ChatGPT is citing Reddit far less: reddit.com's share of ChatGPT Search citations collapsed from about 4% to about 0.5% on August 14, 2026. Daily data for ChatGPT, Google AI Overviews, and Google AI Mode from July 7 to August 17.
While I’m not a fan of scrapping you would think reddit would want this to push traffic to the site. Get a good answer and follow the referral link and you have a new monkey.
I mean just about every metric I’ve seen suggests most websites that see an increase in scraping see a decrease in actual human visitors. I doubt there’s much distinction between chat and search users - the goal of these services is to drop click-through rates.
Oh, definitely. They are reinventing the old portals concept. That’s why I don’t think Reddit actually wants to encourage the scraping (partly also because they sell data themselves to build LLM, but that’s another thing)
I don’t know how many do, but they’re fools if they don’t.
You’d think that since those links are there, that the information on those links matches what that LLM says, and thus not a hallucination, but at least twice, I’ve found the information straight up wasn’t there, or the LLM had misread / misinterpreted the page’s content.
But then I may not be a regular user. I ask a question maybe once every one to two weeks because a regular search finds nothing and, like I say, I prefer to check the sources.
Great question, I wish I knew. I know when I’m on a deep dive I will hit the links as there is other information or other articles the scrapped site has.
I do for things where I want to be accurate, but sometimes it seems like I’m the only one
On the other hand I advised my teen to buy the wrong battery because I didn’t click into the results for “what battery do i need to replace in my Subaru key?”
I think it had the opposite effect. I believe the recently didn’t renew their contract with Google because they were upset people weren’t coming to the site, just getting answers from Gemini. I could be wrong, but I thought I read that
Tonight I spent the time spinning up a silverbullet instance and converting the PDFs to .md files. You’re right, PDFs aren’t searchable and will be a pain as the database grows.
While I’m not a fan of scrapping you would think reddit would want this to push traffic to the site. Get a good answer and follow the referral link and you have a new monkey.
Are there any statistics on how many LLM chatbot users actually click on the source links?
I mean just about every metric I’ve seen suggests most websites that see an increase in scraping see a decrease in actual human visitors. I doubt there’s much distinction between chat and search users - the goal of these services is to drop click-through rates.
Oh, definitely. They are reinventing the old portals concept. That’s why I don’t think Reddit actually wants to encourage the scraping (partly also because they sell data themselves to build LLM, but that’s another thing)
I don’t know how many do, but they’re fools if they don’t.
You’d think that since those links are there, that the information on those links matches what that LLM says, and thus not a hallucination, but at least twice, I’ve found the information straight up wasn’t there, or the LLM had misread / misinterpreted the page’s content.
But then I may not be a regular user. I ask a question maybe once every one to two weeks because a regular search finds nothing and, like I say, I prefer to check the sources.
Great question, I wish I knew. I know when I’m on a deep dive I will hit the links as there is other information or other articles the scrapped site has.
I do for things where I want to be accurate, but sometimes it seems like I’m the only one
On the other hand I advised my teen to buy the wrong battery because I didn’t click into the results for “what battery do i need to replace in my Subaru key?”
I think it had the opposite effect. I believe the recently didn’t renew their contract with Google because they were upset people weren’t coming to the site, just getting answers from Gemini. I could be wrong, but I thought I read that
The guys on LTT mentioned this during their Linux challenge, but I’ve found it to be true -
Scouring forums for answers to obscure errors used to take hours. Now I just throw it at Gemini, troubleshoot in real time and find the fix.
I’ve even started asking it to give me a PDF of the steps we took and filing them away for future me.
I like to ask it for a beautiful html page instead of a pdf. PDFs are a pain later on.
Tonight I spent the time spinning up a silverbullet instance and converting the PDFs to .md files. You’re right, PDFs aren’t searchable and will be a pain as the database grows.
Thanks for sharing :)