A tidy comparison can mix measurements made under different conditions. Gemini's file-analysis guidance notes limits when handling large inputs. Treat an AI table as a draft whose cells need source checks, not as evidence that every specification was read correctly.
Compare the same object
Here is an invented example: Device A is listed as 1.2 kg without its charger; Device B is 1.4 kg including a 0.3 kg charger. Comparing the displayed weights alone gives the wrong impression about the devices themselves. Under those stated conditions, B without the charger is 1.1 kg.
Before ranking anything, add a “conditions” column. Record whether accessories are included, whether a value is typical or maximum, and which exact model variant the source describes. Leave unavailable values marked “not stated.”
Units are part of the fact
NIST distinguishes decimal GB from binary GiB. Do not drop the unit and compare the remaining numbers. Likewise, a percentage without its reference quantity is not a complete measurement.
For each decisive row, keep the manufacturer's source and the original wording beside the normalised value. If conditions differ and cannot be reconciled, say so. This method checks the fairness of a table; it does not turn the table into a purchase recommendation or prove that one device is best for every reader.