In short: The record is a spreadsheet with one row per run, and its value comes entirely from nothing about it changing between quarters.
Nine columns cover it: date, market, question, platform, run number, whether you were named, the sites that were, the position of your mention if there was one, and who ran it.
A dashboard is not this. A dashboard shows you a number now; a sheet lets somebody in six months check what was asked and what came back, which is the only thing that makes a trend defensible.
Why A Sheet Rather Than A Tool
Tools change their method without telling you, and a trend line that crosses a methodology change is not a trend.
The deeper problem is that most tools do not record the answer, only their score for it. When somebody asks in six months why the number moved, a sheet can show the actual answers from both dates and a dashboard can only show two numbers and a slope.
Use a tool for what it is good at, which is running many prompts quickly. Then record the answers yourself. The thing worth owning is the record, not the automation, and a record you can hand to somebody who does not have your login is worth more than one you cannot.
The Nine Columns
Rules That Keep It Comparable
Four, and each one exists because breaking it silently invalidates the comparison.
Fresh session every run. History changes the answer, so a second question asked in the same window is measuring the first question's effect as well as its own.
Same person, or record who. Accounts carry memory and settings. Two people running the same set produce two different sheets, and neither is wrong.
Never edit the question text. Improving a question's wording between quarters means the trend spans two different questions. Add a new row instead and start its history from zero.
Record what came back, not your reading of it. Paste the sites named. A column that says "we did well here" is a memory, and memories are what the sheet exists to replace.
Cadence
Quarterly for the full set, with the noise floor re-measured on the same day.
Monthly is tempting and usually wasteful: the slowest of the four steps moves in quarters, so eleven of twelve monthly readings will be indistinguishable from the floor and the twelfth will be read as a result when it is not. If a stakeholder needs monthly contact, report the work done rather than a re-measurement.
The exception: re-measure immediately after a platform announces a model release or an index change, and mark that row. It is the cheapest way to build the record that separates their change from yours, which is the subject of the next lesson.
Where It Lives
Somewhere the next person can find it without asking you: a shared drive, not a local file, and not inside a tool's account. Include a tab that states the method in six lines, because a sheet whose method lives in somebody's head stops being comparable the moment they leave.
Key Takeaways
- The record is a sheet, and its value comes from nothing about it changing between quarters.
- Nine columns: date, market, question, platform, run, named, the sites named, the position of your mention, and who ran it.
- Record the answer rather than a score. A dashboard cannot show you what was said six months ago.
- Fresh session per run, never edit a question's wording, and note who ran it, because sessions carry history.
- Quarterly for the full set, plus an immediate marked re-measurement after a model release.
Check yourself
Before you move on
Not scored, not recorded, and not part of the certificate. Both answers are settled by a sentence in this lesson, and the reasoning appears whichever option you pick.
- 01
Why record the sites named in each answer rather than only whether you appeared?
- 02
A question's wording could be clearer. What do you do between quarters?