Why a Lighthouse Score Is Only Half the Story
Someone just forwarded you a screenshot: a Lighthouse score of 98, emerald green, on a website that loads (they assure you) faster than you can blink. Somewhere on the page there is a statistic about how each extra second of load time costs conversions. There is a ranking. A list. Verified by "real scans."
Web performance matters. Faster sites deliver better user experiences, higher engagement, stronger SEO signals, and improved conversions. Google Lighthouse is one of the most valuable tools available for diagnosing and improving your site, when it is used correctly.
Here is what Lighthouse actually is, what it is designed for, and how to interpret its results responsibly, so you get real value instead of a misleading headline.
TL;DR
- Lighthouse is a powerful lab diagnostic tool, not a real-user performance measure.
- Pair it with field data (CrUX and Core Web Vitals) for the full picture.
- The "20% per second" conversion stat is overstated. Real measured impact lands closer to 2% to 8%, depending on context.
- Scores can vary 10 to 15 points between runs, so always disclose methodology and use multiple runs.
- The Deloitte 8.4% study is real but narrow in scope (enterprise retail, 2019).
- Use Lighthouse to identify issues and track improvements during development and optimization.
The Purpose of Google Lighthouse
Lighthouse is Google's open-source, automated tool for auditing the quality of web pages. It runs a series of tests and generates a report with scores and actionable recommendations across several categories:
- Performance
- Accessibility
- Best Practices
- SEO
- Progressive Web Apps (PWA)
You can run it in Chrome DevTools, from the command line, as a Node module, or through PageSpeed Insights. It simulates a page load under controlled conditions (typically emulating a mid-tier mobile device on a throttled network) and produces detailed diagnostics.
Its primary purpose is diagnostic and developmental: helping developers, designers, and teams identify specific issues and opportunities to improve page quality before or while users experience them. It is excellent for catching problems like render-blocking resources, unoptimized images, or excessive JavaScript.
Definition: Lab Data vs. Field Data
Lab data (Lighthouse) comes from synthetic, repeatable tests in a controlled environment. It is great for debugging and consistent tracking. Field data (CrUX) comes from real-user measurements collected from actual Chrome visitors under real conditions, and it powers Google's Core Web Vitals ranking signals.
A high Lighthouse score is a strong signal of technical health, but it should be read alongside real-user metrics for a complete assessment.
The Zombie Statistic: "Each Second Costs 20% of Your Conversions"
This claim appears in many performance discussions. While speed clearly affects conversions, the specific "20% per second" figure lacks a clear primary source and overstates most controlled research.
Google and SOASTA's analysis of real mobile e-commerce sessions showed an average drop of roughly 4.42% per additional second in the 0 to 5 second range. Real impact varies by industry, baseline speed, and audience, typically landing between 2% and 8% per second.
Pro Tip: Ask for the Source
Demand traceable studies with a stated methodology. Honest claims about the business impact of speed stay compelling without the exaggeration. If a number cannot name its study, treat it as decoration, not data.
Lighthouse Is a Diagnostic Tool, Not a User Experience Oracle
Lighthouse excels at repeatable diagnostics. It helps you:
- Identify performance bottlenecks
- Get specific optimization recommendations
- Track improvements over time in a controlled way
- Audit accessibility, SEO, and best practices in one pass
However, because it runs under simulated conditions, it may not perfectly match real-user experiences (which vary by device, network, location, and behavior). A 98 in Lighthouse can coexist with good or "Needs Improvement" Core Web Vitals in the field, and the reverse happens too.
Lighthouse identifies the problem. Field data confirms the real-world impact.
Understanding Score Variance and Cherry-Picking
Lighthouse scores on the same URL can vary by 10 to 15 points between consecutive runs, thanks to network jitter, server timing, cache state, and other real-world variability. This is normal, not a defect.
Credible reporting uses the median of multiple runs (typically 3 to 5) and discloses the methodology: tool version, number of runs, timestamp, and test conditions.
Expert Tip: Best Practice for Reporting
Always specify the tool (Lighthouse CLI or PageSpeed Insights), the number of runs, the aggregation method (median recommended), the exact timestamp, and whether the cache was warm. Without that context, a score is an assertion rather than a measurement.
Scores Change Over Time, So Treat Them as Snapshots
Web performance is dynamic. A great score today can degrade tomorrow after plugin updates, new third-party scripts, A/B tests, or cache changes.
| What Can Change a Score | Typical Impact |
|---|---|
| Third-party widget added | -8 to -20 points |
| Unoptimized hero image | -5 to -15 points |
| New render-blocking script | -10 to -30 points |
Treat any score as a point-in-time measurement, and re-test regularly.
The Deloitte Study: Real but Contextual
The 2019 Deloitte "Milliseconds Make Millions" study (commissioned by Google) found an 8.4% conversion lift from a 0.1-second mobile speed improvement for large retail and travel brands. It remains valuable directional evidence, but context matters: enterprise sites, 2019 data, and specific verticals.
Pro Tip: Honest Framing
State it the way it was actually measured: "A 2019 Google-commissioned study of large retail brands found that a 0.1-second improvement correlated with 8.4% higher conversions." A few extra words, and the claim becomes defensible.
What Credible Web Performance Work Looks Like
Having catalogued the ways this goes wrong, here is what the trustworthy version looks like in practice.
- Use both lab and field data. Lighthouse for diagnosis, plus CrUX and Core Web Vitals for real-user validation.
- Disclose methodology. Timestamps, number of runs, and test conditions.
- Focus on metrics, not just the overall score. Prioritize LCP, INP, and CLS.
- Re-test regularly. Performance is an ongoing process, not a one-time state.
- Combine with business metrics. Conversion rate, bounce rate, and engagement.
Key Takeaways
- Lighthouse purpose: a diagnostic auditing and improvement tool, excellent for development and optimization.
- Best used with: real-user field data (Core Web Vitals).
- Score variance: normal, so use medians and disclose methodology.
- Conversion claims: use sourced, contextual numbers.
- Ongoing process: regular testing beats a one-time score.
Frequently Asked Questions About Lighthouse Scores
Is a Lighthouse score above 90 good?
Yes. It indicates strong technical performance under simulated conditions. Aim for 90 or higher (the green band), but always validate with real-user field data before treating the score as proof a site is fast for actual visitors.
What is the difference between Lighthouse and Core Web Vitals?
Lighthouse is lab diagnostics: a synthetic page load under controlled device and network emulation that returns a composite score. Core Web Vitals are real-user field metrics (LCP, INP, and CLS) measured from actual Chrome sessions via CrUX, and they are the metrics that influence Google's ranking signal. Use both.
How do I evaluate a web performance vendor?
Ask for both lab and field data, request the methodology behind any score (tool, number of runs, aggregation method, and timestamp), and ask how they plan to monitor performance over time. Look for transparency. A vendor who answers with "it is industry standard" instead of specifics is telling you something.
Final Thoughts
Google Lighthouse is a fantastic tool when it is treated as what it is: a diagnostic powerhouse for making the web better. Combined with real-user data and a sound methodology, it helps deliver genuinely faster, more usable websites.
Speed matters. Done right, with the right tools and realistic expectations, it drives real business results. If you want performance work that shows field data next to lab scores and discloses its methodology instead of glossing over it, our web performance services page is a good place to start.
