Run it twice and an engine may answer differently
What this rests on. Early read: one brand audited twice, two days apart, with the same 8 buyer questions on each engine.
What we looked at
We ran the same eight brand-neutral buyer questions for the same brand on two different days, and counted how often each engine named the brand.
What we saw
One engine named the brand in none of its eight answers on both days.
Another named it in 1 of 8 answers the first time and 3 of 8 the second.
A third went the other way: 3 of 8 the first time, none of 8 the second.
What it suggests
AI answers are not fixed. A single run is a snapshot of one moment, and small moves between runs can be chance. That is why every number in a Contested report says how many answers it rests on, and why we suggest judging progress on averages over several weeks, not on one run.
What this can't tell you
- This is one brand and two runs. It shows that answers can move, not how much they usually move.
- The second run also measured a slightly different set of engines, so the comparison covers three engines.
What you can try
- Before reacting to a drop or a rise, run it again and compare.
- Judge progress on averages over several weeks, not on a single run.
Get the next note by email
And tell us what to investigate next.