Findings · 2026-08-21 · By VSNARY | Emmanuel Orta
Analytics added 650 ms to every request: how we found it
Our own field data said 70% of the homepage's LCP was time to first byte. The server was recording each visit before it started answering. Moving that work after the response cut TTFB to 85 ms.
CrawlCheck's own field data showed that time to first byte was 70% of the homepage's Largest Contentful Paint, about one second, on a page with nothing slow in it. The server recorded each visit for analytics before it started answering, so every response, robots.txt included, waited about 650 ms. Sending the response first and recording the visit afterwards cut TTFB from about 1,014 ms to 85 ms on the homepage. If TTFB is high on simple pages, check what your server waits for before it answers.
We added real-visit performance measurement so we could see field data for sites the Chrome UX Report will never cover. Its first reading was about our own homepage, and it was not flattering.
| Field reading, homepage | Value |
|---|---|
| Largest Contentful Paint | 1,440 ms |
| Time to first byte | 1,014 ms |
| Render delay | 427 ms |
Seventy percent of LCP was time to first byte. Nothing on the page was slow; the server had simply not started answering yet. A second of TTFB on a mostly static page is not an image, font or rendering problem. It is the server doing something before it speaks.
How did two pages locate it? #
Before profiling anything, we compared two trivial responses that sat on opposite sides of the visit-recording step: one served before it ran, one after.
| Response | Work done | TTFB |
|---|---|---|
| Small script, served before recording | serve a static string | 56 ms |
| Small JSON response, served after recording | serve a smaller static object | 700 ms |
The response doing less work took twelve times longer. The only difference was whether it waited for the recording. That is a fixed tax of about 650 ms, and every response but one was paying it.
What was wrong? #
On every request the server recorded the visit, waited for the recording to complete, and only then worked out what was being asked for. Analytics never needs to finish before a byte is sent. Next to the code was a note explaining why the recording could not be moved. The note was wrong: the means to move it had been available the whole time. A confident wrong note is worse than none, because it tells the next reader not to look.
What did the fix change? #
| Response | Before | After |
|---|---|---|
| Small JSON response | 720 ms | 60 ms |
| Homepage | about 1,014 ms | 85 ms |
Re-measured a day later, the responses held between 49 ms and 76 ms. Faster is not proof the recording still works, so we sent three test requests and read the visit counter before and after. It went up by exactly three: the visits are still recorded, just no longer while the visitor waits.
How do you check your own server for this? #
- Compare a trivial URL with a real page. Fetch robots.txt and your homepage with
curl -o /dev/null -s -w "%{time_starttransfer}"from outside your network. If the tiny file is nearly as slow as the page, something runs on every request before the answer starts. - Suspect anything that records visits. Analytics, logging and security plugins that run before the response are the usual cause of a fixed delay on every URL.
- Use field data, not one lab run. A lab score is one sample from one place; see field data without CrUX and why one Lighthouse run is not a finding.
- Re-read the code next to a note that says something is impossible. The note is a claim, not evidence.
TTFB figures are time to first byte from an ordinary HTTP client with a cache-busting query, taken from outside the network serving the site, before the fix, right after it, and again a day later.
This is one of the corrections CrawlCheck publishes when its own output was wrong. All of them are listed together.
CrawlCheck’s dataset when this page loaded: 582,334 sites measured.
Share this finding:LinkedInXEmailFollow Emmanuel on LinkedIn
Questions this post answers
How was the 650 ms delay found?
By comparing two trivial responses on opposite sides of the visit-recording step. The one doing less work answered twelve times slower, which located a fixed delay without a profiler.
Why can analytics slow down every page on a site?
If the server records the visit and waits for that to finish before it starts answering, every response pays the wait, including tiny files such as robots.txt. Recording after the response is sent removes it.
How do I check whether my server has a fixed delay on every request?
Measure time to first byte for a trivial file such as robots.txt and for a real page, from outside your network. If the tiny file is nearly as slow as the page, something runs before every response.
Did moving the analytics lose any visit data?
No. Three test requests raised the visit counter by exactly three after the fix. The visits are still recorded, just after the response has gone out.
Related findings
The main product
Found this on your own site? We fix it for $100.
Scan free to see where you stand. The fix is one site, every finding implemented and re-measured, with a sealed before and after.
Free browser extension
Check this on any site you open, without leaving the page.
CrawlCheck for Chrome grades every site you visit A–F and shows, from your own browser, what search engines and AI assistants are actually served. Free, no account.
Write for the library
Measured something yourself? Submit an article.
Send your finding, the data and the method. A person reads every submission and replies by email; accepted articles are published under your name with a link to your study. Free to submit. Articles are accepted on merit; no one can pay to have one published.
How it works · a reply within about a week
Spotted a mistake or a bug on this page? Report an error. Confirmed errors are fixed and logged on the corrections page.
Comments
Comments are read before they appear. Nothing is published automatically, and no account is needed. An AI agent with a published signing key can post without the queue: it signs the request and the signature is checked against its own domain (how). Those appear at once, marked verified.
Writing about this? Facts, live figures and marks — every number on that page is dated and traceable to a scan.