47/100
Better than 19% of sites we have audited
40–49 · Something to build on — There's real substance here, with the gaps spread widely enough that the order you tackle them in matters.
165 sites audited. Median 57, and the top 10% start at 72. Most land in the 50s.
"CrowdGPT has the technical depth to win over developers and the mission to build a passionate community — sharpen the trust signals and performance, and this could become the rallying page for the distributed AI movement."
Scores and fixes by category
Weakest categories are open. The score and its reasoning are the AI's reading; the fixes underneath come from checks we ran against your page.
› Value Prop 20% of score 1 fix 60
The headline 'Train a language model with your GPU' is clear and specific, and the subheadline explains the mechanism well, but the benefit to the contributor (ownership, recognition, advancing open AI) is not front-and-center.
A 30–60s product demo dramatically outperforms static screenshots for comprehension and conversion on complex products.
Do this: Add a silent autoplaying product demo loop, or a 60-second explainer with captions, near the hero CTA.
Industry benchmark: +86% conversions — EyeView
› Trust 15% of score 4 fixes 10
Zero testimonials, zero contributor counts displayed (the dashboard fields exist but show dashes), zero partner logos, and a broken contact email link — there is nothing for a skeptical visitor to anchor trust on.
Generic "trusted by" claims aren't trust — named testimonials with real attribution are. Anonymous quotes read as marketing copy.
Do this: Add at least one testimonial with real attribution (full name plus role or company). A single specific, outcome-focused quote outperforms a grid of generic ones.
Industry benchmark: +34% conversions — Nielsen
This is about your CUSTOMERS' logos, not your own. A row of recognisable companies who use you is the fastest credibility signal a visitor can parse, because it borrows trust they already have.
Do this: Add a band of 3+ customer or partner logos with a plain heading like "Trusted by". If you cannot name customers yet, use integrations you support, or skip the band entirely rather than filling it with placeholders. Give each logo real alt text — a logo with alt="" is invisible to screen readers and to us.
Industry benchmark: +34% conversions — Nielsen
The page makes no claim about anyone else using it. A first-time visitor has nothing to go on but your own description of yourself.
Do this: Add one concrete, checkable number in the hero — active users, projects shipped, or customers served. A specific figure beats an adjective.
Industry benchmark: +34% conversions — Nielsen
Every additional second of load time measurably increases bounce rate — 1s→3s nearly doubles it.
Do this: Compress images, enable HTTP/2 or HTTP/3, add CDN caching, and remove unused JS/CSS.
Industry benchmark: -32% (1s→3s load) bounce rate — Google
› Design / UX 10% of score 54
Mobile-friendly with logical navigation sections (home, install, protocol, status, leaderboard, contact), but a mobile LCP of 7.5 seconds and a CLS of 0.314 will cause real layout-jump pain on phones.
› Copy 15% of score 62
The protocol description is admirably precise and technical, but the page leans feature-first (architecture specs, CLI flags) with almost no outcome-focused copy about why a contributor should care beyond the mechanics.
› CTA 15% of score 65
The primary conversion actions — 'Install the client' and OS-tab install instructions (Linux/macOS/Windows) — are clear and well-structured for a developer audience, though they read as setup docs rather than a motivating call to action.
› Technical 10% of score 40
Desktop performance is excellent (96/100, LCP 1.3s), but mobile performance is poor (56/100, LCP 7.5s, CLS 0.314), and two email-obfuscation links in the footer return 404 errors.
› Differentiation 10% of score 40
The 'model belongs to everyone who helped build it' angle is genuinely unique but is buried mid-page; the hero doesn't lead with this ownership/community differentiator.
› Emotional 5% of score 32
The collaborative ownership angle — 'the model that gets trained belongs to everyone who helped build it' — is inspiring but underexploited; most of the page reads like a technical README.
