AI visibility tools: who ChatGPT and Perplexity cite for your queries

Short version, for the scrollers: a rank tracker can’t tell whether ChatGPT or Perplexity quote a site, so we built a probe that asks them directly. Plus a free pre-check for which queries are worth asking about.

I opened a rank report and a Perplexity answer for the same query side by side, and honestly, it seems the two have nothing to say to each other. The position gets no vote in the answer. That gap is what an AI visibility tool measures: citation, not rank. So the probe asks the AI engines directly for the tracked queries, reads the answers, and shows two things: are you cited, and who is cited instead.

The problem: the tracker sees results, not the AI answer

A classic rank tracker knows one thing — a link’s position in the SERP — and from it you can indirectly guess a Google AI Overview hit. But what about engines that aren’t built on Google’s results? Perplexity, ChatGPT Search are simply invisible to a normal tracker, and that’s increasingly where brands get seen or not. So what was the tracker measuring all this time?

What we built

/seo-matrix:ai-citation takes the same set of tracked keywords and, for each, actually asks an AI engine (Perplexity sonar by default; provider-agnostic), then parses the answer:

• whether the site is cited for that query;
• which domains are cited instead (competitors in the answer);
• citation share — the site’s share of citations across the query set;
• movement vs the last run — who entered the AI answers, who dropped out.

History is journaled (ai-citation-history.jsonl), so what comes out is a trend, not the engine’s mood on one particular morning.

First — a free pre-check

The probe is paid, one call per keyword, so how do we not burn budget on unwinnable queries? With a $0 offline pre-check, /seo-matrix:citation-sim. It takes the best page for each keyword, scores its citation-worthiness (the same 0–100 heuristic as citation-audit), and predicts a verdict — likely / borderline / unlikely / weak / no_coverage — splitting queries into ”🔬 probe” vs ”🛠 fix first.” The paid probe spends only on what’s actually winnable; the rest probably belonged on the fix list anyway.

Why it works this way

Why measure it directly instead of inferring it from positions? Because an AI answer is assembled from extractable passages, not positions: one query fans out into sub-queries, and a source makes the answer by citation frequency, not by its rank line (see our breakdown, What is GEO). So “am I cited” is a separate metric — you measure it with a direct probe.

What to do with it

1. Run citation-sim → get the list of winnable queries.
2. Run ai-citation on them → see where the site stands, and where a competitor stands instead.
3. Where a competitor is cited, strengthen the page (facts, sources, extractable paragraphs) and re-probe. Then re-probe later — a single win might just be noise!

In short

• Positions ≠ AI citation; a normal tracker can’t see Perplexity/ChatGPT.
• ai-citation asks the engines directly → are you cited, and who instead.
• citation-sim ($0) filters out unwinnable queries first — pay only for winnable ones.
• Citation is a separate metric; measure it with a direct probe.

That’s the whole trick: stop asking Google where you rank, start asking the engine who it trusts. (The answer stings the first time. It’s supposed to!)

Sources

• This is a native seo·matrix capability (/seo-matrix:ai-citation + /seo-matrix:citation-sim), validated against the live Perplexity API.
• Why citation ≠ position: Aggarwal et al., “GEO”, KDD ‘24 — arXiv:2311.09735.
• AI-answer architecture (query fan-out, citation chips): US v. Google case materials (2025).