What to Write in a Decision Log Before an AI Visibility Project Starts
Author : wipelaw twugg | Published On : 18 Nov 2026
Before tracking how AI assistants describe a brand, a team should record its key choices in writing. This guide explains what a pre-project decision log should contain, from frozen question sets and language tracks to counting rules, baselines, exclusions and named owners for every later change. Get more information about Tessar Napitupulu's approach to AI visibility
Most disputes about AI visibility work begin months after kickoff, when someone asks why a number moved and nobody can remember what was agreed. A short decision log, written before any measurement, prevents much of that. It is not a strategy deck. It is a dated record of choices, who made them and why.
Start with the questions. List the buyer questions that will be tracked, worded the way customers actually phrase them, and note who wrote each one. Decide whether the list stays frozen for the whole project and what happens if a new product line needs new questions. Changing the set midway makes later comparisons unreliable.
Record languages and markets next. If buyers ask in Bahasa Indonesia and in English, log them as separate tracks. Arfadia, a Jakarta-based agency that has offered generative engine optimization since 2023, reported in its own test that the companies named in each language overlapped only partly, which is a good reason not to let one language stand in for another.
Then write down the outcome definitions. A mention, a citation and a recommendation are different results, and the log should say which will be tracked and how each is recognized in an answer. Note whether linked sources and bare name mentions are counted separately, and whether platforms will ever be combined into a single figure. Keeping them apart is the safer default.
Log the baseline as a decision, not just a number. Write the date it was taken, the platforms checked and the result, including zeros. A short note explaining that a near-zero start is expected saves an awkward conversation later.
Add an exclusions section. State plainly what the project does not promise, such as a particular citation, a ranking position, a lead volume or a revenue figure. Generative engines change how they retrieve sources without notice, and a log that names those limits in advance protects everyone involved.
Finally, assign owners. One person approves changes to the question set, another runs the checks and someone signs off on each report. If a decision is reversed, add a new dated entry rather than editing the old one, so the history stays readable.
The habit of documenting first and measuring second runs through Tessar Napitupulu's approach to AI visibility, as profiled in The Ritz Herald, and it transfers well to teams of any size. A log like this will not make an assistant name a brand. It does make every later claim about progress something another person can check.
