What is a good Lighthouse score?

If you’ve ever run your site through Google’s PageSpeed tool and gotten back a set of colored numbers, you’ve met Lighthouse. It’s the diagnostic engine built into Chrome that grades how well a web page is built. The numbers are useful — but only if you know which ones matter for a business, and which are just for engineers.

Here’s the plain-English version.

What Lighthouse actually measures

Lighthouse scores a page in four classic categories, each from 0 to 100:

• Performance — how fast the page loads and becomes usable. This is the one that matters most to you. It’s the number that reflects whether a visitor waits or leaves.
• Accessibility — whether people using screen readers, keyboard navigation, or with low vision can actually use the page. Also increasingly what lets automated tools read it.
• Best Practices — general web hygiene: secure connection, no broken images, no obvious security issues.
• SEO — the on-page basics search engines look for: a title, a meta description, readable text, a crawlable structure.

Each category gets a color:

• 90–100 = green. Good.
• 50–89 = orange. Needs work.
• 0–49 = red. A problem.

What counts as “good”

Aim for 90+ (green) on all four. That’s the honest target, and it’s very achievable — most of our rebuilds land at 95+ on Performance on mobile.

But here’s the nuance most “what’s a good score” articles skip: Performance is the hard one, and it’s the one that pays. Plenty of template sites score 90+ on Accessibility, Best Practices, and SEO while sitting at 30 or 40 on Performance. That green-heavy report card hides the one red number that’s actually costing you visitors — because Performance is where slow load times show up, and Google found 53% of mobile visitors leave a page that takes over 3 seconds.

So when someone waves a “we scored 100 on SEO” report at you, ask the only question that matters for your phone ringing: what’s the Performance number on mobile?

Is a Lighthouse score the same as your Google ranking?

No — and this trips people up. Lighthouse is a diagnostic tool, not Google’s scoreboard for you. Google does use some of the same underlying signals (notably Core Web Vitals, which are speed and stability measures) as ranking factors, but your Lighthouse number itself is not your ranking. Think of it like a blood-pressure reading: it tells you about your health and predicts problems, but it isn’t the same thing as your doctor’s final diagnosis.

Use it to find and fix what’s slowing you down. Don’t treat the number as a promise about where you’ll rank — anyone who tells you a Lighthouse score guarantees rankings is overselling it.

The new part: Lighthouse’s AI-readiness (“Agentic Browsing”) category

This is the genuinely new development, and it’s why owners are suddenly asking about an “AI score.”

In 2026, Lighthouse added a new category to its default report called Agentic Browsing. In plain terms: it grades whether an AI agent — ChatGPT while it’s browsing, Perplexity, Gemini’s agentic mode, and similar assistants — can actually read your page and act on it. As more customers start asking an AI assistant to find and even contact businesses for them, “can a machine understand this page?” becomes a real question.

A few things worth knowing, so you’re not misled by the hype:

It isn’t scored 0–100. Because the standards here are still brand new, Lighthouse deliberately doesn’t give a weighted score. Instead it shows a pass-ratio — something like 2 of 3 checks passed — plus a pass, warning, or fail on each individual check. It’s a readiness signal, not a grade.

What it checks (per Chrome’s own documentation) includes things like whether your page exposes machine-usable actions (WebMCP), whether its accessibility structure is clean enough for a machine to parse (accessible names, a coherent structure, visible elements), whether the layout stays stable as it loads (Cumulative Layout Shift), and whether the site publishes an llms.txt file. Some of these checks need very recent Chrome versions and opt-in trials to run at all, so the category is early and will keep changing.

It is not (currently) a Google ranking factor. This is the honest, important caveat. Google’s own guidance is that optimizing for AI answers is “still SEO” — the same fundamentals — and Google has said AI-specific add-ons like an llms.txt file aren’t needed for its generative search features. So don’t let anyone sell you an “AI optimization package” as a ranking shortcut. The Agentic Browsing category is a forward-looking readiness check — useful to understand, not a lever that moves you up Google today.

The good news: the things that make you agent-ready overlap heavily with things that were already good ideas — clean accessible markup, a stable layout that doesn’t jump around, and fast, well-structured pages. Build the site right for humans and you’re most of the way to ready for the machines.

How to check your own site

1. Run the classic report. Grade your site or use Google PageSpeed Insights directly — both run Lighthouse and show you Performance, Accessibility, Best Practices, and SEO. Focus on the mobile Performance number first.
2. Fix the red before the orange. A single red Performance score usually costs you more visitors than three orange ones combined.
3. For the AI-readiness category, open Chrome’s built-in Lighthouse panel (DevTools → Lighthouse) on a recent version of Chrome and look for the Agentic Browsing section. It’s newer and not yet surfaced by every speed tool, so a public grader that runs on the PageSpeed API will show you the four classic categories, not this one — for now.

The pattern underneath all of it is the same one we build for: a page that’s genuinely fast, stable, and cleanly structured scores well across the board — for Google, for visitors, and increasingly for the AI assistants doing the searching on their behalf.