Show HN: Cut the crap – remove AI bullshit from websites

Original Article ↗ Hacker News Discussion ↗

A prototype web tool promises to strip articles of clickbait, SEO filler, and visual clutter by fetching a URL and using an LLM to return a concise, text-only summary—something many users find especially appealing for bloated recipe and news sites. Commenters debate whether this “AI to fight AI slop” approach meaningfully improves the web or just produces shorter, less reliable AI-generated content, raising concerns about hallucinations, loss of nuance, and lack of disclosure about the summarizer. Others question the broader impact on publishers’ incentives and suggest alternatives such as better ad blocking, high-quality paid sources, or browser-native reader modes rather than full AI intermediation.

What the tool does

  • Takes a URL and returns a condensed version of the page, aiming to strip “clickbait, filler, and AI-generated SEO sludge” and keep core facts.
  • Positioned as similar in spirit to browser Reader Mode, but more aggressive and LLM-powered rather than just DOM-cleaning.
  • Current prototype has a ~2000-character input limit to control costs and lacks browser integration.

Reception and main use cases

  • Many commenters like the concept, especially for:
    • Recipe sites bloated with stories, ads, and videos.
    • News sites with cramped layouts, autoplay videos, and intrusive UI.
    • Over-marketed SaaS/landing pages where the value prop is unclear.
  • Some find the outputs “RSS-like” and pleasant for terminal browsers or RSS-style consumption.

Accuracy and limitations

  • Mixed results:
    • Works “perfectly” on some recipe/news pages, giving just the needed info.
    • Fails or misleads on others: e.g., local news pages where it pulls in unrelated link titles, marketing sites where it produces generic domain knowledge, or misinterpreted real-estate examples.
    • On some inputs it even gets longer or drops most of the article.
  • Several commenters doubt LLM summaries in general, calling them lossy, shallow, or “AI slop.”

AI vs AI and broader concerns

  • Strong irony noted: using an LLM (“bullshit generator”) to remove LLM-generated and SEO filler.
  • Some see this as a wasteful arms race: AI inflates content, other AI compresses it again, adding energy use with little net value.
  • Others argue it’s still useful, akin to grepping logs or compressing lawyerly legalese into more digestible text.

Impact on creators and the web economy

  • Debate over incentives:
    • Critics worry this disincentivizes content creation and reproduces Google’s “answer without clicking” problem.
    • Defenders say users aren’t obligated to consume content in ad-friendly form and that much current “creator” output is low-value anyway.
  • Nostalgia for a pre-ads internet where people wrote for joy and expertise, not monetization.

Implementation details and integrations

  • Reverse-engineered prompt appears to be very simple (“Condense… remove bloat, clickbait, scaremongering.”), using GPT‑3.5.
  • Requests for:
    • A URL-parameterized API and larger context window.
    • Browser extensions (Chrome/Firefox/Safari, Arc-like, Safari Reader-style) and possibly RSS-bridge integration.
  • Some prefer in-page transformation (like uBlock/reader mode) over off-site summarization to keep visiting original pages.