How to Track Your Brand's Visibility in ChatGPT and Perplexity

Tracking your brand's visibility in ChatGPT and Perplexity means running a fixed panel of 20 to 50 category questions against both engines on a recurring cadence, then logging four things per response: whether you were mentioned, whether you were cited with a link, where you appeared, and the sentiment of the framing. Divide mentions by total responses for your visibility rate, and repeat the same panel every week or month so the number becomes a trend, not a one-off screenshot. Our own daily panel puts this in perspective: run across four AI engines, it currently measures a 7% citation rate for citability.dev itself across 104 query receipts, and 20% for chudi.dev across 69 receipts, both self-published, not client numbers.

Most of what gets sold in this category, ours included, is an automated version of a process you can run by hand this afternoon. Before paying for automation, it is worth understanding what a documented measurement bug in some AI visibility tools can do to the numbers you would be automating: our own audit of that failure mode is the reason we now log a receipt for every automated query we run, and it is the same reason a manual method, run correctly, is a more honest starting point than any dashboard you have not verified.

What "AI Visibility" Actually Measures#

Four numbers get bundled under the label "AI visibility," and each one answers a different question on its own.

Share of voice is the percentage of total brand mentions across your panel that belong to you, calculated as your mentions divided by (your mentions plus every competitor's mentions), times 100. It tells you your relative standing in a category, not your absolute presence.

Citation rate is the percentage of responses where an engine links to your domain as a source, calculated as cited responses divided by total responses, times 100. It is the highest bar of the four, since an engine can mention your brand by name without ever linking to your site.

Position is where you appear within a response when you do appear: named first, buried in a list, or absent entirely. Two brands can share an identical mention rate while one is consistently the first name in the answer and the other is a footnote three items down.

Sentiment is whether the framing around your mention is positive, neutral, or negative. A brand can be mentioned and cited and still be described unfavorably relative to a competitor in the same response, which a raw mention count will never show you.

The 4-Step Method for Tracking AI Visibility Manually#

This is the complete process. It requires no software beyond a spreadsheet, and it is the same process an automated tool runs on a schedule for you.

Step 1: Build a 20 to 50 prompt panel. Write down the real questions your buyers ask in your category, not queries about your brand name. Split them across three intents: informational ("what is X"), comparison ("X vs Y"), and recommendation ("best X for Y"). Group the panel into 3 to 5 topic clusters. A panel under 20 queries gives you a result too noisy to act on; a panel of 50 gives you enough volume to see per-cluster patterns without becoming unmanageable by hand.

Step 2: Choose engine coverage and a cadence. At minimum, run the full panel in ChatGPT with web browsing enabled and in Perplexity, since those are the two engines most buyers mean when they say "AI search." Add Claude or Gemini if your audience uses them. Pick a cadence and hold it fixed: weekly for a category where model updates and competitor content move fast, monthly for a stable one. A cadence that slips defeats the purpose, because you are no longer comparing like periods.

Step 3: Score every response with the same rubric. For each query, on each engine, record four fields: mentioned (yes or no), cited with a link (yes or no), position (first, mid-response, or absent), and sentiment (positive, neutral, negative). This is the same rubric described above, applied response by response. A basic spreadsheet with one row per query per engine handles a 50-query panel without any tooling.

Step 4: Calculate share of voice and citation rate, then trend both. Run the two formulas from the previous section against your logged data, once per engine. Then repeat the entire panel on your chosen cadence and plot both numbers over time. A single run is a baseline. Three or four runs on a fixed cadence is monitoring, and monitoring is where the actionable signal actually lives, since it shows you whether a content or schema change moved the number or whether the shift was a model update unrelated to anything you did.

Why ChatGPT and Perplexity Need Separate Tracking#

Do not average ChatGPT and Perplexity into one number. The two engines are architecturally different in how citations get generated, and collapsing them hides which one is actually the problem.

Perplexity's sonar models are built to search on every query; there is no version of a sonar response that skips retrieval, so a citation-bearing answer is the default behavior, not something the model chooses. ChatGPT's Responses API is different: citation metadata, the url_citation array a tracking method depends on, only populates when the model actually invokes the web_search tool during that specific response, and by default the model decides whether to search at all. A panel run without forcing that tool invocation will under-report ChatGPT citations for reasons that have nothing to do with whether the engine actually knows your brand. Chudi's breakdown of exactly how ChatGPT and Perplexity differ in citation behavior goes deeper into the mechanics behind this asymmetry if you are building your own scoring pipeline.

Track the two engines as separate rows in your log, and expect the numbers to diverge for structural reasons before you ever start diagnosing your own content.

Manual vs. Tool-Based Tracking#

The manual method above is the complete process. A paid tool automates exactly that process and nothing more: it runs the same panel on the same schedule against the same engines, logs the same fields, and calculates the same two formulas, just without you doing it by hand every week.

Manual tracking is free and gives you full visibility into every raw response, which matters if you want to catch measurement problems like the ChatGPT citation-tool bug above. Its cost is your own time and the discipline required to hold a cadence for months without letting it slip, which is where most manual efforts quietly die after the second or third run.

Automated tracking costs money and trades away that raw-response visibility for consistency: the panel runs on schedule whether or not anyone remembers to run it, and the trend line accumulates automatically. This is the honest tradeoff, and it is why our own tool reports its numbers the same way a manual panel would: our daily automated panel, run across four AI engines, currently shows a 7% citation rate for citability.dev across 104 query receipts and 20% for chudi.dev across 69 receipts. Neither number is a marketing claim. They are what the same method described in this post returns when it is run daily instead of quarterly, and they are exactly the kind of unglamorous, sub-25% numbers that most brands, including the ones building the measurement tools, actually get.

Start manual. If the discipline of holding a fixed cadence becomes the bottleneck rather than the content work the numbers point you toward, that is the point where automating the same panel starts to pay for itself.

FAQ#

The four questions covered in this post's structured data: how to run the tracking method itself, what counts as a reasonable share-of-voice benchmark, whether ChatGPT or Perplexity cites more brands and why, and whether any of this is possible without paying for a tool. Each answer is written to stand alone so an AI engine can lift it without the surrounding page.

What to Do Next#

Run the free AI visibility scan on citability.dev to check the infrastructure signals (crawl access, structured data, rendering) that determine whether your content can be cited at all before you invest time in a manual query panel. It takes under two minutes and gives you the infrastructure half of the picture this post's tracking method does not cover.

For the engine-level mechanics behind why ChatGPT and Perplexity diverge in citation behavior, the deeper technical breakdown lives on chudi.dev: Perplexity vs ChatGPT Citation Rules.