Six days.
That’s how long it took for one of the most trusted sources in AI search to lose almost its entire footprint. Not because it broke a rule. Not because anyone decided to punish it. It just… stopped being asked. And if you built any part of your visibility strategy on borrowed ground, this week is a warning you don’t want to ignore.
Hi, I’m Jeff Payne. You’re listening to The Jeff Payne Show, Episode 68, where we skip the hype and get straight to what actually helps you as a business owner.
Here’s what happened. For two years, Reddit was the single most-cited source in ChatGPT’s answers — ahead of Wikipedia, ahead of every news publisher, ahead of every review site. If you asked ChatGPT what moisturizer to buy or whether a piece of software was worth it, there’s a good chance the answer traced back to a Reddit thread.In the second week of August, that mostly stopped. Reddit’s share of ChatGPT citations had been holding steady around 3.8% for weeks.
By August 14th, it was down near half a percent. An 86% drop — and it happened in days, not months.Now, here’s where most of the coverage gets it wrong.
The easy story is “Reddit got punished” or “the algorithm turned on them.” Maybe. But the data points somewhere more useful: on August 8th, ChatGPT changed how it searches. Instead of just crawling the open web and seeing what surfaces, it started going directly to named sites and asking them for the answer.
Site-specific queries jumped from under a ½ a percent of its searches to nearly 17% in a single day.
That’s the headline that actually matters to you. Whoever owns the answer is who gets asked. Reddit doesn’t own the answers people post there — the platform hosts them, but no single page belongs to Reddit the way a product page belongs to you.
So when the model started going straight to the source, Reddit had nothing to hand over.
I want to be straight with you about something else, too: the cause isn’t fully settled. There’s a competing explanation that Reddit itself blocked crawler access to protect its licensing deals — which would make this more of a self-inflicted wound than an algorithm shift. And this isn’t even the first time it’s happened; a similar collapse hit Reddit last September, and it recovered.
So don’t build your whole strategy around this week’s headline. Build it around the thing that’s true no matter which explanation turns out to be right.
And here’s the thing: your own website just became retrieval infrastructure. Not a brochure. Not a place people land after they already trust you. The place the model goes to check if you’re the answer.
Independent tracking firms — not just one, several, cross-checking each other — confirm the same pattern: total searches per ChatGPT answer nearly doubled the same week site-specific queries spiked. That’s not a substitution. That’s the model doing more direct work per question and directing that work toward named domains rather than the open web.
Translate that into plain terms: a thin category page, an FAQ that doesn’t actually answer the question someone would ask, a product page written in your internal vocabulary instead of your customer’s — those pages were always weak. Now they’re invisible in a way they weren’t three weeks ago.
The move isn’t chasing this week’s algorithm story. It’s auditing your own site the way a model would — not the way you would.
Open your site and read it looking for the answer to one specific customer question. Not your question. Theirs. You’ll likely find pages written in your language instead of the words your customer actually searches — product names, service tiers, internal terms nobody outside your building uses.
The person asking doesn’t know any of that. They only know the problem they’re trying to solve.
Every page that meets them in their language becomes a page worth pulling from. Every page written in yours stays invisible — to the customer and to the model, at the same time. That’s not a technical fix. It’s unglamorous, cheap, and entirely under your control, which is exactly why most brands will skip it.
Here’s the question worth sitting with: if a customer typed their actual problem — not your product name, not your category, just the problem — into an AI search box right now, would your website be the page it goes to?
If you’re not sure, that’s the whole episode in one sentence.
For two years, Reddit was the most-cited source in ChatGPT’s answers — ahead of Wikipedia, ahead of every publisher, ahead of every review site. In the second week of August 2026, that mostly stopped. What happened next tells you less about Reddit than it does about what your own website needs to become.
What Actually Happened
Reddit’s share of ChatGPT Search citations held steady at an average of 3.8% for weeks. By August 14th, it had fallen below 1% — averaging roughly 0.5% in the days that followed. That’s an 86% relative drop, and it happened in under a week, not gradually over months.
The timing lines up with a separate, verifiable change: on August 8th, ChatGPT’s background search behavior shifted.
Site-specific queries — where the model asks a named domain directly rather than crawling the open web — jumped from under 0.5% of its searches to nearly 17% in a single day. Total searches per answer also nearly doubled over the same window.
Whoever owns the answer is who gets asked.
It’s Not (Just) About Reddit
The easy narrative is that Reddit got penalized. The more useful read is structural: when a model starts asking named sites directly for answers, it favors whoever owns the answer. Reddit hosts millions of individual posts, but it doesn’t “own” the answers the way a business owns its own product pages, pricing, or documentation.
It’s worth being direct about what’s still uncertain here. A competing explanation — one that surfaced days after the initial reporting — points to Reddit itself restricting crawler access to protect its data licensing arrangements, which would make this a self-inflicted change rather than an algorithm shift.
And this isn’t unprecedented: a similar citation collapse hit Reddit in September 2025, and it recovered. The specific cause matters less than the pattern it reveals.
Don’t build your strategy around this week’s headline.
Build it around the thing that’s true no matter which explanation turns out to be right.
This is one of the best written, most informative email newsletters I’ve ever read!
Thank you for sharing your expertise, and hope you’re doing well.
Beth Zimmerman
Founder, Executive Director
Your Website Just Became Retrieval Infrastructure
Whatever caused Reddit’s drop, the practical shift for every business owner is the same: if ChatGPT is increasingly going straight to named domains for answers, then whether your site is crawlable, indexed, and structured to actually answer the questions people ask is no longer a background technical detail. It’s the infrastructure the model relies on to decide whether you’re the answer.
A thin category page, an FAQ that doesn’t answer the actual question someone would type, or content written in internal company language instead of customer language — these were always weaknesses. They’re now weaknesses with a direct, measurable cost in AI visibility.
Your own website just became retrieval infrastructure — not a brochure.
The Move: Read Your Site Like a Model Would
This isn’t about chasing this week’s algorithm news. It’s about a standing discipline: periodically reading your own site the way a model — or a customer who doesn’t know your internal vocabulary — would read it, looking for whether it actually answers a specific question in the words the person asking would use.
The pages that meet customers in their own language are the pages worth pulling from. The pages written in internal shorthand — product names, service tiers, category jargon — stay invisible to the customer and the model at the same time. That gap is inexpensive to close and entirely within a business’s control, which is exactly why most companies leave it unaddressed.
If a customer typed their actual problem — not your product name, not your category — into an AI search box right now, would your website be the page it goes to? That question is worth sitting with regardless of what happens to Reddit’s citation share next month.
A page written in your language rather than your customer’s remains invisible to both the customer and the model.
Promptwatch data: GEO analytics platform; citation-share tracking of ChatGPT Search, Google AI Overviews, and Google AI Mode, July 7–August 17, 2026. Reported on by Search Engine Journal and Axios.
Episode framing inspired by Camille Moore, “Why ChatGPT Stopped Citing Reddit.”, Branding With Benefits (Substack).
Note: Causation for the Reddit citation drop is not fully settled as of publish — a competing explanation holds that Reddit blocked its own crawler access via robots.txt. A similar citation collapse occurred in September 2025 and reversed.
COMPLETE THE FORM TO
BOOK A STRATEGY CALL
"*" indicates required fields
COMPLETE THE FORM TO
BOOK A STRATEGY CALL
"*" indicates required fields
Subscribe and Share – WE APPRECIATE YOUR SUPPORT
