Cursor scored 7.4/10.

The page claims two buyers simultaneously and converts neither cleanly. The hero and download CTA target an individual engineer who wants to try the tool now — “Download for macOS” is a self-serve, zero-friction action. But the body copy, the “Request a demo” CTA, the Fortune 500 claim, the SOC 2 badge, and the “Explore enterprise” link all signal an enterprise procurement path that requires a champion, a security review, and a budget approval. These are not the same buyer, and they don’t make the same decision in the same session. The individual engineer who lands here sees “Trusted by over half of the Fortune 500” and reads it as a product for big companies — not for them. The enterprise buyer who lands here sees a macOS download button and no clear path to a sales conversation above the fold. Fix this by splitting the conversion surface: put the download path first and unambiguously for the individual engineer, then give the enterprise buyer a distinct section with a named path to the demo — not a secondary CTA buried below the testimonials.

The testimonials are the strongest asset on the page and they’re doing the wrong job. Jensen Huang saying 40,000 NVIDIA engineers use Cursor, Patrick Collison describing thousands of Stripe employees, Greg Brockman calling it the place where GPT-5 shines brightest — these are enterprise adoption signals, not individual conversion signals. They answer the question “is this safe to roll out at scale?” not “will this make me faster today?” But the page’s primary CTA is a personal download. The buyer who’s deciding whether to download and try it doesn’t need to know NVIDIA uses it; they need to know it will make them noticeably faster in the first hour. The page has the wrong proof for the action it’s asking the visitor to take. Move the Karpathy and shadcn quotes — both of which describe the product experience directly — above the fold or into the first scroll, and push the Huang and Collison quotes to the enterprise section where they belong.

The page names four distinct differentiation angles — autonomous agents, specialized Tab autocomplete, codebase understanding, and multi-model flexibility — and defends none of them. Each is introduced with a short label and a “Learn more” link, then abandoned. GitHub Copilot has autocomplete. Claude has multi-model options. Codeium has codebase indexing. The page lists the same features the category already offers without making a claim about why Cursor’s version is different. The one thing Cursor could own that competitors can’t easily match is the claim that the agent runs autonomously on its own computer, builds and tests end-to-end, and hands you a finished feature to review — that’s a genuinely different category claim, not a feature delta. The page buries it in the third section. Move the autonomous agent claim to the hero, make it the single thing the page is fighting for, and let every other feature serve as supporting evidence for that claim.

The changelog is present and public, which is a trust signal for developer buyers — but it’s positioned as a footer element, after the testimonials and blog highlights, where most visitors won’t reach it. Developer buyers, especially those evaluating against GitHub Copilot or Codeium, want to know the product is actively maintained and moving fast. The changelog entries visible on the page — PR Review, Cloud Agents, Cursor in Microsoft Teams, Bugbot Effort Levels — show a team shipping weekly. That velocity is a competitive advantage that the page treats as an afterthought. Pull the changelog strip up to immediately after the feature sections, frame it as evidence of shipping speed, and let the dates speak for themselves.