# Methodology and limitations

## Snapshot

All values in this repository use the July 19, 2026 snapshot cited in the [2026 Business Signal Benchmark](https://soxoa.com/research/2026-business-signal-benchmark), published July 20, 2026.

## Dataset separation

The MentionedOn, RegSentry, and Intakra totals come from separate product systems with different units of analysis and collection methods. They must not be interpreted as a joined dataset.

## MentionedOn

MentionedOn measures whether and how businesses appear in answers produced for sampled local-intent prompts. The aggregate release describes 25,220 answers organized into 1,261 category-by-city leaderboards. Model output can vary over time and between otherwise similar prompts. The sample is not a census of every AI answer or every U.S. market.

Detailed source: [State of AI Search](https://mentioned-on.com/state-of-ai-search).

## RegSentry

RegSentry loads public websites under controlled scan conditions and records detected tracking technologies that load before the scanner observes a consent choice. Detection depends on the page, region, timing, scripts, and scan environment. A finding is technical evidence from that run; it is not legal advice or proof that a law was violated.

Detailed source: [State of Website Tracking](https://regsentry.com/state-of-website-tracking).

## Intakra

Intakra retains cited public-market events such as hiring, funding, and product activity as potential buying signals. A signal indicates a time-relevant public event under the product's definitions; it does not establish that a company intends to purchase, that a deal exists, or that outreach will convert.

Detailed source: [State of Buying Signals](https://intakra.com/state-of-buying-signals).

## Reuse

When citing a value, include its snapshot date, preserve the metric definition, and link to the relevant source report or the Soxoa benchmark. Do not combine denominators across the three source datasets.