Your brand was getting cited in AI answers, and now it is not. Before you rewrite the site or blame an algorithm, work this diagnostic in order. It runs from the cheapest and most common explanation to the rarest, so most drops are explained in the first two steps. The order is the point: skip it and you can spend a week fixing something that was never broken.
One thing to hold onto before you start. AI answers are noisy by design. The same prompt, asked twice, can return different sources, even at the most deterministic settings, because of how the models actually run. So a single missing citation is not a trend. The first job of a diagnostic is to decide whether anything really changed.
Step 1: Rule out noise. Did anything actually change?
Look at the trend, not the snapshot. If your brand showed up in 7 of 20 tracked prompts last week and 6 of 20 this week, that is noise, not a drop. Because AI answer generation is nondeterministic and returns different citations run to run, you judge movement over several checks, not one. If the line has slid for two or three checks in a row, it is real and you continue. If it dipped once and recovered, you are finished here; go do something useful.
Step 2: Rule out a measurement artifact. Are you comparing like with like?
The second most common drop is not a drop at all. It is two different measurements set next to each other. A Daily Check reviews a focused set of prompts for recent movement; a Full Check reviews your complete tracked set and is the baseline for rank and reporting. Hold a Daily Check up against a Full Check and the smaller, focused review can look like a collapse when nothing actually moved. Compare like with like: Full Check to Full Check, same prompts, same platforms, same country. Only a same-scope comparison gives you a real before and after.
Step 3: One platform, or all of them?
Now that the drop is real, localize it. Did you fall on every engine or just one? This matters because the engines barely look at the same web. In our own citation study, the five platforms agreed on only a tiny fraction of the sources they cite. A drop on a single engine has a single-engine cause: that platform changed, or your source on that platform moved. A drop across all five at once points at something on your side, usually access or a single page many engines were relying on.
Step 4: Did bot access change?
This is the fastest real cause to check, because it is binary and you control it. A one-line change to robots.txt, or a new firewall or CDN bot rule, can stop the assistants from fetching your pages, and the citations decay within days. Confirm that ChatGPT, Claude, Gemini, and Perplexity can still reach you with our robots.txt AI bot checker, and check that nothing in front of your origin started returning errors to their crawlers. We made the case for why this comes before any other AI work; it is doubly true when something breaks.
Step 5: Did your cited pages change?
Open the specific pages that were earning the citations and check what moved. A URL that 301-redirected, a page that quietly picked up a noindex, a redesign that buried the quotable section, or a post that was deleted will each drop the citations that pointed at it. The model cited a particular page. If that page changed underneath it, the citation leaves with it. This is also why a CMS migration or a content cleanup so often shows up as an AI visibility drop a week later.
Step 6: Did a competitor move?
AI visibility is relative. If a competitor published stronger coverage of the exact questions you were winning, the model can swap them in for you, and your own pages never changed. This is not a stretch: in our data, competitors already make up a large share of the sources AI cites about a brand. Check the queries where you slipped. If a rival now owns the answer you used to own, that is your drop, and the fix is content, not configuration.
Step 7: Did the platform itself change?
The rarest and most frustrating cause: the engine changed how it picks sources and you did nothing wrong. It happens, and it happens at scale. Google’s move to Gemini 3 in AI Overviews in late January 2026 is the clearest recent example. Across the months around it, the share of AI Overview citations coming from top-ranking organic pages fell from about 76 percent to 38 percent, according to a large Ahrefs analysis. Nobody who lost a citation in that shift had a bug on their site. If your drop lines up in time with a known platform change and the earlier steps came up clean, the work is to adapt to the new source behavior, not to keep hunting for a mistake you did not make.
What to do once you know which branch you are on
- Noise or measurement artifact: nothing. Close the tab and stop looking.
- One platform, or platform drift: adapt your sources to that engine and leave the others alone. Changing all five because one moved is how you create the next drop.
- Bot access: fix the rule, then wait a full re-crawl cycle before you judge whether it worked.
- Content: restore or rebuild the exact page that lost the citation, with the quotable part intact.
- Competitor: treat it as a content gap and out-cover them on those specific questions. The end-to-end loop for that is in our GEO playbook.
The thing that makes this fast is having a clean baseline to compare against: a real before-and-after at the same scope, same prompts, same platforms. That turns a panic into a ten-minute diagnosis. If you have not checked the simplest cause yet, start there now and confirm the assistants can still reach you with the robots.txt AI bot checker. A surprising share of real, fixable drops show up right at that step.
