The Visibility Audit produces a heuristic score from public-resource checks and generated assessments. It is a diagnostic aid for reviewing website information. It is not a citation probability, search rank, market-share estimate, or census of answers across AI services.
Inputs and nominal weights#
The current audit uses the model identified in the report; the implementation reviewed on November 1, 2026 uses gpt-5 with web search enabled. Each pillar receives a component score, and the stored breakdown contains its weighted point contribution.
| Pillar | Nominal weight | Inputs used |
|---|---|---|
| Foundation | 10 points | Successful robots and sitemap fetches on the primary and derived AI host, plus AI-host llms.txt. |
| Entity | 20 points | Generated alignment estimate, summary-length heuristic, market-hypothesis completeness, and reachable evidence links. |
| Market | 25 points | Presence/position in a generated vendor list, category overlap, list completeness, reasons, and evidence reach/diversity. |
| Inclusion | 25 points | Mentions and primary-host citations in generated question/answer samples, answer-length heuristic, and evidence reachability. |
| Fanout | 20 points nominally | Generated llms.txt presence/syntax judgments and awareness, evaluation, and decision coverage. |
The total is the rounded sum of weighted contributions. For example, a component score of 60 in the 20-point entity pillar contributes 12 points. The breakdown values should therefore not be read as independent percentages.
Sampling and evidence checks#
The audit asks the model for a company assessment, a market/vendor assessment, a set of approximately ten questions and answers, and proposed new content. These generated questions are suggestions and samples; they are not observed private queries from real users.
Link checks fetch a bounded set of supplied URLs and record success or failure. A successful fetch does not validate the truth of the associated claim. Sitemap discovery uses the declared sitemap or a default sitemap location, follows a bounded number of files, and caps collected URLs. Large or unusual sitemap structures may be incompletely represented.
Current scoring limits#
The audit derives an ai. companion hostname from the primary host's last two labels. That assumption can be inappropriate for standalone sites, public-suffix structures, and organizations using another companion hostname. Missing resources on that derived host can reduce foundation and content-coverage results even when the real site publishes valid discovery resources.
The current fanout calculation uses only part of its nominal scoring range. Its contribution can reach 10 of the nominal 20 points; with all other components at their maximum, the implemented total reaches 90. Do not pursue 100 as an attainable completion requirement for this version.
Use the stored date, model, domain, evidence, and individual recommendations when comparing runs. Provider behavior and generated assessments can vary. Validate changes with public-page checks and separately measured analytics. Contact support when a checked host or scoring assumption does not fit your site.