October 28, 2026

How I Pushed This Webflow Site Past a 95 Performance Score

Cover for the Design by Alex article on taking a Webflow site past a 95 performance score

The short version

This website, Design by Alex, scores above 95 on performance. The motion, the type, and the case studies stayed. What changed was the first few seconds of the load. I did the pass with the public performance report in one hand and an AI assistant in the other. The assistant did not redesign the site. It made the report readable, so the cuts were obvious. This is that story, and it is the same pass I run on a client Webflow site that looks finished and still feels heavy.

Written by Alex Nouri, the designer and Webflow developer behind this site. Published October 28, 2026. The score in this article is the performance score from a lab test of the live homepage, the kind Google's public tools report. It is not a promise that every page, on every phone, on every visit, will show the same number.

What a score above 95 actually means

A performance score is a lab snapshot. It asks how quickly the page can paint something useful, how soon that content stops jumping around, and how quickly the page answers a tap. Google groups the field version of those questions under Core Web Vitals: largest contentful paint, interaction to next paint, and cumulative layout shift. The lab score is the number people screenshot. The field numbers are what a real visitor feels.

On this site the lab score sits above 95. Getting there did not mean stripping the design down to a white page and a system font. It meant the browser could show the page before it had finished every flourish.

The site that already felt fast

I build this site the way I build client work. Large type. Scroll motion. Case-study imagery. A navigation that has a life of its own. In the studio, on a warm cache, it felt fine. That feeling is a trap. A designer's laptop is not the test. The test is a cold visit, on a modest phone, with the scripts, the fonts, and the images all arriving at once.

The first honest look was a performance report, not a guess. The report listed what blocked the first paint. An AI assistant sat next to that report and did one job well: it translated a long waterfall into a short list of causes. It named files. It did not invent a new visual direction, and it did not ship the fixes. I did.

What the report kept pointing at

The same few families of weight show up on almost every Webflow site I open, including this one before the pass.

  • Images that are larger than the rectangle they sit in, or that arrive in a format the browser has to work harder to decode.
  • Fonts that hold the text back until a heavy file has downloaded, so the first paint waits on type the visitor has not asked to admire yet.
  • Scripts and interactions that run for the whole page even when the visitor is still looking at the hero.
  • Third-party tags that have nothing to do with the first view and still sit on the path to it.

None of those is a Webflow limitation. Webflow will happily publish a careful site and a careless one from the same Designer. The score follows the load, not the logo in the footer.

What changed on this website

I am not publishing the recipe. A public checklist goes stale the week the browser or Webflow changes a default, and a half-applied checklist is how a score gets worse. What I will say is the shape of the work, because that is the part worth understanding before you hire anyone.

The pictures were brought down to the size of the frame they actually fill. Type was allowed to appear without waiting on every weight in the family. Motion stayed, and it stopped demanding the browser's attention before the first paint. Anything that was not needed for that first view was moved off the critical path. The page you are reading is the result: the same studio site, with a performance score above 95.

The AI tool mattered at the start and at the check. It read the report with me, argued with my guesses, and later compared the new report with the old one so I could see which change had earned its keep. It never replaced the judgment about what the brand was allowed to lose. That call stays with the person who designed the page.

What I do when the site is yours

The pass on a client project follows the same story, on their URL.

  1. Measure the live site, not a staging guess and not a screenshot of someone else's score.
  2. Separate the weight that moves the score from the motion and the imagery the brand actually needs.
  3. Change the load. Leave the look.
  4. Measure again, on the same kind of test, and keep only the changes that earned the points.

If the design itself is the problem, I say so. A score above 95 on a page nobody understands is not the job. The job is a site that still feels like the brand and no longer makes the visitor wait for it.

Questions people ask before they send a URL

What is a good performance score for a Webflow site?

Treat 90 as the line where the lab test stops complaining. Above 95 is the range I hold this site to. The number moves a few points between runs, so I care that it stays in that range, not that one screenshot says 100.

Do I have to delete the animations?

No. This site still moves. Animations become a problem when they run before the page has anything to show, or when they shove the layout around after the text has already painted. The fix is timing and weight, not a ban on motion.

Can an AI tool optimize the Webflow site by itself?

It can read the report and suggest where to look. It cannot decide which interaction your brand is willing to lose, and it should not be given the publish button. On this site the assistant was a reader. The edits were mine.

Who is this for?

A founder or a marketing lead with a Webflow site that already looks right and still loads like it is thinking about it. Also a studio that would rather hand the performance pass to someone who builds in Webflow every week.

Send the URL

If your site is the one that feels finished in the Designer and heavy on a phone, that is the brief. Bring the live link to a discovery call. I will tell you what the report is actually angry about, and whether a performance pass is the right job or whether the page needs a redesign first.