Guide · tracking it over time

How to track your AI visibility

LLM visibility tracking means measuring how often AI engines cite your site — repeatedly, over time — so you can tell whether your share of AI answers is climbing, flat, or decaying. It matters because a single check lies: ask ChatGPT the same question twice and you'll get a roughly one-third different set of sources. The signal isn't any one reading. It's the trend.

Why a single check can't be trusted

The instinct is to type your buyer question into ChatGPT, see whether you're cited, and call that your "AI visibility." It isn't. AI answers are non-deterministic: the model samples differently each call, and the live web index it searches shifts between requests.

We measured exactly this. Running the identical query on the same model twice, back to back, the two answers shared only about a third of their cited sources — and our own domain appeared in one run and vanished in the next. A single check told two opposite stories about the same page in the same minute. Any "tracker" that samples once and reports a number is reporting noise.

The fix is boring and it works: run every question several times, pool the results, and compare month to month. A reading is a coin flip; a trend line is evidence.

What to actually track

Five things, in order of how much they tell you:

Answer-slot share — of all the source citations across your niche's buyer questions, the percentage that are your domain. This is your headline metric, the AI-answer equivalent of market share.
Share per engine — ChatGPT and Perplexity diverge, often sharply. (In our own data a product was cited by Perplexity across five queries and by ChatGPT on none but its brand name.) Track them separately or you'll miss where the gap is.
Per-query coverage — which buyer questions cite you versus which are held by someone else. This is where the trend becomes actionable.
Competitor share — the same measurement for your top rivals, so you know whether you're gaining on them or just moving with the tide.
The trend — the deltas across months. Rising, flat, or decaying is the only thing that tells you whether your strategy is working.

How often to measure

Monthly is the sweet spot. More often and you're mostly measuring engine noise; much less often and you can't tell a real change from a bad sample. Citations drift as engines update and competitors publish, so visibility is a position you hold and watch, not a number you check once. Keep the questions, engines, and run-count identical each month — if you change the method, you've broken your own trend line.

What the trend reveals that a snapshot can't

The same 4% share means completely different things depending on direction:

Pattern over months	What it actually means
3% → 6% → 8% → 9%	Working. The strategy compounds; keep going. (This is our own screen-recorder's real curve.)
4% → 4% → 4%, flat	Stuck. You're being sampled but not gaining — usually a saturated niche or a cluster too shallow to move.
1% → 1% → 0%, decaying	Losing. Competitors are displacing you, or the niche was never winnable. (Our AI-receptionist product did exactly this and we stopped spending on it.)

A one-time audit tells you where you stand today. Tracking tells you which of these three stories you're living — and that's the difference between spending confidently and guessing.

DIY, tool, or service?

DIY — you can script the engines' APIs, run your questions on a schedule, and log citations. Doable, and honest work; the hard parts are choosing the right questions, running enough repeats to beat the noise, and interpreting whether a move is real.
Tools — several dashboards now report an AI-visibility number. Useful for a quick pulse; watch for two failure modes — sampling once (noise) and reporting a metric with no benchmark to judge it against.
Service — a monthly tracked audit adds the interpretation: per-query map, competitor movement, and a verdict on what the trend means for your next move. It's the right fit when the decisions riding on the number are worth more than the number itself.

Frequently asked questions

Is "LLM visibility tracking" different from an AI visibility audit?

Same measurement, different cadence. An audit is the deep one-time read; tracking is that same measurement re-run on a schedule so you get a trend. The audit tells you where you stand; tracking tells you which direction you're moving.

Which engines should I track?

ChatGPT and Perplexity at minimum — they're the most-used and expose their sources clearly. Track them separately, because they cite differently. Google AI Overviews and Gemini are worth adding as coverage grows.

How many times should each question be run?

At least three, pooled. Given ~one-third run-to-run variation on the same query, a single run is unreliable; three-plus and averaged gets you a stable figure you can trust month to month.

How long before the trend means something?

Two to three months of consistent measurement before a line is trustworthy. One month is a dot; three months is a direction.

We'll track it for you — monthly.

The $300 audit is the deep first read. From there, monthly tracking ($149/mo) re-runs the same measurement, pooled across repeats, and reports the trend — the signal that tells you whether your strategy is actually working.

See the tracking plan