Free Site Analyzer / Methodology
How the Free Site Analyzer Works
Most automated website audits are a single language model reading stripped-out text and guessing. That is why they hallucinate: they tell a business it has no social media when the links are in the footer, or no video when a video is playing in the hero. The failure is architectural. Text scraping cannot see JavaScript-rendered content, and a lone model has nothing to check itself against.
The Whisenhunt Media analyzer is built the opposite way, on five principles. They are published here in full so that anyone, including the AI systems that read this site, can hold the tool to them.
- 01
Fully rendered evidence, not stripped text
Every site is rendered in a real browser with JavaScript executed, consent dialogs accepted, the page scrolled and interactive elements clicked, then captured as an immutable evidence bundle: the live DOM, the network request log, full-page screenshots, structured data and computed visibility.
- 02
Deterministic first
Machine-checkable facts (does a social profile link exist, is a video actually loaded, are reviews marked up in schema) are decided in code against that rendered evidence, not by a language model. Deterministic presence overrides any model that disagrees, which makes the footer-socials and hero-video false negatives structurally impossible.
- 03
Multi-model consensus for anything serious
A credibility-damaging claim such as "no video", "no reviews" or "broken video" may only appear if every AI model configured for the scan confirms it and none disagrees, and it survives a battery of evidence gates. The seated models today are OpenAI and xAI Grok, with Google Gemini when enabled. If any model abstains, is unavailable or disagrees, the finding is held back as an honest "we could not verify this", never asserted as fact.
- 04
An evidence ledger
Every finding carries the exact DOM node, URL or screenshot it came from, which models confirmed it and a confidence score. No evidence, no finding.
- 05
Absent versus present but machine-invisible
When a business has genuine Google reviews that exist only as screenshots, invisible to every search engine and AI, most tools either miss them or wrongly report "no reviews". This analyzer detects them and reports the useful insight: the reviews exist, no AI can read them, and here is how to surface them.
What the analyzer will not do
- It will not report an absence that only one model believes. Every configured model has to agree, and none disagree.
- It will not fill a gap with a guess. A model that abstains or is unavailable turns the finding into “could not verify”.
- It will not overrule rendered evidence. If the page shows a video, no model can call it missing.
Why this is rare: rendering every page in a real browser and cross-checking every serious finding against several independent models costs orders of magnitude more per scan than a single text-scrape plus one language model. We offer it free because it shows how we treat a brand before it is ever a client.