As AI eats the web, the internet’s collective memory is disappearing
TL;DR Highlight
As AI summaries replace original pages, search and web archives are both weakening.
Who Should Read
A backend developer who uses AI search summaries or web data as source material for RAG answers
Core Mechanics
- AI summary got sunset time wrong
- Google's new AI summaries invented sunset times, and a user in Colorado Springs was told the sunset had already happened.
- The author uses this to argue that search now struggles even with basic facts.
- AI blocks the way to the source
- The author says an error-prone AI between users and sources can make an existing page practically undiscoverable.
- 404 Media reports that companies plant content on Reddit to influence AI search answers.
- FiveThirtyEight's archive was deleted
- Disney owned the site and, by March 2025, the remaining staff had been laid off.
- Disney no longer saw it as an active asset and deleted the entire archive.
- Wikipedia loses traffic too
- AI systems scrape Wikipedia and present its content directly, so users don't click through.
- Less traffic means attention and donations no longer flow back.
Community Reactions
- Docs are hard to find
- One commenter said that when writing a Python script, they ask AI about a library's API because findable documentation is gone.
- They called documentation dead.
- AI cited another AI's hallucination
- One commenter asked Claude about Roman history and got a fact they knew was wrong.
- When challenged, Claude said it came from an Encyclopedia Britannica page flagged as an AI-generated summary, and admitted the fact was not historically supported.
- AI summaries are bad, AI mode is better
- One commenter said search summaries often misread the query and push results down the page.
- They said switching to AI mode gives much better results.
- A counterpoint on usefulness
- One commenter said Google search is the most useful it has been in years because of AI mode.
- They added that you need to know AI can be wrong.
How to Apply
- Filter AI-generated pages out of RAG sources
- Today, flag AI summary pages and pages with unclear provenance in your RAG source list.
- Log the domain and collection date of every document cited in an answer.
- Snapshot key source documents yourself
- When you ingest an external document used as evidence, store a copy in your own storage.
- Record the URL, collection time, and copy path so you can verify the evidence if the link dies.
- Don't use AI summaries in place of sources
- Attach original links to the answers you show users.
- For fact-sensitive questions, add a source verification step instead of returning only a summary.
Try It
# Paste into Claude or ChatGPT (a simple check prompt applying the article's point)
Below is the list of documents my RAG system used as evidence for its answers.
{{document list: URL, title, collection date, excerpt}}
For each document, make a table covering:
1. Whether the list shows any clue that the page may be AI-generated or AI-summarized
2. Whether the original source is clear
3. Whether the collection date is old enough that the link may have died
Don't guess about anything not in the list; write 'unknown'.
Finally, rank which documents I should archive copies of first.Terminology
Related Papers
You could have invented transformers from memory
Viewing linear, MLP and attention layers as memory stores: capacity goes ~64 to ~2,700 to over a million
Pointing AI at archives found a forgotten meteorite, lost rhinos, and more
Embedding search plus a cheap filter model surfaced unrecorded events in 400 years of archives
EmbeddingGemma 2: An open, lightweight multimodal embedding model
One open 740M model embeds text, images, audio and video into a single shared vector space
RAG-NAROK: Retrieval-Aware Knowledge Corpus Poisoning in RAG with Source-specific Refutation
Refutation documents targeting retrieved sources took top rank and evaded anomaly detectors up to 88.37%.
Text, Pixels, or Both? Evaluating Input Representations for Multimodal Document QA
Page images beat OCR text on accuracy but cost more; a question-only router cut latency 30%.
Microsoft exec called AI scraping 'the largest theft of labor in human history'
NYT lawsuit unredacted filings reveal Microsoft and OpenAI internal views on scraping