Your AI visibility check has an expiration date

A blended AI visibility score gives you a number. It does not tell you which engine drove a change, when that change happened, or whether the number held steady because one engine improved while another quietly declined. In a category where the four major engines pull from meaningfully different source pools, a single point-in-time check is a snapshot of a system that keeps moving.

Why the number changes without warning

Each of the four engines runs its own search step, pulls from its own mix of pages, and updates that mix on its own schedule. No engine publishes a changelog for its ranking behavior. A directory that ChatGPT indexes heavily this quarter may carry different listings than it did six months ago. Perplexity's reliance on social and community content means its answers shift as discussions in your category accumulate and age out.

We ran 46 buyer-intent prompts against four engines using live web search for one regional brand. The blended mention rate across all 184 responses was 28 percent. But that number averaged together four very different engine results.

Data from a 46-prompt buyer-intent panel run against all four engines in the same week. 184 total responses.

These are starting positions from a specific point in time. Each one will change, and the engines will change on different schedules for different reasons.

Why a blended score masks per-engine movement

Because the engines start from such different positions, a meaningful shift in any one engine produces a smaller blended impact than it looks like on paper. If Perplexity climbs from 17 percent to 25 percent while the other three hold steady, the blended number moves from 28 percent to 30 percent. Two points blended. Per-engine, that is nearly a 50 percent improvement in Perplexity coverage.

The inverse is equally masked. If Perplexity falls from 17 percent to 9 percent while ChatGPT holds at 33, the blended number slips from 28 to 26. That reads as noise in a monthly report. For buyers who search on Perplexity, the brand has moved from weak presence to near-invisible. Neither the gain nor the loss would trip an alert in a system tracking only the blended figure.

Perplexity's source pool is the most likely to drift

We extracted the citation domains from the same 46-prompt panel. Perplexity cited facebook.com 44 times across all responses. ChatGPT cited it zero times. Claude cited it 6 times. Social content drives a much larger share of what Perplexity surfaces than what the other engines surface.

Social content is noisier than directory listings. A community discussion about internet service in a particular county gets written once, generates replies for a few weeks, and is buried by newer threads. ISP aggregator databases update on a slower, more predictable cycle, and once a brand is listed it tends to stay listed. This means the source pool feeding Perplexity turns over faster than the source pool feeding ChatGPT or Claude.

A one-time check captures where Perplexity was looking on the day you ran the panel. In a category where social and community content is a primary Perplexity source, that picture will look measurably different in 90 days without any action taken by the brand or the engines.

What tracking catches that a one-time audit cannot

A one-time audit answers one question: where do I stand today? That is a useful baseline. Monthly tracking answers a different question: which direction am I moving, on which engines, and at what rate?

Knowing that your Perplexity mention rate dropped from 17 percent to 11 percent over 60 days, while your ChatGPT rate held steady at 33 percent, tells you something specific. The source mix feeding Perplexity is drifting against you. The fix is not to improve your website or get more aggregator listings, because those channels were not the problem to begin with. Perplexity was already pulling from your site and ignoring it. The community content surface is what changed.

A blended score in both periods would read as 28 percent declining to 26 percent. That registers as a rounding error. The per-engine trend is where the actual signal lives.

How to use an audit as a starting point

An audit gives you the per-engine baseline and the source gaps: which pages each engine was pulling when it recommended a competitor instead of you. That is the diagnostic. The fix work starts from the source gaps, not the blended score.

But a fix applied to one engine's source gaps will not automatically move the others. Getting listed on an aggregator that ChatGPT favors does not touch the social content pool that Perplexity draws from. You need both the per-engine diagnosis and a way to track whether the fix worked on the engine you targeted, without being misled by movement on the others.

The starting-point audit is the map. Monthly tracking tells you whether you are moving.