What this methodology covers
This page describes how Paz measures AI shopping visibility for the customers of its AI visibility service. It covers which assistants Paz asks, how the questions are built, what Paz records and how the results are reported. The service itself, in which our experts investigate the misses and fix the product information behind them, is described on the AI visibility service page and priced on the pricing page.
Which assistants Paz covers today
Paz reports on ChatGPT, Google AI Mode and Perplexity. Google AI Overviews, Microsoft Copilot and Gemini are captured as diagnostics and reported separately. Assistants are added to this list when Paz runs them for customers.
| Assistant | Role in the visibility report |
|---|---|
| ChatGPT | Primary reporting assistant. Per-engine appearance counts, found rate, position and trends. |
| Google AI Mode | Primary reporting assistant. Per-engine appearance counts, found rate, position and trends. |
| Perplexity | Primary reporting assistant. Per-engine appearance counts, found rate, position and trends. |
| Google AI Overviews | Diagnostic. Captured with a tri-state outcome (present, not shown, failed) and reported separately from the three primary assistants. |
| Microsoft Copilot | Diagnostic. Captured with the same tri-state outcome and reported separately from the three primary assistants. |
| Gemini | Diagnostic. Captured with the same tri-state outcome and reported separately from the three primary assistants. |
Monitoring cadence
Monitoring scope and cadence are agreed for the job, including the tracked questions, assistants and product coverage.
- Managed service: AI visibility, catalog-related work and recurring commercial-performance investigations within the agreed scope.
- Enterprise: larger-scale, multi-brand and custom requirements with individually agreed scope.
Cadence reflects when Paz runs the tracked questions. Results are compared across completed runs on the scheduled cadence agreed for the job.
How the questions are built from buyer personas
The tracked questions are full shopper questions, written the way a named buyer persona would ask them, agreed with the customer before the first run.
Our experts start from the customer's buyer personas and selling priorities: the collections being promoted, the margins that matter, the season, and who buys. For each persona Paz writes the questions that buyer would put to an assistant, in full sentences with the constraints a real shopper adds, such as "I need a reliable waterproof hiking boot under $200, what are my best options?" for a deal-seeking camper, or "a wide-fit hiking boot for men" for a wide-feet hiker. Each question is tagged with its persona and with how broad or specific it is, so a miss on broad questions can be told apart from a miss on specific ones.
The customer reviews the set before the first run and can add, remove or reword questions at any time. Competitors are tracked for the same questions, so the report shows who else appears for each question.
The tracked questions are configured runs: Paz writes them, Paz runs them, and Paz records what each assistant answered at that moment in the agreed locale. The value of the set comes from how closely it matches what the customer's buyers actually ask, which is why it is built from personas and revised as the business changes.
Appearance types: what Paz records
| Term | Definition |
|---|---|
| Product card | A structured product result rendered by the engine, such as an image, price, merchant name, and link where the engine provides them. |
| Product mention | A natural-language reference to a merchant product in the response without a structured card. |
| Brand mention | A reference to the merchant brand in the response without a specific product. |
| Source citation | An outbound URL the engine cites as a source for the response. |
| Position | Where the merchant or its product appears in the response, recorded so changes can be tracked over time. |
| Found rate | The share of the tracked questions that surface the merchant as a product card, product mention, brand mention or source citation. |
Paz records brand mentions and source citations separately, so a merchant can see being named and being cited as two different results.
How Paz reports per-engine visibility
For each tracked question and each primary reporting assistant (ChatGPT, Google AI Mode and Perplexity), Paz records the count of each appearance type (product card, product mention, brand mention, source citation), the found rate (the share of the customer's tracked questions that surfaced the customer), and the position (where the customer appeared in the response, as observed). These are reported as observed values and as trends over time.
The found-rate denominator is the set of tracked questions agreed for that assistant, so found rate is reproducible from the question set and the recorded appearances. Appearance counts and positions come from the same recorded results. Trends compare those per-engine observations across completed monitoring runs. The interval follows the cadence agreed for the job. Google AI Overviews, Microsoft Copilot and Gemini are captured with a tri-state outcome (present, not shown, failed) and reported separately.
Use the per-engine appearance counts, found rate, position and trends described above to interpret and reproduce reported visibility. Every change our experts ship is measured against the baseline these runs establish, and what moved shapes the next round of work.
How Paz treats the diagnostic assistants
Google AI Overviews, Microsoft Copilot and Gemini are captured and displayed with a tri-state outcome: present, not shown, or failed. They are reported separately from the three primary per-engine visibility reports.
Google AI Overviews appears for some questions only. A question with no Overview is recorded as an observation.
How to interpret the results
- Treat each result as an observation of a tracked question at the time and in the locale where Paz ran it.
- Monitoring runs on the schedule agreed for the configured work.
- Compare changes across completed runs using the same question set and assistant.
- Use visibility changes as the signal for the next investigation. Our experts trace each change back to the work that preceded it.
Related
- AI visibility service : what our experts investigate, fix and report, month after month
- Pricing : the managed service and what it includes
- Editorial policy : primary-source, author and reviewer, and corrections standards