Field notes
Write-ups of individual measurements.
Each one states what was measured and what the result does not mean — the second part is usually longer than the first.
For three of the four questions we ask about Thailand’s dive market, AI names no business at all.
We read 115 answers from three engines over ten days.
89 of them named no dive operator — not ours, not anyone’s.
Which question was asked decided it almost entirely.
We applied our own rule more strictly. The number went up.
Our methodology says a day with an incomplete set of engines is excluded whole.
From the day the index went live, the code excluded the incomplete rows and kept the day.
Applied as written, Thailand has one comparable day fewer and a higher score.
Quoting someone beats claiming it yourself
In the same experiment, making your own text sound more authoritative gained a little.
Adding a quotation from a credible source gained 3.6 times as much.
It scored highest of the nine tactics on the paper’s main table.
Keyword stuffing does nothing to AI search. Here is the measurement.
A KDD 2024 paper tested nine ways of editing a page to see what changes an AI answer.
Stuffing the page with the query’s keywords was the only one that failed to move the number it was meant to move.
Quoting a source moved it most.