You Used to Show Up in ChatGPT. Now You Don't.

On August 14, 2026, Reddit's share of ChatGPT's citations fell from an average of 3.83% to 0.52% in four days — an 86% collapse — and nobody at OpenAI told Reddit why. Not a heads-up, not a changelog entry, nothing. Reddit had a signed licensing deal with OpenAI at the time. Its stock dropped roughly 10% in an afternoon session as the tracking data circulated. If a platform can do that to a company with a paid content agreement and a direct line to the people making the change, it can absolutely do it to a business with neither — and most businesses that lose AI visibility never get an explanation either. They just notice, months later, that the citations stopped.

That's the actual problem this piece is about. Not "how do I get cited by AI" — you've probably already been cited, once, before whatever happened, happened. The real question is diagnostic: something changed, and you need to know whether it was something you broke, something a competitor did better, or something a platform changed on its own without asking anyone. Each of those has a completely different fix, and treating them as the same problem is how businesses spend months rewriting content that was never the issue.

Four Days, No Warning, No Explanation

The Reddit collapse is worth sitting with because it's the cleanest documented case of the third category — a platform quietly changing its own evaluation criteria — that exists right now. According to citation-tracking data reported by Search Engine Journal, Reddit held steady at an average 3.83% of ChatGPT citations between July 18 and August 7, 2026. By August 14–17, that had fallen to 0.52%. There's a partial explanation and it doesn't fully add up: on August 8, OpenAI changed ChatGPT Search's fanout behavior, and the share of queries using a site-specific search operator jumped from 0.37% to 16.8% in a single day, while the average number of fanout queries per response nearly doubled. Promptwatch co-founder Klaas Foppen read that as ChatGPT deliberately targeting specific sites rather than casting broadly across the open web. The problem, as the same reporting points out, is timing — the change landed August 8, and the cliff didn't show up until August 14, a six-day gap nobody's fully accounted for.

A separate write-up adds the detail that should worry anyone treating an AI-citation percentage as a stable metric: Ahrefs measured Reddit's citation share at 16.7% around the same period, while an earlier Semrush analysis had it as high as 60% at its peak. Those aren't rounding differences. They reflect different prompt sets, different account states, and different definitions of what even counts as a citation — which means before you panic about a number dropping, you have to ask whether the number ever meant what you thought it meant in the first place.

It's Happened Before, and It Will Happen Again

This wasn't even Reddit's first time. Eleven months earlier, growth advisor Kevin Indig traced a similar Reddit citation collapse to Google quietly removing the num=100 search parameter in September 2025 — a change that broke how several third-party data pipelines, including the ones feeding some AI search tools, could see deep search results, where Reddit threads often rank. Neither collapse had anything to do with Reddit's content getting worse. Reddit didn't publish less, didn't break a redirect, didn't let a competitor out-write it. A parameter got removed on one platform, or a retrieval strategy got adjusted on another, and citation share moved by double digits overnight.

The reason this matters for a business with none of Reddit's scale or leverage: if OpenAI can do this to a company holding a signed licensing agreement with no advance notice, it will absolutely do it to you with no notice, no agreement, and no press coverage to tip you off. The only way you find out is by actually checking — which is a different discipline than watching your Google Analytics dashboard, because none of this shows up there at all.

Suspect One: Something Broke in a Migration

Now to the causes that actually are within a business's control, starting with the most common one. A documented case study covered by Search Engine Land traces a 90% traffic collapse at a multinational cryptocurrency news publisher back to a January 2022 domain migration — moving from a country-code path to a new subdomain structure — that never properly consolidated authority or handled redirects. Google ended up splitting its crawl budget between the old and new structure instead of transferring it cleanly. By January 2023, the Brazilian version of the site alone had 513,369 pages sitting in "crawled — currently not indexed," and soft 404 errors — pages returning a normal 200 status while containing effectively no content — were climbing fast enough that Google's crawl rate on the French version of the same site dropped from 60,000–70,000 requests a day to 20,000–30,000. Daily clicks on the Brazilian site fell from a 15,000–25,000 range before the migration to 2,000–4,000 after — a 90% year-over-year decline that the team only fully understood after pulling Google Search Console's indexing reports domain by domain and forecasting what traffic should have looked like against what it actually did.

