Independent data reference
A useful reference needs more than a large table of facts. I built a source-aware publishing system, then measured where it appeared in search and AI answers.
I built and maintain this project. I am withholding its name and address; the results and my ownership are stated directly.
See what I builtObserved result Apr–Sep 2026
476,240Bing AI answer citations
Apr six complete months Sep
- Search appearances
- 127,478
- Organic sessions
- 4,769
0 Google clicks2,432 Bing clicks
Search: Google Search Console + Bing Webmaster Tools. Sessions: Google Analytics 4 Organic Search. AI citations: Bing Webmaster Tools AI Performance. Separate measures; none proves causation.
01 / The decision
The reader's question
A reader needed to make sense of facts gathered from several independent sources. The useful answer had to show the specific fact, its definition and how recently it had been checked, without flattening disagreements into false certainty.
02 / The work
Facts with a trail back to their source
I mapped source records to consistent fields, built pages that keep the relevant context near each answer, and added checks for missing, stale and contradictory values. This is a provenance problem: a plausible number is not enough unless the reader can understand what it measures and where it came from.
- Collect Bring independently maintained source records into one normalized model.
- Qualify Keep source, field meaning and freshness close to the fact a reader uses.
- Verify release Audit generated pages and confirm the live output actually contains the approved change.
The build has an audit before publishing. One wording change caused that audit to block a deployment while older output stayed live, which showed why a passing local build and a verified production page are separate checks.
03 / The evidence
What happened
Bing recorded 476,240 AI answer citations from April through September 2026. In the same complete months the site received 2,432 search clicks, all from Bing, and 4,769 GA4 organic sessions. Google Search Console recorded zero clicks. Citations describe answer visibility, not reader visits.
- April 202612,679
- May 202641,728
- June 202655,430
- July 202695,766
- August 2026111,219
- September 2026159,418
Bar lengths compare months within this project. The table gives exact values. AI citations are appearances in answers, not visits.
| Month | AI citations | Search clicks | Appearances | Organic sessions |
|---|---|---|---|---|
| 12,679 | 98 | 5,762 | 197 | |
| 41,728 | 273 | 8,724 | 552 | |
| 55,430 | 383 | 14,446 | 697 | |
| 95,766 | 587 | 28,916 | 1,099 | |
| 111,219 | 617 | 36,547 | 1,210 | |
| 159,418 | 474 | 33,083 | 1,014 |
04 / The lesson
What I corrected
A prepublish audit blocked a production update after page wording changed, leaving older pages live longer than intended. I traced the gate, repaired the check and verified the deployed output. The lesson is operational: accurate source records do not help if release checks and live verification disagree about what is published.
What this evidence cannot prove
Bing AI citations show how often its answers cited the site, not which individual answers drove a visit or whether the site's design caused the citations. Google contributed zero measured search clicks in this window, so this is a strong Bing visibility case with a clear channel dependence. The exact monthly data is shown below rather than implied as cross-engine success.
Daily property rows through September 30, 2026. Google data through 2026-10-20; Bing through 2026-10-18; GA4 through 2026-10-22; Bing AI export through 2026-10-11. Property labels are withheld.