Fix the baseline before the work starts, because after launch every number moves and none of them is attributable on its own.
Decide the measures before the work starts, not after the launch. Fix 4 baselines and 1 date, counted from the readings you can still take today, then read the same four at a set interval. Anything you start measuring afterwards is compared against a memory, and memory moves in whichever direction you were hoping.
What to fix before anything changes
Decide what you are measuring before the work starts, because after launch every number moves and none of them will be attributable. Fix a baseline first: current search visibility, current direct traffic, current shelf conversion if you can get it, and the date the change goes live. Then read the same measures at a defined interval against that baseline rather than against the previous month. Two cautions. Give indexing and distribution time to catch up before reading anything, and separate the metrics a rebrand can move from the ones it cannot, because a brand system does not fix a price problem or a distribution gap.
Four baselines, and where each one comes from
Take these readings in the week before launch and keep the raw export, not a screenshot of a chart.
| Baseline | Where you take it | Why it moves for reasons other than the rebrand |
|---|---|---|
| Search visibility | Google Search Console, whole-property export | Seasonality, algorithm updates, competitor activity |
| Direct traffic | Analytics, direct and branded search together | Campaigns, press, offline activity |
| Shelf conversion | Retailer data if you can get it | Price changes, promotion, facings, stock |
| Launch date | Your own record, to the day | Nothing, and that is why it matters |
The fourth is the one people skip and the only one you can be certain about. Without the exact date, every later reading is an argument about when the period started.
What a rebrand can move, and what it cannot
A brand system changes how a product is recognised and understood. It does not change what it costs, where it is stocked, or how many facings it gets.
So separate the two lists before you read anything. If your distribution changed in the same quarter, that quarter cannot answer the rebrand question, and no analysis of it will. The honest move is to say so in advance rather than to attribute the result afterwards.
Why one reading is never enough
This matters more than it used to, because founders now check whether AI assistants mention their brand and treat one answer as a verdict.
A University of St. Gallen study published in April 2026 (arXiv:2604.07585) measured how much the sources behind an AI answer change when the same question is asked again on the same day. Same-day source overlap came out at 0.233 for one engine, 0.282 for another, 0.318 for a third and 0.505 for the fourth. Asking twice churns the source set about as much as waiting a day.
What that means for you is simple. A single check of “does the assistant mention us” is a sample, not a state. If you are going to measure it, ask several times across several days and record the distribution, or do not use it as evidence at all.
The interval, and the mistake inside it
Read at a defined interval against the baseline, not against the previous month. Month-on-month reading turns ordinary variation into a story, and the story usually confirms whatever you expected.
Give the change time to propagate first. New pages need to be found, distribution needs to refresh, and a reading taken while that is still happening measures the transition rather than the result.
One page to write down before you start
One page is enough, and it has five lines.
- The four baselines, each with the date of its export.
- The launch date, to the day.
- The interval you will read at.
- The list of things happening in the same period that could move the same numbers.
- A signature, before the work begins.
MarkaWorks asks clients for that page at the start of a rebrand, and the reason is self-interested as well as honest: without it, a good result is unprovable and a bad quarter is unexplainable, and both of those end up in the same conversation.
An answer you should be willing to accept
Sometimes the honest reading is “we cannot tell”. If distribution, price and the brand all changed together, that is a real answer and it is more useful than a confident one built on a number that three things could have caused.
In the MarkaWorks view a rebrand that was worth doing shows up in several measures pointing the same way over a period long enough to rule out noise. One metric moving once is not that.
