As AI eats the web, the internet’s collective memory is disappearing
As Google leans into AI-generated answers and SEO spam proliferates, many users report that web search quality has sharply declined, making it harder to find niche, archival or technically precise information. Commenters worry that AI systems trained on the open web now summarize and remix human-written content without sending traffic back, undermining the economic and social incentives to publish original material and accelerating a shift toward AI-generated “slop.” Some see this as a threat to the internet’s collective memory and future knowledge creation, while others argue it may push people toward smaller, curated indexes, personal archives, and alternative search tools.
Search quality and AI summaries
- Many see Google search as steadily deteriorating since late 2000s: aggressive synonyming, refusal to return true “0 results,” ad-heavy layouts, and preference for big brands and “intent/commercial” pages.
- AI summaries on Google are widely criticized as confidently wrong, off-target, and pushing real results below the fold. Some say they now default to AI only because classic results are worse.
- Others report that “AI mode” (Gemini, ChatGPT integrated search, etc.) is currently more useful than raw Google for many tasks, but note this is likely a temporary “golden age” before monetization/enshittification.
Impact of AI on content creation and incentives
- Strong concern that AI assistants answer from scraped content without sending traffic back, destroying ad‑ or reputation‑based incentives to write documentation, tutorials, and blogs.
- Several creators say they’ve stopped or reduced public sharing (code, blog posts) or pulled repos because they don’t want to “train their replacement” or lose credit.
- Counterpoint: some posters say they’ll keep publishing because they value sharing ideas, regardless of whether humans or LLMs read them; others think manuals and official docs will still be produced for product support.
AI slop, training data, and “eating the web”
- Many observe a surge of AI‑generated content farms: thin, plausible but wrong articles now dominate results in areas like health, pet care, recipes, and niche tech.
- Worry that future models will train on prior AI output, amplifying errors and low‑quality patterns; some communities are explicitly trying to poison training data.
- Calls for carefully curated “trusted corpora” and private archives; fear that public internet will become unusable for training or reference.
Preservation and “collective memory”
- Debate over what deserves preservation: some dismiss Instagram stories and casual posts as junk; others note graffiti‑level ephemera is invaluable to historians.
- Broad agreement that link rot, paywalls, DRM, lawsuits (e.g., against Internet Archive), and platform churn are already erasing the record, independent of AI.
- Several advocate personal archiving: PDFs on local disks, tools like Zotero, Hister, Kiwix/OpenZIM, and even private or VPN‑based “closed nets.”
Alternatives and future of search
- Users report mixed experiences with DuckDuckGo, Brave Search, Kagi, Yandex, and meta‑searches (SearXNG, EXA, Tavily). No clear consensus winner; tradeoffs between index completeness, spam filtering, and business models.
- Paid search (e.g., Kagi) is seen by some as aligning incentives toward users rather than advertisers, but it cannot fix underlying degradation of the web itself.
- Some foresee a split between an “AI web” and a smaller, semi‑hidden human web (personal blogs, small forums, mesh/overlay networks).