My Site Scored 75 out of 100. This Is Exactly What I Fixed

An audit scored my own site 75 out of 100. The most uncomfortable line in the report was not the number. It was that searching "AI consultant Panama", my site did not appear anywhere in the top ten. Smaller competitors did.

I set out to fix it and documented every change. This article is that log, with the numbers before and after, so you can check the same things on yours.

Everything here is verifiable: open PageSpeed Insights, enter chiragx.com, and look.

The problem was not technical. It was architecture

I offer six services. My site had one URL with six anchors inside the home page. They were anchors, not pages.

One URL cannot compete for six topics at once. They cannibalize each other and none of them wins. That is why I did not show up: there was no page answering "AI consulting" better than someone who actually had one.

That is not fixed with keywords or by writing more blog posts. It is fixed by creating the pages. I went from 6 anchors to 14 real URLs, one per service in Spanish and English, plus the two index pages.

If your company does several things and they all live on the home page, that is your first problem and probably your most expensive one.

The bug I found in my own code

Reviewing the head I found three Google Fonts tags loading the same typeface. Two were set up correctly, loading asynchronously. The third was a plain stylesheet, and that one blocks rendering: the browser has to wait on someone else's server before painting the first pixel.

Someone set up the correct pattern and left the old one underneath. It is a silent failure: no console error, nothing visible, and on a slow mobile connection it costs you seconds.

I fixed it by self-hosting the fonts. And there I found the second surprise: Google serves a single variable file per family. I was downloading the same file three times under three different names, one per weight.

From 8 files and 273 KB to 4 files and 123 KB. And zero third-party connections needed to paint the page.

Layout shift, the one that actually scares people off

My CLS was 0.137. That means content moved while the page loaded. It is the metric that annoys real users most: you go to tap a button and the text jumps.

The cause was typography. Text painted in the system font and then switched to the real one, with different metrics, and everything below moved.

The fix was not dropping the typeface. It was measuring in the browser how much space my real font takes versus the fallback, then generating a fallback with those exact metrics, so the text occupies identical space before and after.

CLS: 0.137 → 0. Zero shifts, measured in a real browser, not estimated.

Images: 60% of the weight, and usually the easiest win

I converted the whole catalog to AVIF, a format most sites in Panama still do not use. The catalog went from 6.7 MB to 2.1 MB. A service page dropped from 675 KB to 128 KB.

But what surprised me most was not the format. It was this: two event flyers weighed 308 KB combined and were displayed at 118 pixels wide. The browser downloaded the full image and shrank it with CSS. Regenerated at the real size: 19 KB.

And the podcast thumbnails were pulling the 1280 pixel version to display at 384. Switched: from 263 KB to 43 KB.

Neither of those is a format problem. It is that nobody checked what size they were being displayed at.

The final numbers

• PageSpeed desktop: 92 → 100
• PageSpeed mobile: 77 → 88
• CLS: 0.137 → 0 on desktop and mobile
• First paint on mobile: 3.0s → 1.7s
• Home page images: 1,013 KB → 533 KB
• Full load time: 1,166 ms → 734 ms
• Metadata: 102 of 102 URLs within limits

What you can check on your own site this week

Without hiring anyone, in under an hour:

1. Run PageSpeed Insights on mobile, not desktop. Desktop always looks fine and is not where people visit you from. Run it three times and take the middle value: it varies between runs.
2. Look at CLS. If it is above 0.1, your content moves while loading and you are losing people who were about to tap something.
3. Count your services and count your URLs. If you offer five things and have one page, that is your ceiling.
4. Search your own term on Google. Not your brand: the service. Whatever you actually sell. If you are not on the first page, look at who is and what they have that you do not.
5. Open your site on your phone using mobile data, not wifi. It is the most honest test and the most ignored one.

Why I publish this with open numbers

Because in SEO it is easy to talk well and hard to prove anything. Anyone can say they improve sites. Few show the before and after of their own, with the public tool so you can verify it.

This is the same work I do for companies in Panama and across Latin America: find what is holding the site back, fix it, and measure it before and after. Not with promises of first place in thirty days, because that is smoke.

If your site loads slowly, does not show up when people search for you, or simply has never brought you a client, I can tell you on one call what is going on. And if you only need two fixes, I will tell you which two instead of selling you a redesign.