Nobody decided to publish worse content. Nobody's writing got sloppy. A URL structure changed, redirects didn't cover everything they needed to, and the compounding effect — soft 404s eating crawl budget, crawl budget starvation slowing indexing, slow indexing meaning fresh articles took 24 hours to even appear — quietly strangled a business that was, by every content-quality measure, unchanged. For AI citation specifically, the mechanism is even more direct: if a model's retrieval step can't find your current URL, or finds an old one that now 404s, it simply won't cite you, no matter how good the page underneath used to be.

Suspect Two: A Competitor Just Published Something Fresher

Sometimes nothing broke at all — a competitor's page just got more current than yours, and current is doing more work in AI citation than it ever did in traditional ranking. AirOps' analysis of more than 4,000 pages cited by ChatGPT across 900 high-intent queries in 15 industries found more than 70% of cited pages had been updated within the past 12 months. For commercial queries specifically, 60% of citations came from content updated in the last six months, and pages that hadn't been touched in over a year were more than twice as likely to lose their citation entirely — for commercial intent queries, a stale page loses over half its chance of being cited at all.

Separate research from Quattr on citation "half-lives" — the time it takes for a page's citation frequency to drop by half — found ChatGPT's decay is the fastest of any major platform at roughly 3.4 weeks, against 5.7 to 5.8 weeks for Perplexity, nearly 70% longer. Most pages peak in citations within 30 to 90 days of publication and, without an update, approach near-total disappearance by the one-year mark. None of that requires your page to have gotten worse. It requires a competitor's page to have gotten newer, more specific, or better structured while yours sat still — and AI systems, unlike Google's index, don't give old-but-authoritative content much benefit of the doubt against a fresher answer to the same question.

Suspect Three: The Platform Changed What It Rewards

The fourth pattern sits between the other three — not your fault, not exactly a competitor's win either, just the ground shifting under an entire category at once. HubSpot is the clearest documented case of this at scale. Its own data, disclosed in April 2026, showed customers seeing a 27% year-over-year organic traffic decline, with organic click-through rates on queries triggering Google's AI Overviews falling from 1.76% to 0.61% — and even queries without an AI Overview saw organic CTR drop 41% year-over-year since the feature's May 2024 rollout. CEO Yamini Rangan put it plainly: "How buyers search is fundamentally changing. They are asking questions in places like ChatGPT and Gemini, and the companies that show up in those answers are already winning."

HubSpot's own blog traffic tells the sharper version of that story. One detailed breakdown puts the drop at roughly 75% between March 2023 and January 2025, accelerating hard after Google's December 2024 core and spam updates, which shifted weight away from the kind of broad, generic content HubSpot had built its blog around — the analysis specifically calls out pages like famous-quotes roundups and resignation-letter templates as exactly the topical-depth-light content that update targeted. Here's the part that actually matters for diagnosis, though: a separate analysis found HubSpot still holding a 35.3% share of voice in AI-generated answers for its category, cited in 95–100% of top prompts related to CRM. The traffic collapsed. The AI citation didn't — because those are genuinely different measurements, and a business tracking only one of them would draw exactly the wrong conclusion about what's actually happening to it.

What Site Owners Are Actually Arguing About Online

It's worth reading how people who actually run these sites are diagnosing this in public, rather than only the vendor write-ups. A Hacker News thread on tech publications losing over half their Google traffic since 2024 turned into a genuine argument about causes, and nobody in it agreed. A commenter using the handle aboutplants blamed the sites themselves — pages "bombarded with pops, ads everywhere, autoplaying video" — arguing readers left because the experience got worse, not because AI took anything. Another, npilk, reached for a literary comparison, describing a "vicious cycle" where departing readers force more desperate monetization, which drives away more readers.

tracerbulletx pushed back hard on the ads theory with a personal data point: their own ad-free tech blog "lost more than half the traffic" too, which they took as evidence the cause runs deeper than monetization choices — AI systems extracting value from published content without sending anything back. doug_durham countered that the aggressive-ads pattern comes from financial desperation, not greed, and questioned whether readers would actually pay subscriptions instead. xrd offered the uncomfortable middle ground: they'd personally found Gemini genuinely better than clicking through multiple articles for an answer, while admitting flatly that "Gemini wasn't paying the authors." marginalia_nu raised the question nobody in the thread could answer — how ad revenue stays stable industry-wide when traffic is down 60% before any AI-driven ad revenue exists to replace it. And gortok closed with the line that frames the whole diagnostic problem in this post: "What does AI do when the content it relies on stops coming?"

