How BrandCrux measures brand visibility
Working notes from the team. Why we sample the way we sample, how the classifier decides what counts, what the composite score is actually doing under the hood. Written for operators who want the numbers behind the number.
Which bank does AI recommend in India?
We put 40 questions Indian banking customers ask to five AI assistants. ICICI Bank was named in 173 of 200 answers and IndusInd Bank in 26, and Claude still links to the banks' old .com addresses.
Read →
Who AI recommends for Maharashtrian sweets
Forty questions about Maharashtrian sweets and snacks, for buyers in India and the US, 400 answers. Chitale Bandhu was named in 105 of 200 in India and 89 of 200 in the US.
Read →
What AI tells Detroit car buyers
We put the 50 questions a Detroit buyer asks before choosing a dealership to five AI assistants, twice, and read all 500 answers. One group is the default answer; each assistant keeps its own shortlist, and several top selling stores are rarely named.
Read →
Authority Topics: how we decide what your brand should be known for
Topics, not keywords, are the unit of measurement inside BrandCrux. Here is how the taxonomy gets built, how the classifier keeps it clean, and how the per-topic composite blends seven separate signals into one score.
Read →
Why one scan does 830 readings, not 80
AI engines are not deterministic. A 40-prompt scan that looks like 160 work units is actually 830 readings. Here is the variance we measured and why five samples per cell was the number we settled on.
Read →