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. Its name and URL stay off this page; the numbers are its own.
See what I builtObserved result Apr to Oct 2026
602KBing AI answer citations
Apr seven complete months Oct
- Impressions
- 166K
- Visits from search
- 5.7K
0 Google clicks2.9K Bing clicks
Search: Google Search Console + Bing Webmaster Tools. Visits from search: Google Analytics 4. AI citations: Bing Webmaster Tools AI Performance. These count different things, and none of them shows what caused a change.
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.
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.
What happened
Bing recorded 602K AI answer citations from April through October 2026. In the same complete months the site received 2.9K search clicks, all from Bing, and 5.7K GA4 organic sessions. Google Search Console recorded zero clicks. Citations count answers that named the site, clicked or not.
- April 202612.7K
- May 202641.7K
- June 202655.4K
- July 202695.8K
- August 2026111K
- September 2026159K
- October 2026126K
Bar lengths compare months within this project. The table gives exact values. An AI citation counts an answer naming the site, whether or not anyone clicked.
| Month | AI citations | Search clicks | Impressions | Visits from search |
|---|---|---|---|---|
| 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 | |
| 126,247 | 469 | 38,369 | 913 |
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. They can't show which 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 above rather than implied as cross-engine success.
Complete months, April through October 2026. Google data through Nov 5, 2026; Bing through Nov 4, 2026; Google Analytics through Nov 7, 2026; Bing AI export through Nov 4, 2026.