Vercel scored 6.0/10.

The page opens with “Build and deploy on the AI Cloud” and then immediately fractures into five parallel tracks — AI Apps, Web Apps, Ecommerce, Marketing, Platforms — each with its own feature pitch. The hero claims a single category (“the AI Cloud”) but the body delivers a product catalog. A developer landing here to solve a specific problem — deploying an AI app, scaling a multi-tenant platform, routing model requests — has to do the work of figuring out which of the five tracks is theirs before they can evaluate whether Vercel solves it. The move is to pick the one buyer this page is built to convert first and build the scroll path around their specific problem; the tab-switching architecture works in a product UI, not on a homepage where the visitor hasn’t yet committed to staying.

The performance metrics in the social proof strip — “build times went from 7m to 40s,” “95% reduction in page load times,” “24x faster builds” — appear without a single named team behind them. These are the strongest numbers on the page and they’re doing almost nothing because the visitor has no way to verify or contextualize them. Vercel’s own footer says “Trusted by the best teams” and the page elsewhere references enterprise customers, which means named proof exists somewhere in the brand’s asset library. Move at least two named attributions — company name, role, specific context — into the proof strip directly beneath the hero, where the visitor is still deciding whether to scroll.

The AI Gateway leaderboard — a live ranked list of top models by usage percentage on June 7, 2026 — is a genuinely differentiated product signal buried in the middle of the page with no framing that explains why it matters to the visitor. The implicit claim is powerful: Vercel sees real-time model traffic across the ecosystem and can show developers which models are actually being used, not just marketed. That’s a category-defining data asset. Right now it reads as a widget. Promote it above the leaderboard with one sentence that names the claim — something like “we route [X] model requests per day across [Y] providers; here’s what developers are actually running” — so the visitor understands what they’re looking at before they look at it.

The page carries two structurally incompatible conversion paths without distinguishing who each one is for. “Start Deploying” is a self-serve, zero-friction action aimed at an individual developer who wants to ship something today. “Get a Demo” and “Talk to an Expert” and “Get an Enterprise Trial” are sales-assisted paths aimed at a team buying decision. Both paths appear at the top of the page, side by side, with no signal to the visitor about which one fits their situation. The self-serve developer reads “Get a Demo” and feels like they’ve landed on an enterprise vendor’s site; the enterprise buyer reads “Start Deploying” and wonders if there’s a real sales process. Separate the paths visually and contextually: self-serve above, enterprise below, with a brief qualifier (“deploying solo?” / “scaling a team?”) so each visitor self-selects without friction.