Your AI visibility can vanish overnight, and often it isn't your fault
· Algorithm Updates
Three incidents in five weeks. One cause was never explained, one was a model bug and one never actually happened. If you'd reacted to any of them with a content overhaul, you'd have been fixing the wrong thing.
Late summer 2026 gave anyone tracking AI citations a crash course in volatility. None of what happened was caused by the sites affected, and in one case nothing happened at all. Together, the three incidents make the strongest case yet for a simple discipline: when your AI visibility moves, verify before you act.
Incident 1: Reddit falls out of ChatGPT
Reddit's share of ChatGPT search citations had been steady at an average of about 3.8% through 7 August. By 14 August it was under 1%, a relative drop of around 86% in a week (Search Engine Land, Axios).
The leading explanation was a backend change in how ChatGPT runs query fan-out, the process of breaking one question into many sub-searches, including heavier use of site:-style domain targeting from around 8 August. OpenAI hasn't commented. A follow-up analysis by Search Engine Journal then found the drop happened in two distinct phases, and that the fan-out explanation only accounts for the first, smaller dip (Search Engine Journal). The rest is still unexplained.
It isn't the first time, either: a near-identical collapse happened in September 2025, and Reddit recovered.
Incident 2: Google's AI Mode stops citing anyone
On 2 September, Google rolled Gemini 3.8 Flash into AI Mode, and for many top-of-funnel queries, AI Mode stopped citing or linking to sources altogether. Google Search VP Robby Stein confirmed the bug on 3 September; by 4 September, citations had largely reappeared (Search Engine Land).
A fortnight earlier, the move to Gemini 3.7 Flash had already shown a subtler version of the same thing: testers found similar lists of entities in answers but different top citation URLs for some queries (Search Engine Roundtable). Every model swap reshuffles citations to some degree. This one just did it all at once.
Incident 3: the drop that never happened
From 12 to 13 August, Search Console's generative-AI performance report showed a sudden cliff-edge fall in AI Overview impressions. It coincided with an unconfirmed spike in ranking volatility on third-party trackers, which made it look like something real. On 17 August, Google confirmed it was a logging bug, not a change in visibility (Search Engine Roundtable).
Why AI citations are so fragile
These aren't flukes. They're the normal behaviour of systems that choose sources on the fly:
- Every assistant cites a different web. Ahrefs' study of 17 million citations across seven AI platforms found that only 7 of the 50 most-mentioned domains appeared across Google AI Overviews, ChatGPT and Perplexity. Roughly 86% of cited sources were unique to a single assistant (Ahrefs).
- Even one assistant cites differently depending on how hard it's thinking. In ChatGPT, quick-answer and high-reasoning modes shared just 25.6% of cited domains, as we covered in Thinking, fast and cited.
- Models change constantly. Google changed the model behind AI Overviews and AI Mode at least three times between July and September. OpenAI shipped GPT-5.6 in July, made new versions of it the defaults in August and moved ChatGPT to GPT-6 Astra in September.
A strategy built on one number from one platform is built on sand.
Real drop or platform noise? A verification checklist
Before you rewrite anything, work through these questions:
- Is it one platform or all of them? If you've dropped in ChatGPT but held steady in AI Overviews and Perplexity, the cause is almost certainly on ChatGPT's side, not yours.
- Is it you, or your whole category? Track a small control set of competitors and neutral queries. If everyone in your space moved at once, the platform changed.
- Do two measurement sources agree? Compare Search Console's generative-AI reports with a third-party tracker. If only one shows the drop, suspect the measurement.
- Has anyone else noticed? Check Search Engine Roundtable, Search Engine Land and the usual social channels. Platform bugs usually surface within 24 to 48 hours.
- Did a model or product change just ship? Model swaps are the single most common cause of sudden citation reshuffles.
- Did business outcomes change? If AI-referred sessions, leads and conversions held steady, a visibility metric that dropped may matter less than it looks.
- Did you change anything? Only once the first six questions point back at you should you look at your own recent deploys, robots.txt, CDN rules or content changes. When the cause is on your side, it's usually something technical. See Is your site blocking AI search without knowing it? (coming soon)
Measure AI visibility like weather, not like rankings
The practical lesson is to change what you report:
- Use rolling averages (four weeks or more) rather than daily snapshots.
- Report citation share across platforms, not a single figure from the platform your tool happens to sample.
- Separate visibility from outcomes. Citations are a leading indicator; referrals, conversions and brand searches are what the business actually cares about.
- Annotate your dashboards with known model releases and platform incidents, so next year's you can tell a strategy problem from a Tuesday.
Frequently asked questions
Why did Reddit disappear from ChatGPT citations in August 2026? Reddit's share of ChatGPT search citations fell from about 3.8% to under 1% between 7 and 14 August 2026. A change to ChatGPT's query fan-out is the leading explanation, but OpenAI hasn't confirmed it, and follow-up analysis suggests it only explains part of the drop.
Why did Google's AI Mode stop showing citations? A bug that came with the Gemini 3.8 Flash rollout on 2 September 2026 stopped AI Mode from citing or linking sources for many queries. Google confirmed it on 3 September, and citations had largely returned by 4 September.
How can I tell if a drop in AI citations is real? Check whether it affects one platform or several, whether competitors moved too, whether two measurement sources agree, whether a model change just shipped and whether business outcomes changed, before looking at your own site.
Why do different AI assistants cite different sources? Each platform uses its own retrieval system, index and model. Ahrefs found roughly 86% of cited sources were unique to a single assistant, so visibility in one doesn't imply visibility in another.
The kicker
In August, the most-cited community on the internet nearly vanished from the world's most-used AI assistant, and nobody, including the company that runs it, can fully say why. If Reddit can't control its AI visibility week to week, neither can you. What you can control is whether you panic.
Want a baseline to compare against next time something moves? Run a free CiteSite audit and keep the report. It's a snapshot of what AI systems can see on your site today.
Sources: Search Engine Land: Reddit's ChatGPT search citations fall; Axios: ChatGPT, Reddit citations and GEO strategy; Search Engine Journal: why Reddit's ChatGPT citation drop isn't fully explained; Search Engine Land: Google to fix citation bug with Gemini 3.8 Flash in AI Mode; Search Engine Roundtable: Gemini 3.7 Flash in AI Mode; Search Engine Roundtable: Search Console performance report drop; Ahrefs: do AI assistants prefer to cite fresh content?.