Lighthouse Performance of 1,497 Vibe-Coded Project Pages
78.8% of 1,497 measured VCL project pages scored below Lighthouse’s Good threshold. A desktop lab benchmark, not a verdict on every app or its real-world users.
Updated November 1, 2026. This article uses dataset v10, with a November 1 cutoff and measurements recorded through October 31. This is a new frozen edition; changes in sample membership, recorded classifications and measurement timing are not within-project growth.
1,179 of 1,497 measured public project pages—78.8%—scored below 90 in Lighthouse performance in VCL’s November report. The median score was 66.
Most measured pages therefore fell short of Lighthouse’s “Good” band. But “below Good” is not synonymous with “Poor”: the largest group scored between 50 and 89.
Download the chart · Download the source CSV
Three score bands, three different results
| Lighthouse performance score | Project pages | Share of 1,497 measured pages |
|---|---|---|
| 0–49 · Poor | 327 | 21.8% |
| 50–89 · Needs Improvement | 852 | 56.9% |
| 90–100 · Good | 318 | 21.2% |
These bands follow Lighthouse’s published scoring definitions. The 78.8% finding combines the first two groups: 327 plus 852 pages. It should not be reported as the percentage of “broken apps.”
Measurements were available for 1,497 of the 1,502 project URLs in the report. The other 5 were not assigned a score of zero or counted as Poor.
Measurement selection. This edition excludes stored Lighthouse results with recorded runtime errors or incomplete core scores, then uses the latest remaining eligible observation. Older results without full execution records retain uncertainty. This rule does not certify every historical score.
At the cutoff, the median measurement was 5.7 days old; the oldest was 53.5 days old.
What was actually tested
Each result is a single desktop lab run against the submitted public page, using the latest successful recorded measurement at or before the cutoff. If a project submitted an app-store or marketplace listing, the score describes that listing page—not the application behind it.
This is not a Core Web Vitals field assessment, a mobile-user study or a test of every authenticated screen in a product. Lighthouse scores can also vary with test conditions. Historical runtime-version records are incomplete, so the edition preserves the recorded results without claiming exact reproduction of every run.
The sample consists of VCL-listed projects, with no matched non-AI-built control group. It therefore cannot establish that AI coding caused the observed scores or that one coding assistant produces faster websites than another.
What it does provide is a concrete public-page benchmark: 318 measured pages reached the Good band, while 1,179 did not under the recorded desktop tests.
The three-benchmark comparison shows how this performance threshold overlaps with backlink strength and agent readiness among custom-domain projects with complete measurements.
Data and source. State of Vibe Coding 2026, Dataset v10: Lighthouse performance, cutoff November 1, 2026 at 00:00 UTC. See the edition’s methodology for scanner settings and submitted-page scope.
The full State of Vibe Coding report presents this measure separately from search visibility and agent readiness.