01 / Define the baseline
Write the query exactly as it will be repeated
Spacing, quotation marks, spelling, and added qualifiers can produce different result sets. Save the precise query as text and resist improving it between checks. A baseline is useful because it stays still while the index changes.
Keep the query set small. One exact-name query and a few genuinely useful topic variations usually reveal more than a long list of near-duplicates that cannot be compared cleanly.
02 / Normalize the view
Separate personalization from broad visibility
Signed-in history, language, approximate region, device, and search mode can influence what appears. Choose a repeatable viewing condition and label it clearly rather than treating one screen as universal.
A private window reduces some account history but does not create a neutral laboratory. Use it as one consistent lens, then compare another lens only when the difference would change a decision.
Working principle
A baseline is a controlled observation, not a claim about what everyone sees.
03 / Capture structure
Log categories before positions
Record the result URL, publisher type, and whether the preview is current. Category changes often reveal more than a one-position move: an official page replacing an aggregator is a meaningful structural improvement.
Avoid copying personal details or inflammatory wording into the log. A neutral label such as controlled page, directory, document, or unrelated same-name result preserves the decision value without multiplying exposure.
04 / Compare deliberately
Use intervals that can reveal a trend
Check after a source change, an index update, or a planned interval. Constant refreshing produces many observations but little additional evidence and can tempt a response to ordinary volatility.
Close the loop by writing the next action beside each material change. If no plausible observation would change the action, wait instead of collecting another snapshot.