Why did my visibility drop?

A sudden drop in organic Share of Voice, mention rate, or citation rate is often a measurement change rather than a real loss of presence. This playbook walks the causes in order of likelihood and gives the fix for each, so you finish with a confident diagnosis instead of a guess.

First, do not panic

Visibility is probabilistic: answers vary run to run. One low day is rarely meaningful on its own. Before you investigate, confirm the drop shows up across more than a single run.

Causes, ranked by likelihood

LikelihoodCauseFix
Most likelyYour prompt set changed, so the denominator shiftedCompare like-for-like windows from the same prompt set
CommonAn engine changed its model or source mixCheck whether the drop is isolated to one engine
PossibleA competitor published content that engines now citeOpen the answers and check the sources panel
Always presentRun-to-run sampling noiseWait for more runs before concluding anything

1. Your prompt set changed

This is the most common cause. Visibility metrics are computed over the prompts in your set, so adding, removing, or editing prompts changes what you are measuring. If you added six broad category prompts where you are weak, organic SOV can fall even though nothing about your old prompt set changed. Fix it by comparing windows that use the same prompt set: look at unchanged prompts before and after, not the blended total. See build a prompt set for how edits affect the denominator.

2. An engine changed its model or source mix

Engines update their models and the sources they lean on without notice, and they disagree with each other, which is why AppearIn tracks them separately. If your drop is concentrated on a single engine while ChatGPT, Gemini, and Perplexity otherwise hold, an engine-side change is the likely cause. The fix is to read it as an engine-specific shift, not a brand-wide loss, and watch whether that engine re-stabilizes over the next several runs.

3. A competitor published content engines now cite

If the drop is real and spread across engines, a competitor may have earned new citations. Open the answers for the prompts that fell and check the sources panel. If a new roundup, review, or competitor page is now feeding answers where you used to appear, that is your cause and your cue to act. The improve Share of Voice playbook covers the response.

4. Genuine run-to-run sampling noise

Even with everything stable, individual runs vary. A small wobble on one snapshot is usually just sampling. The fix is patience: wait for more runs. If the metric returns to its prior band, it was noise; if it stays down across runs and engines, treat it as one of the causes above.

Worked example

Acme's organic SOV drops from 31% to 19%. Acme checks the ranked causes in order. The prompt set is unchanged, so cause 1 is out. The drop is almost entirely on Perplexity while ChatGPT and Gemini held steady, which points to cause 2. Opening the Perplexity answers shows the same brand mentions but a reshuffled source mix. Acme waits four more runs; Perplexity settles back to 29%. Diagnosis: an engine-side source change plus normal noise, not a loss of presence and no action needed.

What good looks like

You have diagnosed the drop well when you can state, in one sentence:

  • Whether the drop is a measurement change or a real loss.
  • Which engine or engines it affects, and whether it is isolated.
  • Whether it persisted across several runs or reverted, and therefore whether to act.

The trap to avoid

Reacting to a single low run. Most overnight drops are denominator changes or noise. Compare like-for-like windows and wait for the trend before you change anything.

Related