Course lesson 1: what does one AI answer tell you?
Broadcastwell measures and fixes whether AI search engines name B2B software companies.
Run the free checkThe first lesson of our AI visibility course: one AI answer records three separate events on one engine at one moment. It shows whether your company was named, which sources were cited, and who was named beside you or instead of you. It does not show whether a buyer saw it, whether it will repeat, or why the engine chose as it did.
Updated 11 November 2026.
Reading time: about 4 minutes. Lesson 1 of 5 in the course.
Three events, kept apart
When you look at an answer, separate what happened into three lines before you interpret anything.
- Was the company named? A mention of the company as itself, counted once per answer. Our glossary entry for named gives the exact rule.
- Which sources were cited? The links the engine attached. A citation to your own site is a different event from being named, and either can happen without the other.
- Who else was named? The companies a buyer would see in your place. This list is often more useful than your own line, because it tells you whose pages and listings the engine drew on.
Writing these three lines separately stops one event from standing in for another. "We were cited" is not "we were recommended", and "we were not named" is not "the engine dislikes us".
A real answer from the public Index
On 22 September 2026, ChatGPT was asked "best construction management software for a small residential builder" as part of the Absence Index. That one answer named five checked vendors, in this order: JobTread, Contractor Foreman, Buildxact, Foundation Software and Buildertrend. It cited four of their own domains.
Procore was not in that answer. On the same question across ChatGPT, Claude, Perplexity, Google AI Overviews and Google AI Mode, Procore was named in 3 of 15 valid answers, all three from Google AI Mode, while across the whole September category it was named in 109 of 150, more than any other checked vendor. The ChatGPT answer was a true record of one engine's choice; it could not tell you that another engine answers the same words differently. The figures come from the public Index release 2026-09.
What one answer cannot carry
- Frequency. One answer is an observation. A rate needs repeats and a base, which is lesson 3.
- Audience. Nobody recorded who else asked the question that day or what their screen showed.
- Cause. The answer does not explain itself. A cited page is where the engine pointed, not proof of why it chose a name.
- The future. The next run can differ, which is lesson 2.
Our guide on reading an AI answer without overreading it goes further into what one answer can justify.
Exercise
Use the fictional Kalvenor sample (Sample data, fictional company), a made-up field service software company we publish to show the format. In its fictional baseline, Kalvenor was named in 125 of 518 scored answers, its own pages were cited in 66 of 518, and its most named rival, Northvale FSM, appeared in 329 of 518.
Write three sentences, one per event, in this form: "In the fictional Kalvenor baseline, [event] happened in [count] of [base] answers." Then write one sentence on what the rival's count tells you that Kalvenor's own count does not. Keep the word "fictional" in each sentence; a figure that loses its label is how sample data turns into a false claim.
To see how a saved answer looks with its time, engine and sources attached, open our fictional receipt demonstration.
Sources
- Broadcastwell, Absence Index release 2026-09, published 23 September 2026; the answer above is in the long-format benchmark file, captured 22 September 2026.
- Broadcastwell, measurement method v1.1, naming and citation rules, read 9 November 2026.
- Broadcastwell, fictional Kalvenor sample, values as shown on our completed-case format page.
Next lesson: Lesson 2: why ask the same question three times?