A sudden loss of AI citations could reflect a change to a website or a change to the engine itself. We wanted to see whether a ChatGPT model upgrade changed the number of source domains behind its answers.
We compared 423,341 answers from steady windows before and after the flagship upgrade, including 16,161 questions observed under both versions. We also tracked an unchanged control model to help distinguish the upgrade pattern from broader changes.
Across 16,161 matched questions around ChatGPT's early March 2026 flagship upgrade, average cited source domains per answer fell from 23.4 to 12.1. An unchanged control model held steady, pointing to a platform change rather than a site-specific loss.
A model upgrade halved the sources behind each answer
A model upgrade on the vendor's side is a real cause of AI-visibility change, and it can move more than anything a brand does to its own pages. When ChatGPT's flagship version upgraded in early March 2026, the evidence behind each answer thinned out across the changeover.
Sources cited per answer fell from a mean of 23.4 to 12.1, a 48% cut. The drop was not gradual. It lined up with the model change, not with anything on the sites being cited.
Sources cited per answer
- Before the upgrade23.4
- After the upgrade12.1
Takeaway
On a matched prompt set, 99.7% of answers cited fewer sources
Window averages can hide a shifting mix of questions. To rule that out we held the questions fixed and compared 16,161 prompts that ran under both versions of the flagship.
Of those matched prompts, 16,116 cited fewer sources after the upgrade and only 45 rose. That is 99.7% moving the same direction, with a mean drop of about 11 sources per prompt. This was not noise.
- Prompts run under both versions
- 16,161
- Cited fewer sources after
- 99.7%
- Mean drop per matched prompt
- −11.3
An unchanged control model held steady, isolating the cause
If the drop were a quirk of measurement, a model that did not change should have fallen alongside the flagship. A smaller, unchanged model ran throughout the same weeks as a control.
The upgraded flagship fell 48% while the control held steady, even ticking up 4% over the same window. The gap between the two is about 12 sources per answer. The drop tracks the model upgrade, not how we measured it.
- Upgraded flagship−48.1%
- Unchanged control+4.4%
Takeaway
The control sat where the flagship used to, then the flagship dropped under it
Before the upgrade the two models cited almost the same number of sources, 23.4 for the flagship and 23.6 for the control. After, the control still cited 24.6 while the flagship had fallen to 12.1. Same weeks, same measurement, two very different paths.
- Upgraded flagship, before to after
- 23.4 → 12.1
- Unchanged control, before to after
- 23.6 → 24.6
- Sources apart after the upgrade
- ~12
Every industry category fell
The cut was not concentrated in one kind of question. Every industry group in the sample fell, and answers after the upgrade landed at roughly 11 to 13 sources regardless of topic.
The biggest losers were the categories that had been citing the most. Software and artificial intelligence answers, which led before, fell the hardest. The model pulled the top categories down toward the rest.
Sources before · 6 results
- Software25.4
- Artificial intelligence25.4
- Financial services23.5
- Health care22.2
- Commerce and shopping22
- Community and lifestyle20.4
Explore the data (6 rows)
| Software | 25.4 | 12.5 | -12.9 |
| Artificial intelligence | 25.4 | 12.7 | -12.7 |
| Financial services | 23.5 | 12.8 | -10.7 |
| Commerce and shopping | 22 | 11.4 | -10.7 |
| Health care | 22.2 | 11.8 | -10.4 |
| Community and lifestyle | 20.4 | 11.6 | -8.8 |
After the upgrade, answers landed in a narrow range, whatever the topic
Across all 31 industry segments in the sample, the averages after the upgrade landed in a tight 11 to 13 source range. Where a question came from used to shape how much evidence it drew. After the upgrade it barely did.
Citation counts went from a wide spread to a near-fixed number
Before the upgrade, source counts spread widely. Answers ranged from the mid-teens to past 40, with a long tail of deep-citing answers above 35. The spread, measured as standard deviation, was 7.6.
After, the range collapsed. 88% of answers now cite between 10 and 15 sources, and the spread roughly halved to 3.0. Answers stopped occasionally reaching deep and settled into a fixed band.
- Of answers now cite 10–15 sources
- 88%
- Spread in source counts, before to after
- 7.6 → 3.0
- Of old answers cited 20–25 sources
- 47%
The cut was broad and even, not a shift toward big domains
Citation counts fell, but the domain shares barely changed.
The share of citations held by the top 20 domains barely moved, 9.6% to 9.9%, and the top 100 share moved just as little. The drop landed roughly proportionally across the whole field rather than favoring the biggest sites.
- Top 20 share, before9.6%
- Top 20 share, after9.9%
Takeaway
A third fewer citations, and a fifth fewer distinct domains
Halving sources per answer shrinks the whole citation footprint. Total citations across the window fell 37% and the number of distinct domains cited fell 23%. The web that AI draws from got both shallower per answer and narrower overall.
- Total citations
- −37%
- Distinct domains cited
- −23%
- Top 20 domain share (held flat)
- 9.6% → 9.9%
What this means for your AI visibility
A citation drop can be the model, not you. Before you treat a fall in citations as a content problem, check whether it lines up with a model change. The cause may be upstream of anything on your site.
Fewer citations raises the bar. When each answer cites fewer sources, the contest to be one of them is sharper. Breadth of presence matters more, not less.
Track across model releases. A single visibility number read in isolation will blame the wrong cause when a model shift happens. Watch the trend across releases, not just week to week.
How we measured this
We counted the number of distinct source domains behind each ChatGPT answer across a steady window before the flagship upgrade and a steady window after, then compared 16,161 prompts that ran under both versions. A smaller, unchanged model ran throughout as a control. The upgrade is anchored on the observed changeover in early March 2026.
This covers one engine and one upgrade event, so it does not predict the direction of future upgrades. Source count measures breadth, not influence per source, so fewer citations does not strictly mean less downstream effect. Figures describe a measured sample of AI answers, not Perplexity, Gemini, or Copilot.
How we measured
We compared 423,341 ChatGPT answers from steady windows before and after a flagship model upgrade, with 16,161 prompts run under both versions and an unchanged control model.
- Sources cited per answer
- 23 → 12
Mean, before and after
- Drop on the upgraded flagship
- −48%
- Of matched prompts cited fewer sources
- 99.7%
- Change on the unchanged control
- +4%