VisibilityOS by PodPitch

Matched AI answer comparisons

The public aggregate data from the VisibilityOS study covers 817 matched comparisons, 36 questions, and eight engine views from October 19 to November 17, 2026 in UTC.

Download the counts

Pooled comparison outcomes

Counts for gaining, losing, or keeping the same presence status, and a separate count for link changes.

Download outcome CSV

For a reproducible analysis, use the version 1.0.0 files from jsDelivr. Download the versioned engine-view CSV and versioned outcome CSV, or browse the audit template on jsDelivr.

The VisibilityOS Study Reader lets compatible assistants read published totals or compare two model views. Try it in the Glama browser inspector or find its connection details in the official MCP Registry.

The Postman documentation shows how to fetch both files. Download the Postman collection to run the checks against the published counts.

The generated study guide on DeepWiki explains how the repository organizes the data, explorer, and tools for reproducing the results.

Download the VisibilityOS study container archive for Linux on AMD64 and ARM64. The release includes the aggregate data, calculator, and a checksum for the archive.

Field definitions

engine_view
The engine and mode in the published comparison table.
comparisons
The number of matched pairs for the engine view or pooled measure.
presence_changes
Pairs where the tracked brand was present on one date and absent on the other. A mention or link counts as presence.
measure and count
The pooled outcome and its count. The gains, losses, and unchanged counts add to 817. The link-status measure can overlap any of those outcomes.

How the data was collected

VisibilityOS matched observations with the same question, brand configuration, engine view, recorded model, measurement mode, country, language, and location. It kept the last eligible observation for each UTC date and paired consecutive eligible dates.

The files contain study totals only. The sample uses monitored questions about PodPitch-owned websites. Questions and observations can contribute to more than one pair, and coverage varies by engine view.

See the VisibilityOS guide to comparing the same engine mix for a worked example of how coverage differences can change a pooled rate.

Read the source study and methodology ยท Open the study explorer

Run the study notebook in Binder to reproduce the per-view rates and pooled result in a temporary Jupyter session.

Use the runnable VisibilityOS paired mention example to calculate before and after rates from matched counts. The example uses fictional data.

The Internet Archive research record contains a one-page methods brief and both published CSV files.