Everyday AI tasks: summarising, extracting, rewriting and getting formula help

Most people who use AI at work use it for a small set of jobs: shortening something long, pulling details out of messy text, changing the tone of a draft, and getting help with a formula or snippet. Each has a prompt shape that works and a check that must follow.

Summarising

Say who the summary is for and what decision it supports. Example 1: "Summarise this three-page vendor proposal for a manager who will decide whether to take a call with them. Five bullets: what they offer, price model, contract length, risks, and anything they avoided stating. Quote exact figures." The last instruction matters: a summary that says "competitive pricing" when the document says "12% above last year" has lost the point.

Check: pick two bullets and find the sentence in the original that supports each. If you cannot, the bullet is not a summary, it is a guess.

Extracting into a table

Example 2: twenty customer emails about delivery problems. "From each email below, extract: customer name, order number, the problem in under ten words, and whether they ask for a refund (yes, no or unclear). Output a table with one row per email. If a field is missing, write 'not stated'; do not guess." The "do not guess" line stops the tool inventing order numbers that look plausible.

Check: count the rows against the number of emails, and spot-check three rows against the source.

Rewriting and formula help

Example 3: a blunt internal note has to go to a client. "Rewrite this for the client. Keep every fact and date unchanged. Polite, direct, no apology unless we caused the delay. Under 150 words." Listing what must not change is more important than describing the tone.

Check: read the original and the rewrite side by side and tick each fact. Rewrites quietly drop the inconvenient ones.

Example 4: "In Google Sheets, column A has dates and column C has amounts. I need the total for March 2024 only. Give the formula and explain each part in one line." A good answer uses SUMIFS with two date conditions. Then test it: put a known small set of values in, including one date on 31 March and one on 1 April, and confirm the total is what you expect.

Common mistakes and a practice habit

• Pasting customer data that includes personal details, card numbers or anything your workplace policy forbids. Strip or replace them first.
• Trusting a summary's numbers without opening the source.
• Letting the tool guess missing fields instead of flagging them.

Once a day, run one of the four tasks on something real but low-stakes, then do the matching check and note whether it caught anything. Keep a short list of prompts that worked, along with the checks that go with them.