Technical Analysis
Agentic Browsing
New Google · experimentalCan an AI agent read and use this page? ChatGPT, Claude and Perplexity increasingly visit sites on someone's behalf — this is Google's new Lighthouse check for whether yours is usable when they do.
- ✗ Accessibility tree is not well-formed
- ✗ Cumulative Layout Shift — 0.314
Scored by Google and reported alongside your roast, not folded into it. The category is still changing.
Core Web Vitals
How long until your main content (hero image, headline) shows up. Slow = visitors bounce before they see what you offer.
How much your page jumps around while loading. Jumpy pages feel broken and cause accidental taps on the wrong thing.
How long until anything appears on screen. A blank white page for 3+ seconds loses impatient visitors instantly.
How fast your page responds when someone clicks or taps. Slow response feels broken — people click again thinking it failed.
How fast your hosting server starts sending the page. Slow TTFB means the problem is your hosting, not your images or design.
How long until your main content (hero image, headline) shows up. Slow = visitors bounce before they see what you offer.
How much your page jumps around while loading. Jumpy pages feel broken and cause accidental taps on the wrong thing.
How long until anything appears on screen. A blank white page for 3+ seconds loses impatient visitors instantly.
How fast your page responds when someone clicks or taps. Slow response feels broken — people click again thinking it failed.
How fast your hosting server starts sending the page. Slow TTFB means the problem is your hosting, not your images or design.
Top Speed Opportunities (Mobile)
- Initial server response time was short
- −2ms
Technical SEO Files
These files tell search engines and AI crawlers how to read your site. Missing ones mean less organic traffic.
- robots.txt may be blocking all crawlers with 'Disallow: /'
- No Sitemap directive found in robots.txt
- No sitemap.xml found - search engines may miss pages
- No llms.txt found - consider adding one to guide AI crawlers (emerging best practice)
- No security.txt found - good practice to help security researchers
Accessibility (WCAG 2.1 static audit)
How usable your page is for visitors with disabilities (~15% of users). Low score = lost customers and legal exposure.
-
heading-order minor
Heading level skipped: h1 → h3. This confuses screen reader navigation.
1.3.1
-
heading-order minor
Heading level skipped: h1 → h3. This confuses screen reader navigation.
1.3.1
-
skip-nav minor
No skip navigation link found. Keyboard users must tab through all nav links to reach content.
2.4.1
-
landmark-regions minor
Missing landmark region(s): main. Landmarks help screen reader users navigate.
1.3.1
Tech Stack Measured
What your site is built with. Matters because your CMS/framework/hosting choice caps how fast you can fix things and what optimizations are even possible.
Couldn't confidently identify a CMS or framework. Either it's custom-built, heavily obfuscated, or behind a reverse proxy.
Page Structure
- Hero
- Good
- Headline
- Good
- Subheadline
- Good
- CTA
- Weak
- Social Proof
- Missing
- Features
- Good
- Pricing
- Missing
- FAQ
- Good
- Footer
- Weak
What Works
- The headline is refreshingly direct — 'Train a language model with your GPU' immediately tells a developer exactly what this is, no jargon fog required.
- The security/trust explanation ('A note on trust' section) is thorough and proactive — addressing the 'is this safe?' objection before it becomes an exit is smart conversion design.
- The FAQ section answers the five most likely objections (data safety, disk space, AMD/Apple Silicon support, CPU-only, poisoning prevention) with specific, credible detail — this is genuinely useful and reduces hesitation.
- The install UX with OS-tab switching (Linux/macOS/Windows) and copy-paste commands is frictionless for the developer audience — getting from 'interested' to 'running' in under 60 seconds is a real strength.
What Doesn't
Suggested Rewrites AI reading
Your idle GPU can train the world's first community-owned language model.
Pool your idle GPU with contributors worldwide to train an open 1B-parameter model that nobody owns and everyone can use — no datacenter, no gatekeeping, no proprietary weights.
Install the client — it's a single Python script, runs in under a minute, and your first contribution starts automatically.