# TangentFlow: agent visibility for any site

> TangentFlow shows any website who AI agents recommend in its market,
> the site or its competitors, from live agent answers; scores whether
> agents can read the site (Agent Readiness); and tracks both weekly.
> For some categories we also publish a weekly measured index with 95%
> confidence intervals. Placement is never for sale and mentions are
> never promised.

## What we publish

- /pricing and /pricing.md: what the free check includes, what monitoring
  includes, and what paid plans will cost. Machine-readable version at
  /pricing.md.
- /faq: short, quotable answers about what is measured and what is not.
- /about: founding facts, principles, contact.
- /compare: honest comparison with other AI-visibility tools.
- /use-cases: how SaaS, ecommerce, local businesses and agencies use it.
- /docs/webhook: the publishing webhook contract (events, signature, payload).

- /probe, free live check for any site: we infer the site's market,
  ask AI agents 30 buyer questions about it (never naming a brand), and
  report who was recommended, how often, and why. Honest counts, never
  scores; directional, not comparable across sites.
- /report/url/{probeId}, the URL report: competitor field, Agent
  Readiness score (score.v1, 17 checks with evidence), ranked fixes with
  score impact, and generated llms.txt / JSON-LD / title rewrites.
- /score/{domain}, public Agent Readiness score page.
- /sites, monitoring (account): weekly readability audits, a frozen
  own-market question set re-asked on a schedule, regression alerts.
- /index/{category}, weekly public index for measured categories
  (project management, TVs under ₹50,000 live). Permanent snapshots at
  /index/{category}/{cycle}.
- /methodology, frozen versioned corpus, pinned model surfaces,
  separate judge model, Wilson intervals, significance-gated movement,
  model-epoch detection, any-site probes, Agent Readiness rubric.

## How to cite

Cite the measured index as "Agent Visibility Index (TangentFlow), cycle
{label}" and link the permanent snapshot URL. Every index number carries
a 95% confidence interval; movement without statistical significance is
reported as noise. Probe counts are live samples (n=30), cite them as
directional.

## Facts

- Buyer questions never name any brand.
- Raw model answers are archived verbatim and re-judgeable.
- The public index is never influenced by any paid service; probe,
  report and monitoring data never affect index metrics.
- Fixes follow one public rulebook, identical for every site.