Skip to content

AI for office work: e-mails, minutes, spreadsheets

By the KI können editorial team8 min read

Written with AI and checked against the sources by a second AI. As of November 15, 2026. More in the notes on AI.

A lot of office work repeats itself: the same e-mails, the same minutes, the same spreadsheets. An AI can give you drafts and explain formulas. What it needs is a precise instruction, and what you need is a firm habit of checking.

In an office, a woman writes on a laptop, a colleague brings a folder; a copier and a shelf of binders stand nearby.

Key points

  • The AI writes e-mails, minutes and templates from your facts. Content and promises remain your responsibility.
  • For formulas, name the program, language, columns, rule and rounding. Try new formulas in a copy first.
  • Names, personnel files, job applications and client data don’t belong in a personal AI account.
  • Built-in assistants like Copilot or Gemini see everything you are allowed to open. Your company’s rules come first.
  • Have the AI mark gaps instead of filling them. That way you see what is still missing.

What AI is good for in the office

In the office, an AI is above all a help with phrasing, sorting and explaining. But it knows nothing about your business, your customers or where a case stands. What it needs to know goes into the instruction, the text you give it. The more precise it is, the less you have to rework.

Typical office tasks and what you check
TaskWhat the AI needs from youWhat you check
Drafting an e-mailFacts as bullet points, tone, lengthPromises, dates, amounts
Shortening an e-mailThe text and what must stay inWhether something important is missing
Sorting minutesYour bullet points with initialsDecisions, deadlines, who is responsible
Explaining or building a formulaProgram, columns, rule, roundingTwo rows by hand
Cleaning up a spreadsheetRules for the cleanupSpot checks and totals
Writing a templateOccasion and placeholdersWhether it fits several cases

In the trial lesson you put your first instruction together and get feedback. No account, in a few minutes. Start the trial lesson

Drafting and shortening e-mails

Sophie is an office administrator at an electrical wholesaler. Many of her e-mails are too long because she is still thinking while she writes. She has the AI shorten them, but with clear limits: anything that has been promised must not disappear.

Sample instruction
Shorten my e-mail below to 120 words at most.
Keep all dates, amounts, order numbers and promises word for word.
Don’t change any content; only cut repetition and filler sentences.
Leave the greeting and sign-off as they are.
Below the text, list what you cut.
E-mail:
[paste the e-mail here, replace names with placeholders first]

The list at the end is the real gain: Sophie sees at a glance whether a sentence she cut was important after all. For new e-mails, she gives the AI her bullet points and two of her own e-mails that went down well, as a model for the tone. She replaces names, addresses and numbers with placeholders first.

Check before you send

  • Do all dates, amounts and numbers match the case?
  • Is there anything in it that you cannot or may not promise?
  • Does the e-mail sound like you, and is the greeting right?
  • Have all placeholders been replaced with real details, and only in your e-mail program?

Minutes from your own bullet points

Isabel is an administrative officer in a municipal administration and takes the minutes of the weekly meeting of department heads. Her notes are brief, and sometimes a deadline is missing. So that the AI doesn’t invent a fitting one, she tells it explicitly to mark gaps.

Sample instruction
Turn my bullet points below into minutes of results.
Structure: date and participants as initials, decisions, tasks, open questions, next meeting.
Tasks as a table with the columns What, Who, By when.
Only include what is in the bullet points.
If a deadline or a responsible person is missing, write “?” in the cell.
Anything that was only mentioned and not decided goes under “open questions”.
Bullet points:
[paste the bullet points here]

Isabel writes her bullet points with initials instead of names. She leaves out sick notes and personal matters entirely; they don’t belong in any minutes that pass through an AI chat. Then she compares every decision with her notes and fills in the question marks herself.

Having spreadsheet formulas explained and built

Mark is an industrial clerk in the purchasing department of a machine builder. He has taken over a spreadsheet from his predecessor containing a long formula that nobody understands any more. Instead of deleting it, he has it explained, part by part. Then he has a new formula built.

