AI visibility tracker: the trend is the signal

AI answers are not stable. Model updates reshuffle which brands an engine trusts, retrieval-grounded engines change their answers as the live web changes, and a competitor's well-placed article can displace you from an answer without your Google rankings moving at all. A snapshot captures none of this motion; it tells you where you stood at one moment, on one phrasing, on one engine.

Tracking turns the same measurement into a line. When the line moves, you can tie the change to a cause, a model release, a content launch, a competitor entering, and respond while the shift is fresh instead of discovering it a quarter late.

• Mention rate per prompt: the share of repeated checks where the engine named your brand, per engine, so you see which surfaces know you and which don't.
• Share of voice: who else gets named on your prompts, and how your slice compares. Displacement shows up here first.
• Answer position: first-mentioned and last-mentioned are different outcomes; the tracker records where you land.
• Cited sources per answer: the URLs the engine drew from. These pages decide the answer, and changes in this list explain most visibility moves.
• History with timestamps: every past answer stored, so any change is auditable rather than anecdotal.

Three patterns cover most of what a trend line shows. A step change on a model releasemeans the engine's training data recut its answer set; whatever coverage you earned in the preceding months either crossed the threshold or didn't. A gradual slide with a rising competitor means you are being out-published on the sources the engine cites, and the cited-source list tells you exactly where. A fast gain after publishing, usually on retrieval-grounded engines first, is the confirmation signal that your content strategy is landing, and it typically leads the slower training-data engines by months.

A typical displacement reads like this. A brand holds a 70% mention rate on its ten money prompts, steady for two months. In week nine, the citation set on three prompts gains a new URL: a competitor's just-published comparison page. Mention rate hasn't moved yet, answer engines are still mid-digestion, but the tracker flags the citation-set change. Two weeks later the competitor appears in the answers beside the brand; two weeks after that, on two prompts, instead of it.

The intervention window was week nine, not week thirteen. Because the tracker showed which page earned the competitor their entry, the response was specific: a deeper, fresher page targeting the same question, plus outreach to the two publications the engines cited alongside. By week sixteen the mention rate recovered, and the whole episode, cause, response, and recovery, is legible in one trend line. A monthly manual check would have discovered only the loss.

Fewer than you think, chosen better than most do. Ten to twenty prompts covers a single product's buying journey: three or four discovery questions ("what is...", "how do I..."), five to eight comparison and category questions ("best X for Y"), and a few high-intent decision questions ("X vs Y", "is X worth it"). Past fifty prompts per product, additions mostly duplicate signal: engines answer adjacent phrasings from the same sources, so track distinct intents, not phrasing variants. Spend the saved budget on checking more engines instead, because visibility divergence between engines is a bigger blind spot than variety within one.

AI visibility follows evidence. Engines name brands they have seen reinforced by sources they trust: well-cited industry coverage, structured content with clear authorship on your own site, and presence in the pages they retrieve when answering live. That is why a tracker pays for itself in direction rather than measurement alone: the cited-source lists across your prompts are a ranked to-do list of the exact pages and publishers shaping your category's answers. Earn your way into those, and the mention rate follows.