Getting started
What GenAuthority measures
The five engines, what a scan records, and what it deliberately refuses to guess.
GenAuthority answers one question: when someone asks an AI about your market, what comes back?
To answer it we take the questions your buyers actually ask, put them to five AI engines on a schedule, and record every answer in full along with every source each answer cited. Everything else in the product is arithmetic on top of that record.
The five engines
Every plan scans all five. There is no tier where you get three engines and pay extra for the rest.
| Engine | What we ask | What it reports back |
|---|---|---|
| ChatGPT | The prompt, through the live ChatGPT interface | Answer text, cited sources, any ads shown |
| Google Gemini | The prompt, with web grounding on | Answer text, cited sources, the searches it ran |
| Anthropic Claude | The prompt, with web search enabled | Answer text, cited sources |
| Perplexity | The prompt | Answer text, cited sources |
| Google AI Overviews | The prompt as a Google search | The AI Overview block and its sources |
Each engine gets asked independently. We do not ask one model to guess what another would have said.
What one scan records
For a single prompt on a single engine we store:
- The full answer, as text. Not a summary of it.
- Every cited source, deduplicated by URL, with its domain.
- Which brands were named, yours and your competitors’, by matching against the brand names and aliases you configure.
- Where in the answer each brand appeared, which is what position and share of voice are computed from.
A scan of 100 prompts across five engines is 500 answers. We call one prompt-on-one-engine a prompt-run, and that is the unit your plan meters.
What we refuse to do
This is a measurement product, so the failure mode that matters most is producing a number that looks real and is not. A few rules we hold:
- An unmeasured brand has no score. Not zero, not position 50. It shows as
nullor “Unknown”. - A headline percentage is withheld when fewer than half the intended engine calls came back. A visibility score built on two engines out of five is not a visibility score, and rounding it up to look complete would be lying.
- Sentiment is never inferred from absence. If no answer discussed your brand, there is no sentiment to report.
- We say when an engine goes quiet. If a scan completes but one engine returned nothing at all, that is surfaced rather than silently absorbed into a lower score.
What happens after a scan
The raw record feeds four things:
- A visibility score: one number per brand, per period.
- Share of voice: you against the competitors you named.
- Citation gap: sources that cite a competitor and have never cited you.
- Next best actions: an ordered list of moves, each one traceable back to a specific prompt, citation or competitor.
The software does the calculating. Language models are used to explain and to classify, never to invent a measurement.