For a formula, the AI needs more than you might think. Depending on the program and language setting, the same function is called SUMME (in German) or SUM, and its parts are separated by semicolons or commas. If such details are missing, the AI guesses, and the formula doesn’t fit.

Sample instruction
I use Excel in English.
Sheet “Orders”: column A item number, column B quantity in pieces, column C unit price in euros. Data from row 2.
Sheet “Discounts”: column A item number, column B discount in percent.
In column D of Orders I need the total price after discount, rounded to the cent.
If an item has no discount, 0 percent applies.
Give me the formula for row 2 and explain each part in one sentence.
Also tell me what happens if an item number appears twice in “Discounts”.

Mark first pastes the formula into a copy of the file. Then he calculates two rows by hand: one with a discount and one without. Only when both are right does he put it into the real spreadsheet.

Preparing data without giving away people

An AI can also clean up spreadsheets: find duplicate rows, standardize spellings, bring dates into a fixed format. For this, only upload what doesn’t concern people, such as item numbers, quantities or cost centers. If you only replace names with initials and keep the list that goes with them, it is still data about people. Such files belong only in an AI that your company has approved for them.

Staff, job applications and clients

Some offices work with especially sensitive data. Robert is an HR officer at a logistics company. He uses AI for an onboarding plan, a template with placeholders for interim job references (Zwischenzeugnis) and the outline of a training session. He never puts personnel files, sick notes or job applications into an AI chat.

With job applications there is one more thing: if an AI screens or pre-sorts applications, the EU AI Act usually treats this as a high-risk use. The strict obligations for it apply from December 2027. Whether and how this can be done is not decided by an individual but by the company, with experts.

Olivia is a tax assistant. Her clients’ data is covered by the professional secrecy of the tax firm. So she only has the AI write general material, for example a list of the documents employees should gather for their income tax return. She checks herself whether the list is complete and up to date before it goes to clients.

Templates you use again and again

What you explain every week, you write down once: as a template for an instruction that you save and paste in each time. It says who you are, what should come out, which rules apply and where the placeholders are for whatever changes.

  1. Date and version, so you know which version you are using
  2. Role and context, such as “purchasing department of a machine builder”
  3. What should come out and what for
  4. Rules for tone, length and form
  5. Placeholders in square brackets for everything that changes

Test a new template with two or three cases, including one where a detail is missing. The same template doesn’t produce the same text every time. You can share it with your team if it contains no data about people and your employer allows it.

Built-in assistants: what they may see

Many office programs have an assistant built in. In Microsoft 365 it is called Copilot, in Google Workspace Gemini. With your work account it can access everything you are allowed to open: e-mails, calendar entries, chats and shared files. Depending on the license less, but never more. ChatGPT and Claude now also have add-ins, for Excel for example; whether you may use them is up to your employer, as with any other AI.

This is a bit like giving a colleague your bunch of keys. They can get into every room you can, including one that has been left open by mistake.

Before you use a built-in assistant for real work, settle these points:

  1. Read your company’s AI rules, if there are any.
  2. Sign in with your work account, not a personal one.
  3. Report folders and files that are shared with too many people.
  4. Start with a harmless task, such as an announcement to all staff without names.

With the work account, your company’s contracts with the provider apply. Whether a personal subscription may access company files is for IT to decide. If there is a works council (Betriebsrat), it usually has a say when such programs are introduced.

How to practice this

Whether it’s an e-mail, minutes or a formula, what matters is telling the AI exactly what should come out and what it must not invent, and then checking the result carefully. In the app “KI können” you practice this in short lessons with examples from everyday office work and put the instructions to work in your own AI, including your company’s. The trial lesson needs no account. The AI finder helps you work out which AI suits your work.

Sources

Information as of November 15, 2026. Laws and offers can change; this article is not legal advice.