Can AI analyse a spreadsheet?

Yes. For a handful of rows, paste them into the prompt as plain text — no upload needed. For a large file, use an upload feature, which on most vendors works better on a paid plan. Either way, recompute one figure yourself before trusting the rest.

Why does the AI sometimes get simple maths wrong?

It predicts text rather than calculating, so a subtraction can come out plausibly wrong. Paid plans on the major vendors can run actual code for arithmetic, which is more reliable. On any plan, asking it to show each step and checking one number catches most slips.

Will the plan be personalised to me?

Only as much as you tell it. Starting point, time available, and what went wrong last time are the three details that turn a generic plan into a usable one. Then report back after week one and let it adjust.

Create a plan, then analyse a small file — 10 Things With AI

Plans and data are where AI starts to feel like an assistant rather than a writing tool. They are also where the two failure modes of this whole course meet: a plan can be generic if you do not say where you are starting from, and a data answer can be wrong in a way that looks exactly like right. This lesson shows the guard for each.

Win 7: create a plan

Say your starting point, your target, your time, and anything that has gone wrong before. A plan built for a generic beginner is easy to get and useless to follow; a plan built for *you* needs those four facts.

Try it

A four-week plan to run 5 km

Three free runs a day on the site key. Add your own key for unlimited runs.

Did it work? Week 1 should look easy — mostly walking with short runs — because you told it you gave up last time by doing too much. If week 1 has you running 20 minutes, it ignored your history.
Then: reply with what actually happened after week 1. *"I did all three sessions but my knees hurt on the third."* A plan that adjusts is worth ten that do not.

Win 8: analyse a file

You do not need to upload anything to analyse data — for a small table, paste the rows straight into the prompt. Ask a specific question with a specific unit ("in dollars and percent"), then check one number yourself. The check takes thirty seconds and tells you whether to trust the rest.

Try it

Which spending category grew the most?

Three free runs a day on the site key. Add your own key for unlimited runs.

Did it work? Check it yourself: eating out went from 180 to 340, which is +160, and 160 ÷ 180 is about 89%. If the model's numbers match, trust the rest of its answer. If they do not, do not trust any of it — a model that slips on one subtraction has slipped on the others too.
The second thing that should stand out: subscriptions jumped from 58 to 79 in August. A good answer mentions it; a lazy one only talks about the biggest number.

The thirty-second check. Do this for every data answer, every time.
Step	What you do	Why
Pick one number	The headline figure — here, +160	One check tells you whether the model was careful
Recompute it	340 − 180 on your phone	Models are fluent; they are not always careful with arithmetic
Match?	Yes → read on. No → discard the whole answer	One wrong number means the method was wrong, not just the number

Knowledge check

The model says eating out rose by $170 and 94%. Your own subtraction says 340 − 180 = 160. What do you do?

Create a plan, then analyse a small file

Win 7: create a plan

Win 8: analyse a file

Lesson FAQ

Stop overpaying for AI tools