1 minute readReference note
How are incorrect company figures in AI answers addressed?
In brief
Incorrect company figures need the measure, date, unit, and business entity checked before a correction is proposed. Two different numbers can both be valid if they describe different reporting periods or definitions. An employee count for a subsidiary is not the same fact as a group total.
The accuracy review should trace the displayed figure to the best available original reference. Revenue, headcount, customer totals, and geographic reach can be rounded, estimated, or reported using different boundaries. A confident number in an AI answer is not made reliable by the presence of a citation that mentions the company but does not contain that number.
Where public correction is appropriate, the replacement should carry enough context to prevent the same confusion. A figure without a date can become misleading even when initially correct. An undisclosed metric should not be invented merely to fill a public gap, and private information need not be released solely to repair an answer. Severity depends on the decision affected: a numerical error in investor or procurement research can matter much more than a small harmless rounding difference. Later checks should test the exact claim and its relevant context.