Prompt
How do I tell which pages LLMs are pulling from?
Latest observation
Nov 8, 2026
OpenAI APIWeb search: off
There isn’t a reliable, universal way to see the exact web pages an LLM “pulled from,” because it depends on how the model was used:
1) If it was a normal chat model
Most LLMs do not search the web in real time. They generate answers from patterns learned during training, so there may be no source pages at all.
2) If it was a search/browsing-enabled model
Then you may be able to see sources, but only if the product exposes them. Look for:
- citations / footnotes / source links
- a “sources” or “references” panel
- the model stating which documents it used
- logs in your app if you built the retrieval system
3) If you’re using RAG or your own app
Then you can usually track sources directly by logging:
- the document ID
- URL
- chunk text
- retrieval score
- timestamp
That’s the most reliable way to know what it used.
4) If you want to infer likely pages
You can:
- ask the model to cite sources
- compare the answer against likely search results
- search key phrases from the answer in quotes
- use plagiarism/origin-checking tools cautiously
But this only gives likely matches, not certainty.
Best practice
If you need traceability, build or use a system that:
- retrieves from a known corpus
- returns citations with every answer
- stores retrieval logs
If you want, I can show you:
- how to check sources in ChatGPT or another specific LLM, or
- how to build source tracing into a RAG app.