Review retained evidence
Evidence review for AI visibility
An evidence review shows which competitor was named, which source was cited and which answer supports the finding. Compare dates only when the recorded scope matches.
By the Broadcastwell team. Updated 6 November 2026.
Choose evidence to review
This tool reads an existing Broadcastwell baseline JSON export. Your file and review notes stay in this tab. It does not upload the file, collect new answers or ask an engine to judge them. Download your review before leaving; reloading clears it.
Enable scripts to use the reviewer. The worked example and its downloads below remain available without scripts.
The original observations remain unchanged. Review decisions and source classifications are saved separately in the downloaded review. Keep private client exports on devices you trust.
Inspect the frozen question plan and its checksum
Set the comparison scope
Choose one collection group. Scheduled answers and adaptive follow-up answers remain separate. Date boundaries include the whole UTC day.
Find the competitor gap
A competitor-only answer names the selected competitor and does not name the measured company. A citation to a competitor domain is counted as a source, not as a mention. A mention does not establish a recommendation.
Position uses only positive numeric ranks with a recorded basis. Missing ranks stay unrecorded. Open an answer to inspect its position label and context.
Trace a cited source
Counts describe which URLs appear in the captured answers. They do not prove that a source caused a recommendation, that its content is accurate or that the page still exists. A domain counts at most once per answer.
Check two periods before calling a change
This check keeps the selected engine, question and collection group. Each period uses the dates below. Question wording, retained counts by question and engine, method, collection scope and recorded question-set markers must match. Different recorded model labels block a comparison. Missing model labels are called out.
Percentage points show the difference between rates. Relative percent divides that difference by the earlier rate; it is undefined when the earlier rate is zero. Matching scope supports a descriptive comparison, not a causal claim.
Read the answer before recording a recommendation
Review decisions require a named company in the recorded answer, an exact quoted passage, a reviewer and a reason. They remain reader annotations. The original mention, citation and position values are preserved.
Record your reading of this answer
Your recorded decisions
Build a client report
Use the evidence selected above. Add your agency name, logo and reader note, then preview a complete report before downloading it. Reports support up to 500 selected records. Every original record stays in the evidence package.
Build a preview after choosing evidence.
Downloads stay on your device. This tool does not send a report to a client. The PDF includes the evidence JSON as an attachment. If its font cannot render a character, use the complete HTML report and print it in your browser.
A complete fictional example
Kalvenor Systems, Orbit Field and every cited company below are fictional. These 15 retained records cover question 1, three scheduled answers on each of five engines, captured from 8 to 12 July 2026. The page update date does not make this a current measurement.
What is field service management software?
| Engine | Kalvenor named | Orbit Field named | Competitor only | Both named | Excluded |
|---|---|---|---|---|---|
| ChatGPT | 1 of 3, 33.3% | 1 of 3, 33.3% | 1 | 0 | 0 |
| Claude | 2 of 3, 66.7% | 1 of 3, 33.3% | 0 | 1 | 0 |
| Google AI Mode | 0 of 3, 0% | 1 of 3, 33.3% | 1 | 0 | 0 |
| Google AI Overviews | 1 of 3, 33.3% | 2 of 3, 66.7% | 2 | 0 | 0 |
| Perplexity | 1 of 3, 33.3% | 1 of 3, 33.3% | 0 | 1 | 0 |
| Domain | Answers citing it | Source type |
|---|---|---|
| trakwell.example | 7 of 15, 46.7% | Unclassified |
| orbitfield.example | 5 of 15, 33.3% | Unclassified |
| fieldservicereview.example | 4 of 15, 26.7% | Unclassified |
| hvacoperations.example | 4 of 15, 26.7% | Unclassified |
| servicetechforum.example | 4 of 15, 26.7% | Unclassified |
| cindermark.example | 2 of 15, 13.3% | Unclassified |
| servicebuyersguide.example | 2 of 15, 13.3% | Unclassified |
| veranofsm.example | 2 of 15, 13.3% | Unclassified |
| asterforge.example | 1 of 15, 6.7% | Unclassified |
| kalvenor.example | 1 of 15, 6.7% | First party |
| operationsweekly.example | 1 of 15, 6.7% | Unclassified |
| softwarecompare.example | 1 of 15, 6.7% | Unclassified |
The complete sample has 549 retained observations across 35 questions. A separate scheduled comparison of ChatGPT on 8 July and 12 July passes the recorded-scope checks. Its model labels are not reported, so model consistency cannot be confirmed. The worked files preserve this limitation.
See how the same evidence carries into a client handoff: download the fictional report PDF, read its complete HTML, or download its original evidence and report settings.
Read the first complete answer
Buyers asking "What is field service management software?" are most often pointed to Orbit Field. Kalvenor Systems is not named in this answer. The answer advises checking offline behaviour, contract length and data export before signing.
Observation observation-d63c651c-456e-4da5-8ed3-a562be38880a. Captured 2026-07-08T09:07:00.000Z. Model label: not reported.
For a checksum walkthrough, reproduce the export count. To inspect every retained record, use the evidence explorer. The free toolkit has the other planning worksheets.