HomeMethodology
How a finding is labeled
Every public claim on Findora AI should be traceable to a label. Observed, Inferred, and Hypothesis describe evidence. They are not a visibility score. Public-page audit plus Grok model checks. We do not query ChatGPT, Claude, or Perplexity yet.
Observed
Observed means text the audit fetched from a public URL, shown with a quote and that URL. You can open the page and look for the same words. Observed does not mean an assistant said it. A field is called verified only when the audit service already labeled it that way. This screen does not promote a field on its own.
Inferred
Inferred means a reading of pages the audit fetched. Competitive-gap rows can carry the service label Inference. Inference is not a measurement of what ChatGPT, Claude, or Perplexity would answer. It is an interpretation of the public text, and the report says so. A gap is not a share of voice.
Hypothesis
Hypothesis means a suggested explanation the fetched pages did not confirm. Hypothesis stays Hypothesis. Findora AI does not relabel it because a later screen wishes it were confirmed. If you want it confirmed, the path is a later fetch that matches the acceptance evidence. This version does not run that rescan.
Grok check status
A Grok model check is shown as the service returned it, including a failure and its reason, such as a spending limit. We do not query ChatGPT, Claude, or Perplexity yet. Mention monitoring and recommendation monitoring are not live. When a check did not run, no rate was calculated. An empty result is not shown as zero, and it is not a recommendation.
The product is not multi-model recommendation monitoring. AI Visibility Growth remains Coming — monitoring not live yet — and checkout for that plan stays disabled. A one-shot Grok check is not that monitoring plan.
What a free run returns
The free summary includes the checks that ran, top findings, and a short evidence quote. Full findings, the fix pack, and acceptance checks are the one-time Fix Pack. Suggested drafts stay marked DRAFT in this browser and are not written to your site.
What we refuse to publish
No fabricated customer counts, no before-and-after recommendation claims, and no screenshot of an assistant we did not query. The sample report elsewhere in the product is labeled as a sample and is separate from this methodology. It does not call a website, and it is not a live scan.