When buyers ask ChatGPT, Claude, Gemini or Perplexity what to buy, the engines cite “sources.” We resolve every one of those sources to its owner. In the first category we audited — AI tools for lawyers — the answer was uncomfortable.
28-day window · 63 sampled answers · 4 engines · 442 resolved citations · baseline across our full 143,424-citation index: 46.2% vendor-owned. Method public: methodology.
For every ranked vendor we compute a self-sourced recommendation rate: of the AI answers recommending it, how many also cite the vendor’s own domain. The public version is anonymized — the audit names every row.
| Vendor | Self-sourced rate | Reading |
|---|---|---|
| Vendor A | 100% | every recommendation accompanied by its own pages |
| Vendor B | 100% | same pattern |
| Vendor C | 86% | near-total self-dependency |
| Category leader 1 | 20% | recommendation earned mostly off-domain |
| Category leader 2 | 31% | majority independently sourced |
The same product’s recommendation share diverges by 28–48 percentage points between engines. Single-engine anecdotes mislead; the audit maps all four, with confidence intervals.
The same instrument, pointed at your category or your brand — every vendor named, every source mapped, the evidence retained for verification.
Turnaround: 5 business days. Every report carries instrument disclosures — if a number moved because our instrument changed, we say so. Orbator’s own product categories are permanently excluded from this series. No pay-to-fix bundling: an audit is never sold with placement services on the same category.
An 8-section forensic report (PDF + private copy): the measured recommendation ranking with confidence intervals, the source-ownership map of the most-cited pages, named self-sourced recommendation rates for every ranked vendor, a per-engine divergence matrix, methodology and instrument disclosures. Plus a CSV evidence pack (every resolved citation and run), a 45-minute analyst walkthrough, and one 30-day refresh of every table.
We do not scrape rankings or social mentions. We sample the AI engines directly with neutral buyer questions, retain the full answers, resolve every citation to its owner, and publish the method. The deliverable is evidence — traceable to retained answer texts — not opinion.
Five business days from intake. Every audit starts with a registry hygiene pass on your category so duplicate entities cannot distort the numbers — that step is logged in the report.
Not as a bundle — deliberately. An audit sold with placement services on the same category stops being an audit. Orbator sells visibility tooling separately and at published prices, our own product categories are permanently excluded from this series, and any relationship with a named vendor is disclosed in the report.
We tell you. Every report carries instrument disclosures — sample sizes per engine, engine staleness, extraction upgrades — because a metric that moved for instrument reasons must never be sold as market movement.