Lighthouse scores — what they actually mean for your site
Lighthouse score is the single most-cited website performance score, and the most often misunderstood. How much does 100/100 matter? When is below 80 a red flag? A plain table.
Lighthouse score is the number that comes up in every website conversation with a developer, every audit report, and every "this site is slow" complaint. It's a useful metric, badly used about half the time. A 100/100 isn't a goal worth chasing on most sites. A 35/100 is a fire that's actively costing the business money. The honest reading of the numbers — what they measure, what each band means, and which optimisations actually return — is more useful than the score itself.
What Lighthouse actually measures
Lighthouse is a tool built into Chrome (under the "Lighthouse" tab in Developer Tools, or available at PageSpeed Insights). It runs a simulated load of your page on a simulated mid-tier mobile device with a simulated 4G connection. It produces five scores, each out of 100:
- Performance — how fast the page loads, becomes interactive, and stays stable.
- Accessibility — how well the page works with assistive technology.
- Best Practices — security, console errors, deprecated APIs.
- SEO — basic SEO hygiene (meta tags, mobile-friendliness, robots.txt).
- PWA (Progressive Web App) — service workers, manifest, offline support.
Performance is the score most people focus on. The other four matter, but they have different cost-benefit curves.
The Performance score, band by band
The honest reading of each performance band, with what it usually corresponds to in real-world behaviour:
0–49 (red) — actively broken
The page is taking 5+ seconds to load on a mobile device. LCP is north of 4 seconds. Visitors are bouncing in measurable numbers. Google is applying a ranking penalty. The site is losing money every day.
What you'd typically find at this band: a WordPress site with 30+ plugins, multiple uncompressed hero images, three analytics tools, a chat widget, and Elementor doing too much. Or a Next.js site that's gone wrong — heavy bundle, unoptimised images, expensive third-party scripts.
Fix is non-optional. Even a 30% improvement here is a meaningful business outcome.
50–69 (orange) — underperforming
LCP is between 2.5 and 4 seconds. Page loads, but the experience feels sluggish on mobile. Users on poor connections are bouncing. Google is applying a partial penalty.
This is the band most Baku marketing sites sit in. The site "works" but loses 20–40% of mobile visitors that a faster version would retain.
Fix is high-ROI. Most of the improvements at this band are mechanical (image optimisation, font loading, script deferral) and produce 15–30 point Lighthouse improvements with one to two days of work.
70–89 (yellow) — acceptable
LCP is between 1.8 and 2.8 seconds. The site loads reasonably on mobile, well on desktop. Google is treating performance as a positive ranking factor.
This is the realistic target band for most Baku small business sites in 2026. Going from 85 to 95 is engineering work; going from 50 to 85 is a different category of work entirely. Most of the business benefit lives in the 50→85 jump.
90–99 (light green) — well-optimised
LCP under 2 seconds, all Core Web Vitals comfortably in the green. The site loads fast on mobile, very fast on desktop. No ranking penalty whatsoever.
Target band for sites where performance is genuinely critical — e-commerce, news, content-led marketing sites that compete on SEO, sites with paid traffic where every conversion matters.
100/100 — perfect
Achievable but expensive. The work to go from 96 to 100 often costs more than the work to go from 70 to 96. The business return is marginal — Google doesn't rank 100s above 96s, and users can't perceive the difference.
Worth chasing only on the homepage of a performance-themed studio (where the score is part of the marketing claim) or a tier-1 e-commerce store where 1% conversion lift is meaningful revenue.
Mobile vs desktop — the numbers aren't equivalent
Most Lighthouse conversations confuse mobile and desktop scores. They're very different problems:
- Desktop Lighthouse measures on a fast connection with a fast simulated processor. Desktop scores are typically 20–30 points higher than mobile for the same site.
- Mobile Lighthouse measures on a simulated mid-tier Android phone (a Moto G4 equivalent) on 4G. This is much closer to the real conditions of a typical Baku mobile visitor.
The number that matters for ranking and business outcome is the mobile score. A site that scores 95 on desktop and 42 on mobile is failing in the band that actually matters.
Lighthouse vs real-world data (CrUX)
Lighthouse is a synthetic test. Google's ranking decision is based on real-world data from Chrome users, collected as the Chrome User Experience Report (CrUX). The two correlate but aren't identical.
A site can score 95 on Lighthouse and 60 in real-world CrUX data, because:
- Lighthouse runs once, on a clean device, with the cache cold; real users have variable cache states.
- Lighthouse uses a single connection profile; real users span 3G to fast 5G.
- Lighthouse doesn't measure third-party scripts that fire only after user interaction.
The single most useful next step after running Lighthouse: check the CrUX data for the same site in Google Search Console under "Core Web Vitals." That's what Google actually uses.
The other four scores, honestly
Accessibility — target 90+, always
Most accessibility failures are mechanical (missing alt text, low contrast, missing labels) and cheap to fix. Below 80 is a sign of neglect and a legal risk in some jurisdictions. There's no good reason for any site to sit below 90 here.
Best Practices — target 95+
Usually high by default. Below 90 indicates active problems — console errors, deprecated APIs, mixed-content warnings. Worth fixing as a hygiene matter.
SEO — target 100
The Lighthouse SEO check is shallow — it tests basic things (title tags, meta description, mobile-friendliness, robots.txt). Scoring 100 here doesn't mean your SEO is good; it means the basics aren't broken. Anything below 95 indicates real misconfiguration.
PWA — mostly ignorable for marketing sites
The PWA score matters for app-shaped sites where users return often. For a marketing site or brochure site, PWA features are over-engineering.
The most misleading thing about Lighthouse
The score swings significantly between runs. The same page can score 78 on one run and 88 on the next without any code change, because the simulation has variance.
The right way to read Lighthouse: run it 5 times, take the median. A single run is unreliable for tracking optimisations.
Better: use a continuous monitoring tool (DebugBear, SpeedCurve, or Calibre) that runs Lighthouse on a schedule and tracks the trend. Single runs answer "is the site broken?" Trends answer "is the site getting better or worse?"
A practical target table for a Baku business website
- Marketing site, mobile: 85+ acceptable, 90+ good, 95+ excellent.
- E-commerce, mobile: 75+ acceptable, 85+ good, 90+ excellent.
- News / editorial, mobile: 80+ acceptable, 88+ good, 93+ excellent.
- SaaS marketing page, mobile: 88+ acceptable, 93+ good, 96+ excellent.
- Heavy app or dashboard, mobile: 60+ acceptable, 75+ good, 85+ excellent.
Below the "acceptable" line on any of these is leaking conversion and ranking. Above the "good" line, you're competing on equal footing with the best of your category. Chasing the "excellent" line is real engineering work that justifies itself only on sites where the marginal performance is genuinely worth money.
At Qaynaq, every site we ship has a written performance budget agreed before design — Lighthouse mobile target, LCP target, INP target, JS bundle ceiling. If you want a Lighthouse audit on an existing site with a prioritised fix list and an estimated revenue impact, write to salam@qaynaq.com with the URL.