How a Clarion audit works.

For every audit, our crawler fetches the homepage and up to two internal pages at both desktop and mobile viewports. We collect rendered HTML, full-page JPEG screenshots, real Core Web Vitals from a PerformanceObserver inside the page, the count and size of outbound network requests, and the list of third-party domains the page contacts. We also check for the presence of HTTPS, robots.txt, sitemap.xml, and ads.txt.

None of this is invasive. We act like a normal browser visiting your homepage once, in a single short session. No persistent cookies. No login.

The crawled data is fed to a vision-capable language model. It returns structured judgments on twelve dimensions:

• Content originality. Is this written by a specific human with a specific point of view.
• Design quality. Visual taste, hierarchy, restraint.
• Editorial voice. Is a distinctive voice present.
• Purpose clarity. Could a first-time visitor describe this site in five seconds.
• Niche specificity. Specialists earn more per impression than generalists.
• Content depth. Substance vs. surface.
• Headline quality. Specific and honest vs. clickbait.
• AI-generated likelihood. A direct estimate of how machine-written the prose reads. Specific phrase patterns and structural tells are listed openly.
• Audience inference. A named guess about who reads this site, plus likely geography and visit intent.
• Brand-safety category. The categories the site falls into and any restricted-content signals.
• Made-for-advertising signals. Whether the site exists primarily to host ads.
• Conversion intent. Content-first, balanced, affiliate-heavy, or sales-first.

Every score is accompanied by the model's reasoning, anchored against calibration examples we hold fixed across audits.

A weighted sum of all components produces an overall score from 0 to 100. The tier mapping:

• Excellent (88–100): premium yields realistic.
• Strong (72–87): good network fit.
• Standard (48–71): admitted at a conservative tier with specific improvements identified.
• Needs Work (0–47): not ready; the report explains why and what to fix.

Certain conditions cap the tier regardless of score: high AI-text risk, dormant publishing cadence, high-severity made-for-advertising signals, and high-severity brand-safety flags. When a cap kicks in, the report names the cap honestly.

We start from a tier base range (Excellent: $8–$25, Strong: $4–$12, Standard: $1.50–$6, Needs Work: $0.30–$2). The range is then adjusted up or down for content category (finance and tech premiums, crypto and gambling discounts), audience geography (US, UK, Canada, Australia lift the multiplier), specialist niche, visit intent, AI-text risk, and conversion intent.

The projection is honest about uncertainty. Unless you connect Google Analytics or another verified audience source, the result is a range, not a point estimate.

We will not assign a higher tier than the model supports because the publisher asked. We will not hide the components of the score. We will not punish a site without naming exactly what we found. And we will not produce hand-wavy "positive trend" or "market-leading" copy in a report — every sentence has to point at something specific in the data.

For sites that install the Clarion tag, evaluation does not stop at the initial audit. The tag continues capturing real performance and audience data, and the scoring engine re-runs on a rolling cadence. If your site improves, your tier rises. If it degrades, we tell you why and demote with notice. This is the opposite of the AdSense pattern of one-time approval and silent enforcement.

This methodology is v0.1 and will evolve. Known limitations as of today:

• External authority signals (domain age, backlinks, organic traffic) are not yet wired. The fields exist in the audit schema; the provider integration is on deck.
• Originality checks against scraped-content fingerprints are run via the vision model rather than a dedicated plagiarism index. We'll add that when the volume justifies it.
• The 15–30% eCPM lift quoted for the Verified Audit upgrade is grounded in standard industry mechanics, not in our own network data — because we don't have a network yet. We will replace this with measured results from our actual demand stack as soon as we have them.