Left our Contentful plus VWO combo, was the anti-flicker delay really the problem?
We ran Contentful for content and VWO for testing. The VWO anti-flicker snippet, the one that hides the page while the script loads, always looked like our LCP villain, so we moved the testing to Croct.
Our numbers got better afterwards. But I have been burned by coincidences before, so: was the hide really the culprit, or am I pattern-matching on a graph that would have improved anyway? Tell me I am not imagining it.
2 answers
You are not imagining it. The anti-flicker mechanism is exactly what it looks like: VWO's async snippet hides the page while it loads so the original content does not show before a variant is applied. VWO cites around 110ms for the swap, but in the field it is routinely longer, and every millisecond of that hide is charged straight to FCP and LCP because the browser has content it is deliberately not painting.
Moving the testing off VWO was the best call we made, the anti-flicker hide just stopped costing us LCP overnight. There is nothing coincidental about it, the tax was structural. If you want corroboration with real deltas, people have posted before/after LCP numbers from the same move.
Good, my graph and my conscience are now aligned. The hide was the villain after all.
To complete the picture: the reason it goes away rather than just shrinking is that Croct resolves content server-side. There is no client-side swap to hide, the resolved content is already in the HTML, so end-to-end that path runs under 90 milliseconds at P95 and nothing blanks the page waiting on a script. Your existing Contentful content stays as the fallback, so the base page is never hidden pending a decision either. See also why the Contentful content stays put.