification field (works in Tencent Docs smart tables and Feishu field shortcuts alike)

text
Read the "Feedback" cell in this row and output a category: Bug / Feature Request / UX Improvement.
Output only one of these three options — nothing else.

② Keyword extraction field

text
Extract 2-3 keywords from the feedback in this row, separated by commas.
No explanations — output only the keywords.

③ English translation field

text
Translate the feedback in this row into English.
Output only the translation, with no added commentary.

(Even in an English-language workflow this one earns its keep on China platforms — a good chunk of customer feedback will arrive in Chinese.)

④ Topic triage field (Feishu "Meeting Topics Intake" table)

text
Read the topic in this row, evaluate the rules below in order, and output exactly one result:
1. The topic is purely an announcement or progress sync; nobody needs to decide anything → output: FYI Only
2. It must be discussed and decided within this cycle → output: Discuss This Week
3. A decision already exists, or it is shelved for now → output: On Hold

⑤ Knowledge Q&A test questions (run these one by one once the Q&A entry is set up)

text
What is the reimbursement standard for business-trip lodging?
What does a new hire need to get done on their first day?
How long is the product warranty?

Acceptance criteria: Q1 should cite the Expense Reimbursement Policy, Q2 the New Hire Onboarding Guide, and Q3 the Product Manual — any question that fails to point at its matching document means your knowledge scope missed one. (The annotations are for your eyes only; paste the questions themselves to the AI, nothing else.)

Tip: keep constraints like "output only..." and "no explanations" — they are the key to a tidy column. The vaguer the prompt, the more the AI field's results "drift."

7. Why It Works: Three Tricks, One Logic

  • AI fields: the table column as AI's structured output port. At its core it's "run the prompt once per row" — the column defines the output format (category / keywords / translation), and every row is one invocation. The payoff is results that are tidy, filterable, and countable — miles better than copy-pasting out of a chat box a hundred times;
  • Minutes: a two-stage pipeline of "speech-to-text + LLM summary." Stage one turns sound into text (so when it mangles a proper noun, fix it in the transcript, then have the AI regenerate the summary); stage two has the AI read the transcript and write the summary and action items. Once you see the two stages, you know which step can go wrong, and where;
  • Knowledge Q&A: restricted retrieval-augmented generation (RAG). It doesn't answer from thin air — it first "retrieves" within the document scope you drew, then "generates" an answer based on what it found. So the tighter the scope and the fuller the citation tracing, the more trustworthy the answers. When it does poorly, the scope is usually the thing that's wrong.

8. Pitfalls: 5 Real Mistakes to Avoid

  1. AI fields burn quota per row (billing differs between versions/plans) — don't run the whole table on day one. Batch-running a few hundred rows can drain your quota in one go. The right move: hand-run 3 rows first, judge the quality, tune the prompt — then unleash the full table. Quota rules vary by version/plan — check the official page; this article deliberately quotes no numbers;
  2. Clear confidential meetings with compliance before switching on Minutes. Transcription means handing the meeting's audio to the platform. Client quotes, compensation discussions, unreleased results — for meetings like these, check your company's confidentiality rules and the security red lines in A5 first. If in any doubt, don't transcribe — hand-writing the minutes is the cheaper loss;
  3. Knowledge Q&A will invent answers with a perfectly straight face. Exactly two countermeasures: constrain the knowledge scope (no free-roaming) + switch on citation tracing (check the source on every answer). Anything without a source — or whose source doesn't check out — counts as never having been said;
  4. The line between free and paid capabilities keeps moving. What's free today can land in a paid tier tomorrow; entry points get shuffled too. Don't weld a critical business process to any single AI feature, and keep your own exported backup of anything important;
  5. AI field results drift. The same column can tag inconsistently between today and tomorrow (especially when the prompt is vague). On important tables, hard-code the output standard into the prompt and spot-check a few rows on a schedule. AI tagging is a "first-pass assistant," not the "final judge."

9. FAQ

Q1: Tencent Docs or Feishu — which should I learn first? Whichever one you actually use. The two table AIs share identical logic (a prompt run once per row); learn one and the other is just a matter of different menu names.

Q2: Can I do this on my phone? Checking results and putting questions to Knowledge Q&A — no problem. But the setup actions, creating AI fields and writing prompts, really should be done on a computer.

Q3: How is an AI field different from asking an AI in a chat box? A chat box is one-off Q&A — you copy the result back into the table yourself. An AI field freezes the requirement into a column, runs it on every row automatically, and the results land in the table. When you're processing many rows, the gap is measured in orders of magnitude.

Q4: Where does this data live, and is it safe? In the cloud of the platform concerned (Tencent Docs / Feishu), with permissions following the document or knowledge base. For anything touching customer privacy or pay and compensation, apply A5's principles: redact first wherever you can, and when in doubt, don't upload.

10. Going Further

  • On to C4 (Build an AI Knowledge Base for Team SOPs): this article's Knowledge Q&A is the entry-level play. C4 teaches you to organize scattered team documents into a systematic knowledge base and take answer quality up another notch;
  • On to C5 (Low-Code + AI Approval Flow): Feishu Base's automation takes you from "AI tags it" to "AI tags it, then routes it" — building leave and expense approval flows. C5 is the full hands-on;
  • The export-format trap when moving between ecosystems: export a Tencent Docs / Feishu table to Excel as a relay and import it into the other product, and plain data survives — but views, formulas, and AI-field configurations mostly won't. Before migrating, write out a "what will be lost" list; AI fields have to be rebuilt on the new platform;
  • Prompts not landing? Go back to A3. An AI field's quality ceiling is the prompt's ceiling. A3's structured prompt method (role + task + constraints) drops straight into field settings;
  • Back to the desktop-spreadsheet track: B1's Excel/WPS expense sheet and this article are sister pieces — one desktop, one collaborative. The two tracks together are the full picture of spreadsheet AI.

11. CTA (Optional)

  • The next article goes deeper: once you've fed your team's documents to the AI, how do you make it answer accurately — and keep answering accurately over time? The answer is C4 · AI Knowledge Base. Start with this article's Knowledge Q&A, and you're already one step ahead of 90% of your colleagues.

This article is part of the "AI-Powered Office: From Beginner to Expert" series, B30 (T1/T2 Collaborative Docs / L1). Author: Cheng Xiao. Translator: Yian. Status: Draft.