If your brand stopped showing up in ChatGPT or an AI Overview this week, do not panic and do not assume you broke something. AI answers are unusually volatile, and a single drop is often noise rather than a real decline. The skill is telling the two apart. Here is why AI visibility swings, how to know whether yours is genuinely falling, and what to do in each case.
First, how volatile these systems really are. A drop of this size is often the model moving, not you.
| Signal | What the data shows | Source |
|---|---|---|
| Reddit's ChatGPT citation share | ~60% (early Aug 2025) fell to ~10% (mid-Sep 2025) | Semrush |
| Same query, repeated | AI Overview result changes ~70% of the time | Industry analysis |
| Two identical queries, same brand list | Under a 1-in-100 chance | SparkToro |
The volatility has a structural cause. AI answers lean on a small set of sources, so when one is reweighted, everyone drawing on it moves at once.
Share of AI citations
Most-cited domains in AI answers
AI answers are volatile by design
The same question can return a different brand list on the same day. AI Overview content changes for the majority of repeated queries, and citation shares move sharply on model updates. Reddit's share of ChatGPT citations fell from around 60 per cent to about 10 per cent in a fortnight in late 2025, per Semrush. None of that was caused by the brands involved. It was the model changing underneath them.
Noise versus a real decline
Tell them apart by pattern, not by a single check. Noise looks like this: you appear in three of five runs, disappear in one, come back the next day, or drop on one model while holding on the others. A real decline looks like a sustained fall across multiple models, over several weeks, on the questions that matter. If you only checked once, you cannot tell yet. Run the question several times, across models, over time. What AI visibility is covers how to read the pattern.
Want to get recommended by AI?
Check your AI search visibility, then let the Honeyb agent write, fix, and earn what gets you recommended. Free to start.
Real causes of a genuine drop
When it is real, the usual causes are a model update that reweighted its sources, a competitor earning more mentions and displacing you, your key sources changing or losing authority, or your own footprint going stale while others' grew. Note that none of these are fixed by editing your homepage.
What to do about noise
Almost nothing, beyond keeping perspective. Do not rewrite your strategy off one bad reading. Track the trend, report ranges rather than single numbers, and set expectations with your team that weekly wobble is normal. Over-reacting to noise is its own failure mode.
What to do about a real decline
Diagnose where you lost ground. Compare which sources the models cite now against before, and which competitors gained. Then rebuild the footprint: earn fresh third-party mentions, re-engage the communities and comparisons that feed the answers, and make sure your content is current and quotable. How to get cited by AI and how to get cited on Reddit are the practical next steps.
Measure so you can tell the difference
The only way to separate noise from decline is a consistent measurement over time. Track a fixed set of buyer questions across ChatGPT, Gemini, Claude and Perplexity on a schedule, and watch named mentions as well as citations. A single screenshot cannot tell you whether you have a problem. Start with the free AI visibility check.
