Monday morning. The report arrives. Twelve charts, forty numbers, three tools’ worth of screenshots, and a sentence at the bottom that says “overall performance was stable”.
You read it for four minutes and know nothing you did not know on Friday. That is not a reporting problem. It is a writing problem, and machines are now perfectly capable of writing it badly at scale.
Here is the report we think a machine should write, and the rules that stop it from lying politely.
The four paragraphs
What moved. Why we think so. What we are not sure about. What we would do.
A weekly report, in order, no charts required.
- What moved: the two or three things that changed enough to matter, with the count next to each number.
- Why we think so: what moved with it, across which tools, and how confident that makes us.
- What we are not sure about: the cells that were too small, the sources that were quiet, the claim we could not check.
- What we would do: one or two drafts, staged, waiting for approval.
That is the whole thing. A founder reads it in ninety seconds and makes a decision, or decides not to. Nothing on it is decoration.
The rules that keep it honest
The report is only as good as what it refuses to say.
Every number carries its sample size. “Bookings from the newsletter: 11, up from 6.” Never “up 83%”. The percentage flatters; the count informs.
Every claim names its sources. If the report says the videos booked the calls, it says which tools it read to conclude that, and how many people the pattern held for.
Every relationship is called what it is. Moved with, not caused. A report that says “X caused Y” from observational data is guessing with a confident face, and it should say “moved with” and let you decide how much to believe it.
And every week has a “not sure” paragraph, even a good week. A report with no uncertainty in it has hidden the uncertainty somewhere, and you will find it later, at a worse time.
The percentage flatters; the count informs.
What changes when the report is honest
Meetings get shorter and decisions get braver.
When the report says what it does not know, you stop arguing about the numbers and start arguing about the decision. When it ends with a draft instead of a chart, the decision is one click. When every claim has a receipt, you can act on a small pattern this week and check it next week, instead of waiting for certainty that never comes at your size.
The best report is not the one with the most numbers. It is the one you can act on before your coffee is cold, and trust after it is.
Questions people ask next
What should a weekly marketing report contain?
Four paragraphs: what moved (with counts), why we think so (with sources and confidence), what we are not sure about, and what we would do next, staged as drafts.
Should a marketing report use percentages?
Sparingly, and never without the count. “11 bookings, up from 6” tells the truth; “up 83%” flatters a small number.
Can an AI write a marketing report honestly?
Yes, if it is built to carry sample sizes, name sources, say “moved with” instead of “caused”, and include what it does not know every week.
Throughline reads across every tool you already pay for and narrates the part that moved. Request a run to see your own line.