Every agency says its website is its best salesperson. We wanted to know how good those salespeople really are, so we measured 771 of them against the same 36 criteria, our own site included. The headline: design is the easy part, and trust is where most of the field, and we, lose points.
What we did
We started from the 2025 Clutch 1000 list of agencies. Of the 1,000 websites, 807 answered at all; 646 of the addresses on our list were generated from company names, and 179 of those were dead. After removing blocked or empty captures, 771 sites had complete scores.
- Six criteria are measured by script: mobile speed (Lighthouse), accessibility (Lighthouse and axe, WCAG 2.2 AA rules), on-page SEO, structured data, AI-search readiness, and language support.
- Thirty criteria are judged from screenshots and crawl data: ten for looks, ten for usability and conversion, six for proof and trust, four for offer and expertise. Each is scored 0 to 20.
- Claude scored every site. Then ChatGPT re-scored the top 40 and our site blind, and both judges ranked the same sites head to head in groups of six.
What the field looks like
- Slow on phones. The median mobile Lighthouse performance score was 56 out of 100, and only 9% of sites scored 90 or more. Median largest contentful paint was 7.2 seconds under the throttled mobile profile Lighthouse uses.
- Accessibility gaps are normal. 88% of sites had at least one serious or critical automated accessibility finding on desktop. The median Lighthouse accessibility score was 92.
- Quiet about price. Pricing and engagement transparency had a median score of 5 out of 20. Almost nobody shows a range, a model or a typical timeline.
- Few tools for the buyer. Interactive decision aids, such as estimators or quizzes, had a median of 4 out of 20. Security and compliance stories had a median of 3 out of 20.
- One language. Only 13% publish hreflang for more than one language, and only 16% mark up an FAQ in structured data.
- Almost everyone is open to AI search. 5% block AI training crawlers in robots.txt, and fewer than 1% block AI search crawlers.
Design gets a site noticed. Proof is what gets a stranger to book a call.
What our own site taught us
On the 22 criteria that make up a web design and UX index, our site scored 304 of 440 and ranked 11th of 772. On speed and accessibility measures it ranked near the top. On proof and trust it ranked 571st. We had almost no client logos, no named testimonials, no awards and no third-party reviews on the site. A buyer who lands on it sees what we built for ourselves, not what we did for others.
The judges also caught real bugs: stat counters that showed wrong numbers mid-animation, a mismatch between “four products” and “eight apps”, low-contrast labels, and a placeholder where a product screenshot belonged. We fixed those the same week.
Limits you should know about
- Both judges are AI models, and one of them (Claude) also helped build our site. The two disagreed on exact ranks for the top sites, with a Spearman correlation of 0.40 on absolute scores. We trust the broad bands more than any single rank.
- The Clutch 1000 ranks agencies by client reviews, not by design quality, so it is not a list of the best-designed sites on the web.
- Screenshots are stills, so motion and micro-interactions are under-scored for every site.
The scoring rubric and the per-attribute results are available on request. If you want your own site measured the same way, book a call.