About Visible Diagnostic

Who runs it

VisibleDiagnostic is built and maintained by Shangyi Duan, an independent developer. The same person designs the question set, runs the samples, reads the answers and writes the report. There is no account team, no junior hand-off and no dashboard to log into.

The practice publishes its method, its limits and, since 2 November 2026, the first 45-answer variance pilot behind the work. If a claim on this site cannot be traced to a dated answer, a public data file or a published price, it does not belong here.

Why it exists

Buyers now ask ChatGPT, Perplexity and Google for a shortlist before they open a vendor's site. Those answers move: in our pilot, only 6 of 15 buying questions returned the same brand set in all three runs, and the “how should I choose” questions had no stable first-mentioned brand in 5 of 5 cases. A single screenshot cannot show that, and a dashboard that averages it away cannot either. So the deliverable is a dated evidence document: ten buying questions, three logged-out runs each over three days, thirty dated answers, a source audit, three prioritised actions and a day-60 re-run — built to be forwarded to a marketing lead and argued with line by line, not to produce a number that looks precise and cannot be defended.

How the data works

The free check on the homepage is a demonstration, not the paid measurement. It turns the brand, category, audience and up to two competitors into three questions, sends them to Perplexity Sonar through the OpenRouter API, and scores the returned citations with the heuristic in api/check.js. One run per question; the answers are shown in the browser and are not stored.

The paid diagnostic is 10 questions × 3 logged-out runs = 30 dated answers, delivered within five business days of the kickoff, with a day-30 execution check and a day-60 re-run included. The published prices are $850 for the audit, $350 per month for monitoring after a baseline, and $600 per white-label report. The free-check surface and the paid sampling surface are different, and the method note says so.

The public pilot is 5 B2B SaaS categories × 3 question types × 3 runs = 45 answers. Its raw CSV is published at ai-visibility-variance-pilot-2026-11-01.csv, and every number on the research page is recomputed from that file at build time. Read the methodology for the sources table, the numbered steps and the three known limitations.

What this site is not

• Not a rank tracker. There is no stable “AI rank” across sessions, locations and models, and this site does not sell one.
• Not affiliated with OpenAI, Perplexity, Google, OpenRouter or any tool named on this site. No vendor reviews or approves this work.
• Not a promise of a citation, recommendation, ranking, traffic or revenue. The evidence is a record of sampled answers, not a forecast.
• Not a client showcase. RelayHQ is fictional; the sample report is synthetic and labelled on every page. No logo, testimonial or result appears without written permission.
• Not a content, link-buying or outreach shop. The diagnostic names the work; it does not execute it.
• Not a newsletter and not a checkout. There is no payment until fit is confirmed in writing.

Data we keep

Contact requests and free-check result emails are stored for up to 180 days with the email, brand, website and note you submit, plus the score, source, timestamp and user agent. Aggregate event counters are kept for up to 120 days. Free-check answer text is never written to the lead store. The retention values come from worker/worker.js, not from a template policy; the privacy page lists the processors.

Corrections and contact

If a page or a data file is wrong, use the corrections page to report it with the URL and a source; every accepted correction is logged with its date and evidence. For a diagnostic request, use the contact form. Both reach the same person. We reply within one business day. The dated record of site and data changes is on the changelog.

Last updated: 2026-11-02

Questions