Nobody in that thread landed on one cause. That's the honest state of the evidence right now — real traffic loss, multiple plausible mechanisms operating at once, and site owners genuinely unable to agree which one explains their own numbers without actually digging into their own logs.

The Documented Scale of It

The tech-publisher pattern isn't one anecdote — it's measured across an entire category. A Growtika analysis using Ahrefs data, covered by Nieman Lab, tracked ten major English-language tech publications — Wired, CNET, Mashable, The Verge, and PCMag among them — from their 2024 peak traffic months through January 2026. Combined traffic across all ten fell from 112 million monthly visits to 47 million, a 58% decline. The Verge alone dropped 85%, from 5.3 million visitors in February 2024 to 790,000 in January 2026. CNET lost 47%, PCMag 41%, Mashable held up best at a 30% decline. Seven of the ten lost at least half their traffic. The report ties the acceleration specifically to Google AI Overviews expanding into how-to guides and gear reviews through the second half of 2025 — exactly the content categories tech publishers built their traffic on.

That's not a company doing something wrong. It's an entire content category — servable, answerable "how do I" and "which one should I buy" questions — moving from ten blue links to one synthesized answer, and every publisher who built a business on being one of those ten links absorbing some share of the loss regardless of how good their writing stayed.

How to Actually Run the Diagnosis

Put the four causes together and a real diagnostic sequence emerges, and it isn't guesswork if you're willing to actually check rather than assume. Start with your own site: has any URL, template, or hosting change happened in the window since visibility dropped, and does Google Search Console's indexing report show a spike in soft 404s or "crawled — currently not indexed" pages that lines up with it? That's the migration-fault pattern, and it's the one most directly in your own control to fix. If nothing changed on your end, check whether the specific query or topic you used to win now returns a competitor's page with a visibly more recent update date — that's the freshness pattern, and the fix there is publishing a genuinely better, newer answer, not a full rewrite. And if neither of those explains it, check whether the drop tracks with a known platform change — an AI Overviews expansion into your category, a core update, a retrieval-behavior shift like the one that hit Reddit — because that's the one you can't fix by changing your own content at all; you can only adapt to the new rules once you know what they actually are.

• Pull your indexing report for the exact window the drop happened in, not a monthly average that smooths it out.
• Check the actual current AI answer for your core queries across at least two or three platforms, not one — measurement disagreement between tools is real, as the Reddit case shows.
• Compare your page's last-updated date against whatever page is winning the citation now.

That third step is the one businesses skip most often, because it means actually looking at what replaced you rather than only auditing your own page in isolation — and it's usually where the real answer is sitting. You can run a free first pass of that comparison yourself with the AEO Score tool before deciding whether the cause is something you can fix in an afternoon or something that needs a proper investigation.

Where This Actually Leaves You

The honest version of this whole piece is that "you lost AI visibility" isn't one problem, it's four different problems wearing the same symptom. A migration that broke your redirects gets fixed with a redirect map and a crawl-budget cleanup. A competitor's fresher page gets answered with genuinely updated content, not a cosmetic edit. A platform rule change — the kind that hit Reddit twice in eleven months with zero warning either time — gets managed by knowing it happened at all, since there's no lever on your side to pull. Treating all four as the same "AI SEO problem" and throwing a generic content refresh at it is how businesses spend a quarter fixing nothing, because they never actually confirmed which of the four things happened to them.

That diagnostic step — figuring out which of those four actually happened before deciding what to do about it — is the entire premise behind our own AI Visibility Recovery service: we don't start from a generic audit template, we start by finding what specifically changed, because a business that used to be cited already has real evidence sitting in the historical record of what worked. The fix is almost always faster and cheaper once you know which of the four suspects actually did it.