AI visibility tracking: definition

AI visibility tracking is a measurement method, not a single result. It records how engines such as ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews answer a fixed set of buyer questions. Then it repeats that measurement over time.

The word that matters is consistent. Tracking holds these conditions steady between runs:

• The prompt set: the same questions, in the same wording and language.
• The engines: the same surfaces, recorded one by one.
• The session state: logged out, with memory off, and the same search setting.
• The scoring: the same method for mentions, citations and factual accuracy.

A snapshot uses the same steps once. It is useful as a starting point. It can't show a trend. For the full mechanics, see the guide on how AI visibility tracking works.

A tracking run sends each prompt to each engine and stores the full answer. It logs whether the brand appeared, which URLs were cited, and whether the stated facts were right. Then it repeats at a set interval.

Consistency is harder than it sounds, because engines shift the test underneath you. As of October 2026, OpenAI says ChatGPT search typically rewrites a question into one or more targeted queries before searching. It also says saved memories can shape that rewrite. A logged-in run with memory on is not reproducible, so trackers should run logged out.

Google also says AI Overviews and AI Mode may use different models and techniques. They can return different links for the same question. Treat them as two surfaces, not one. How often to re-run depends on how fast your category moves. The check cadence playbook covers that choice.

AI answers move even when your site doesn't. Engines decide per query whether to search the web at all. Indexes change, and models get updated.

Otterly's 2026 experiment shows the scale of movement. Otterly tracked 100 AI citations for one test site in April and 1,666 in the first 27 days of July. That happened after Google had removed the site from its index. Most of those citations came from Microsoft Copilot, while Google's AI surfaces almost never cited it. A single snapshot in April or in July would have told two very different stories.

The practical move is to set a baseline, then re-run under the same conditions. Read each engine on its own line. Whether one check is ever enough is covered in is a one-time AI visibility check enough.

"Tracking gives me my AI rank"

It doesn't. Google's reporting for AI features exposes impressions, not positions, so no ground truth exists to check a rank against. Read why there's no AI rank to track.

"A tracker sees every citation"

A prompt-based tracker counts answers to the prompts it asks. In Otterly's test, Microsoft's own AI Performance report credited one site with 42,817 citations, compared with 3,934 in Otterly's prompt tracker. The time windows may differ, but the gap shows that a prompt set is a sample. Compare the two data types in prompt tracking vs engine-reported data.

"My prompts are what buyers type"

Engines rewrite and fan out queries, so your prompts are a proxy for an input nobody outside the engine fully sees.

"One score covers everything"

A blended score averages engines that can behave in opposite ways. Keep the per-engine numbers.

• LLM visibility tracking: the same method framed around measuring across models.
• Free AI visibility tracking: what a free snapshot covers and what it misses.
• Snapshot: a single run with no baseline to compare against.
• Engine-reported data: counts an engine publishes itself, such as Bing Webmaster Tools' AI Performance report.

What is the difference between a snapshot and tracking? A snapshot runs your prompts once and shows what engines said that day. Tracking repeats the same prompts, engines and scoring on a schedule, so you can see direction. The free checker on this site gives a snapshot: up to three buyer questions, scored once. It is a good baseline, not continuous tracking. To see change, you need repeat runs under the same conditions.

What happens to my history if I change the prompt set? The new results stop being comparable with the old ones. A different set of questions measures a different thing, even if the score looks similar. Treat a changed prompt set, a changed engine list or a changed scoring formula as a new baseline. Note the date of the change so nobody reads the jump as a real shift in AI answers.

Can tracking tell me why my visibility changed? Not on its own. Tracking shows that answers changed, and on which engine. The cause sits elsewhere: a crawler blocked at your firewall, a page dropped from an index, content that only appears after JavaScript runs, or an engine update you can't control. Use a drop in one engine as a prompt to check access and index status for that engine specifically.

Should tracking run logged in or logged out? Logged out, with memory off. OpenAI says ChatGPT can use saved memories when it rewrites a search query, which makes logged-in results depend on the account's history. Logged-out runs are the closest you can get to repeatable conditions. Even then, engines decide per query whether to search, so expect some run-to-run variation and judge trends over several runs.