# Broadcastwell Absence Index: B2B software naming in AI answers

Observed answers on stated dates, not a ranking. This is not a judgment of product quality.

We publish observed naming and cited sources from fixed buyer questions on stated dates. September has 14 categories, 286 checked vendors, 140 target questions, five engine products, and three scheduled repeats per question and engine. Its 2,100 target attempts include 2,095 valid answers and five recorded exclusions. The separate 150 control answers are outside this benchmark's target base.

## Files and evidence boundary

- release-2026-09.jsonl: one clean row per target attempt, including failed outcomes, detected checked vendors in first-mention order, and cited registrable domains with their category class.
- release-2026-08.1.jsonl: empty, zero supplied attempt records. The original August answer text and per-engine attempt identities were not retained. An empty file means unavailable evidence, not zero observations.
- release-2026-08.1-aggregates.json: unchanged historical release evidence, including its erratum, question bases, counts, and intervals. Do not expand aggregates into fabricated attempts.
- schema.json: field schema for attempt rows.
- CITATION.cff and LICENSE: citation and reuse terms.

## Fields

Each row identifies its release, category, question, exact question text and type, engine product, repeat and capture time, outcome, detected named vendors with position, cited registrable domains and class, and whether it is a control. Null means a dimension was not recorded. September did not record location, search activation, pass dates, need labels, or paraphrase groups. Its pass field groups the retained collection only; it is not a newly inferred capture date.

Own is relative to a specific checked vendor. For category-level mixes all checked vendor domains are grouped as checked_vendor; match a vendor's canonical domain or recorded alias before applying the other source classes. Each distinct returned URL counts once per answer; this long-format file lists distinct domains, so use the source class file for URL citation bases. Prominence is within the checked-vendor order, not among unchecked vendors.

## Limits

The question bank and checked set are fixed, not a probability sample. Repeats can be dependent. Wilson intervals describe the observed calculation. These records do not measure product quality, causal effects, buyer behavior, clicks, revenue, or future answers. Failed outcomes do not become vendor absences. Historical figures are never recomputed in place.

## Cite

Broadcastwell (2026). The Absence Index, release 2026-09: observed naming of B2B software vendors in AI answers. Broadcastwell LLC. https://doi.org/10.5281/zenodo.22907695. Concept DOI for all releases: https://doi.org/10.5281/zenodo.22150321. Licence CC BY 4.0.

Historical August version DOI: https://doi.org/10.5281/zenodo.22721962.