The Absence Index
Method
The measured engines and configuration are stated separately for each release.
Engines by release
- 2026-08.1: ChatGPT, Claude, Perplexity, Google AI Overviews. 10 fixed questions per category; 3 repeats per question and engine pair. Method: a separate version was not supplied. DOI 10.5281/zenodo.22721962.
- 2026-09.1: ChatGPT, Claude, Perplexity, Google AI Overviews, Google AI Mode. 10 fixed questions per category; 3 repeats per question and engine pair. Method: v1.1. DOI 10.5281/zenodo.22907695.
For method v1.1, verbatim answers are retained privately. Ten target questions determine the scored base; controls were run for CRM and construction project management only for release 2026-09, and remain outside its scored base. The release pages state answer capture dates, exclusions, checked vendors and the method version.
August 2026 historical method
What the release measures
Historical release 2026-08, corrected version 2026-08.1. Answer text was not retained. No answer quotations are published.
Plain word vendor names, such as Close, Copper or Notion, count only with a qualifier or the vendor domain among the cited URLs. Historical answer text cannot be independently checked against this rule.
Observed naming in buyer answers. It does not measure service quality or product merit.
Engines queried: ChatGPT, Claude, Perplexity, Google AI Overviews. Question shapes: best option, specific workflow, top tools, alternative, buyer evaluation. 10 fixed questions per category, 3 documented repeats per question and engine pair.
Evidence and bases
Each category table states the capture date and the observed-answer base. Every percentage is published with its 95% interval and the base it was computed from, in the same cell. Sorted by observed mention rate. These are observed answer counts on stated dates, not a ranking; vendors whose intervals overlap are not statistically distinguishable.
The first two releases publish standings only.
Vendor selection rule
CRM: A vendor is in the checked set if it appears on the relevant category page of at least two major software directories as of the capture date.
Project management: A vendor is in the checked set if it appears on the relevant category page of at least two major software directories as of the capture date.
The checked set is fixed before the answers are collected, so no vendor is added or removed after a result is seen.
Excluded answers
CRM: 3 answers are excluded because an engine returned the recorded reason. Excluded answers reduce their question bases and are not counted as misses.
Project management: No excluded answers were recorded for this category.
Vendor dispute policy
This release measures the frequency with which checked vendors appeared in a fixed set of buyer research answers across stated engines, questions, and dates. It is not a comparative league table. The release data and method are available on the Index site.
A named vendor may request one same-configuration re-run per release at no charge; both results are published side by side, with no silent replacement. Write to index@broadcastwell.com naming the category and the release.
Commissioned engagements
CRM: no vendor in this category commissioned or paid for its inclusion or its result.
Project management: no vendor in this category commissioned or paid for its inclusion or its result.
Read Broadcastwell methodology
Index Verified
A claim verifies control of the canonical domain printed in a record. A claimed record shows Index Verified with the claim date. A claim changes no observation, endorses no vendor and adds no vendor words to the record. Claims, badges, alias corrections and one re-run per release are free for every checked vendor. No payment changes any number on the Index.
Engines badge form
An engine counts for the engines form when the vendor was named in at least one of the ten buyer questions in at least 2 of 3 scheduled repeats on that engine, the same threshold used for the filled marks on each record.
Categories added from a commissioned Category Audit
A buyer of a Category Audit may opt in to publish its category in the next release. Such a category is marked Audit-derived and states its own rule: the questions are the buyer's ten confirmed buyer questions; the checked set is the buyer plus every other vendor named in at least one valid answer, so the set is formed from the answers. The buyer paid for its own measurement and not for any result.
Include the buyer even at zero; it opted in and its zero is the baseline its climb starts from. Exclude every other zero-named vendor.
An Audit-derived category becomes a standing category when it is re-measured under the directory selection rule in a later release. The buyer is disclosed as a commissioned engagement on the category page.
Client climbs
Where a vendor in a category is a Broadcastwell client and has opted in, its record shows a climb strip: the named count release by release with the engagement start marked, and the line "Broadcastwell client since <month>". Counts measured under different engine sets or method versions are shown side by side and never connected into a single series. The numbers are measured under the same method as every other record and are never computed differently for a client.
Agent Gap study
The Agent Gap, October 2026, is a separate study of 70 vendors, three buyer tasks and one agent, ChatGPT Work with Cloud browser. It records 210 primary sessions and 30 preselected repeats, with a single run per task and unknown evidence unscored. Its results are shown on tested records as a study block and are never part of a release's named or cited counts.
Criteria Maps method
The study uses 2100 scheduled source answers across fourteen categories, ten buyer questions, ChatGPT, Claude, Perplexity, Google AI Overviews, Google AI Mode and three runs. Five original exclusions leave 2095 inputs; two documented study exclusions leave 2093 valid answers. Answer capture dates are 22 and 23 September 2026.
A criterion counts once per valid answer when explicitly stated before the first vendor. Canonical labels preserve the coded scope. Runs report the number of three scheduled runs stating the criterion for the same engine and question. Vendor adjacency requires an attributed capability in the same complete answer. Page comparisons cover the dated saved page only; partial or unread scope stays unknown.
Each vendor figure uses the median across the five most named vendors in the frozen category release, including zero counts. Adjacency compares distinct source answers for the same criterion; page coverage compares criteria for the same page state. Medians are descriptive and have no confidence interval.
Perplexity quotations are omitted. Exact original record IDs, offsets and hashes resolve to the deposited source answers. Offsets count Unicode code points; hashes use UTF8. Complete CSV and JSON evidence accompany the study.
Prepared by the Broadcastwell team. Team review: in progress. Zero analyst verified criteria.
The AI written RFP; Study DOI 10.5281/zenodo.23271935. Source DOI 10.5281/zenodo.22907695. CC BY 4.0.
Observed source answer frequencies. Wilson intervals describe the calculation, not independent population inference. Repeated runs are not independent performance tests. Page comparisons cover only the dated page and stated scope. An unstated claim is not an absent product feature.
Any named company may request one no-cost re-run at index@broadcastwell.com; both results are published.
Checklist Audit $490. Every figure comes from the published method, v1.1.
Wrong About You, Study 6
Prepared by the Broadcastwell team
Answers captured under method v1.1 on 22 to 23 Sep 2026 for the Absence Index release 2026-09 (DOI 10.5281/zenodo.22907695). Statements were extracted and classified under the published Record rules and checked against each vendor's own public pages. Every flagged statement is published with its engine, run, question, capture date and the vendor page it was checked against. Team review of the flagged statements is in progress; rows change only if the review changes them, and every change is logged in the public ledger.
The vendor pages were read on 2026-10-08. Every row shows how many of the three runs stated it and its review state. A review date appears only when a review is recorded. These blocks never change a release count.
Any named company may request one no-cost re-run at index@broadcastwell.com; both results are published.
Head to Head, Study 5
Every answer to a head to head question in release 2026-09 was read for the vendor it names the better choice. The vendor page shows the counts for each question that names it, with the category median. Read the study