A landing page is the hardest 800 lines a startup writes.

Walk through the launches on Product Hunt this morning. Twelve of the first fifteen pages will share a gradient. Eight will open with a sparkle pill and the words "AI-Powered". Eleven will use Inter as the only font. Seven will offer the same two CTAs side by side: "Get Started" and "Learn More". This is not a coincidence. It is the average of the training set, rendered.

Lovable, v0, and Bolt are good at building whole applications. Auth, dashboards, data tables, marketing — all of it, in one prompt. The trade-off shows up in the marketing page, because a generalist that has ingested 200,000 React templates emits the median of those templates. The median is the slop. Specialists win the moment that happens, and it has happened.

So we narrowed the surface. Counterslop builds and audits one artefact: the landing page. Nothing else. A page is graded against 13 named tells before it ships. Tell 1 is a blue-purple gradient hero. Tell 2 is a sparkle eyebrow. Tell 3 is the 85 phrases on the ban list — every entry an empty superlative, a decade-old cliché, or future-tense launch copy that ages in a quarter. Tell 4 is Inter as the only font. Tell 5 is the three identical icon cards. The rest are in the kit. If a page trips any tell, the page does not ship until the writer fixes it.

The rubric is not opinion. It is taken from what works: Resend's noun-headline, Linear's restrained motion, Granola's serif display face, Marc Lou's CTA law that no button reads "Get Started". Every rule traces to a page that shipped, earned attention, and converted. The 85-phrase ban list is the inverse: every entry is a phrase that appeared on more than 30 pages we scraped in the last 12 months, and on none of the 11 reference sites we admire.

This is the bet. A landing page is the highest-leverage 800 lines of code in a company. If we own the segment that says "the page that sells your product will not look like the other 14,000 AI pages this year", we have a real product. We do not need to also build your dashboard. We do not want to.

The kit is open source under MIT. Audit your own pages with it tomorrow. The hosted version runs the same audit on every deploy and refuses to ship pages that score over 15. No model on its own knows what restraint looks like. Restraint is a rule set. We wrote one, and we will hold our own pages to it first.

The pages we ship will not look the same as the others. That is the entire pitch.

— Ravindra Kumar, founder