The Absence Manual¶

What this is¶

The Absence Manual is a technical manual on AI search visibility for B2B software: how to measure how often an AI answer names your company, how to read what that measurement does and does not tell you, and what the evidence actually supports doing about it. Every finding in it is drawn from three already-published studies with DOIs, open data and open analysis code. It is free, ungated and permanently online. There is no form, no login, no paywall and no email field anywhere on this site, because an AI crawler issues a plain HTTP request and cannot fill in a form. A gated manual about citability would refute itself.

Author. Sairam Sivakumar, Broadcastwell.

Operator disclosure. Broadcastwell ran this measurement and sells services in the category it measures. Broadcastwell is excluded from the measured sample and from every ranking. The mitigation is not that the conflict is absent, it is that the raw data and the code are public and the result can be recomputed by anyone who disagrees.

What it is built from¶

Volume	What it covers	DOI
Volume I	85 B2B software companies, 61 categories, 860 scored answers and 5,160 traced citations, one engine held constant. The dataset README additionally records 1,753 unique domains cited	10.5281/zenodo.21537014
Volume II	The Absence Ladder. All 616 absence records classified by question shape	10.5281/zenodo.21586091
Volume III	Cross-engine divergence. 280 questions, 40 categories, four engines, 853 answers collected 05:28 to 16:29 UTC on 2026-08-04, with a 25-question repeat subsample run three times	10.5281/zenodo.21789120

All three are CC BY 4.0, with their data and analysis code at github.com/Broadcastwell/state-of-geo-2026.

How to read the evidence labels¶

Every substantive claim in this manual carries one of three labels, so you always know what kind of thing you are being told. This is the manual's central discipline and it is applied without exception.

Measured. The claim comes from one of the three volumes, and the volume is named in the same sentence or the label. You can recompute it from the published data.

Reported. The claim comes from a source outside this research programme, cited with its publisher and date. You should check it at the source.

Reasoned. The claim is an inference drawn from the measured or reported material above it. It is argument, not measurement, and it is marked so you can disagree with the reasoning without disputing the data.

A claim with no label is a defect. If you find one, it is a mistake and it will be corrected.

How to read it¶

Every chapter has its own permanent URL and stands alone. Statistics carry their sample size in the sentence that states them. Limitations are stated in the body of the argument rather than collected at the end where they can be skipped. Where the evidence is strong the claim is unhedged, and where it is thin the chapter says so and names the base it rests on. Where a number is missing or two published sources disagree, the manual flags the gap rather than filling it with a plausible value.

Reading paths¶

I need to understand the category¶

About 20 minutes.

I need to evaluate a provider¶

About 25 minutes.

I need to run a measurement program¶

About 30 minutes.

Chapters¶

No.	Chapter	Status
0	Selection, not ranking	Published
1	One in three vendors is never named	Published
2	The Absence Ladder	Published
3	Cited is not recommended	Published
4	Who the engines actually cite	Published
5	One engine is not four	Published
6	Your score is noisier than your vendor admits	Published
7	The AI Overview does not always appear	Published
8	What a valid measurement requires	Published
9	Getting named at the category door	Published
10	Getting quoted at the comparison gate	Published
11	How long this actually takes	Published
12	How to buy GEO without getting sold a number	Published

Appendices¶

No.	Appendix	Status
A	The ten-question buyer bank	Published
B	Absence classification rules and precedence	Published
C	Scoring and matching specification	Published
D	Glossary	Published
E	References and research provenance	Published

Every chapter and appendix is published. A slug published once is permanent: version numbers change, URLs never do. See the changelog.

The whole manual as one PDF¶

Download the complete manual as a PDF. Every chapter, every appendix, the source table with its DOIs and the dated research provenance, in one file. It is generated from this same source in the same build, so the PDF and the pages cannot drift apart. There is no form, no email field and no login: the link downloads the file. That address always serves the current version, and the versioned filename is preserved alongside it.

For marketing leaders¶

If you are not going to read a technical manual, start with what AI search is doing to your category. Plain language, no jargon, one page.

If you want this run for your category¶

Offer record, current as of 11 September 2026.

The free 10-question check: ten buyer questions for your category on one engine, Perplexity, returning your position on the absence ladder and the chapter that addresses it. It runs in the browser in under a minute.

Category Audit, $490 once. One category, ten buyer questions, five engines, three measured runs. Which vendors the engines named instead with counts, which of the ten questions you lose, the sources the engines read, your position on the absence ladder, and a written findings document with three prioritised fixes, within 48 hours of the category being confirmed. The full $490 is credited against the Diagnostic if the Diagnostic is bought within 30 days. Start with the $490 Category Audit

AI Visibility Diagnostic, $990 once. 35 buyer questions across shortlist, role, use case and evaluation groups across five engines, three scheduled runs per question and engine plus up to two further runs on any pair whose verdicts disagreed: 525 scheduled observed answers, named and cited outcomes reported separately, competitor and source analysis, a receipts appendix and a prioritised plan. Adaptive observations are extra. AI answers vary, so repeated observations and their disagreement are reported. The full $990 is credited against the program's first monthly invoice. Get the Diagnostic, $990

The program, $13,500 per 90 days, billed $4,500 monthly. A 90 day initial term, then month to month. Sold by conversation at broadcastwell.com/contact.

Agencies buy the same work wholesale at $1,000 per client per month for the first client and $490 per client per month for each additional client at broadcastwell.com/for-agencies. Three clients are $1,980 a month; five clients are $2,960 a month. The agency keeps the client relationship and implements the fixes. Every figure comes from the published method, v1.1. The full ladder is at broadcastwell.com/pricing.

If Broadcastwell is not the right fit, you get your money back. A standalone Category Audit or AI Visibility Diagnostic is refundable in full, no questions asked, if requested within 30 days of delivery of the findings. The first month of the program is refundable in full, no questions asked, at any time before month two is invoiced. You do not need to give a reason, and you keep every deliverable produced up to that point. The only exception is third-party costs paid on your behalf, and we tell you about any of those before they are incurred. See the master refund and proof-gate terms.

Notes¶

Short standalone pieces cut from the chapters, each linking back to its parent.

Same score, different problem, from Chapter 2.
Half its own shortlist, from Chapter 6.

Licence and status¶

Prose and figures are published under CC BY 4.0. The site code is MIT. Nothing here has been through academic peer review: it is measurement published with the data and code that produced it, and the standing invitation is to recompute any figure you doubt from the public files and publish what you get. The dated research record and collection designs remain available in the published source files. Compare observations only with their collection dates and run structures attached; a change in run count changes the evidence base.

Version 1.0, August 2026.