How Depais analyzes websites.
Depais combines automated checks of publicly available website infrastructure with community reviews. Technical signals inform the confidence score; user reviews remain a separate measure.
What we check
Depending on what a website exposes, an analysis may inspect HTTP behavior, TLS configuration, DNS and hosting details, page metadata, crawlability, structured data, response performance, detected technologies and public risk indicators.
External dynamic analysis
When the optional Cloudflare URL Scanner integration is enabled, Depais submits only the normalized public homepage root: HTTPS scheme, public hostname and root path, with no credentials, query string or fragment. Untrusted JavaScript is not executed on Depais infrastructure. Depais requests desktop and mobile captures from Cloudflare and stores only a bounded desktop PNG as the report image.
Scans use Public visibility by default. Public scans may appear in Cloudflare's recent-scan list and search results. An operator may configure Unlisted only when the account supports it. In either mode, Cloudflare processes the public URL, result and screenshots.
Dynamic and reputation observations are point-in-time evidence. Missing data, quota limits and provider errors are never presented as proof that a website is clean or safe.
How the technical score works
Each completed deterministic check can contribute a defined number of points. Positive, neutral and concerning observations are combined into a technical score from 0 to 100. The report shows the underlying evidence so the number never has to stand alone. External scanner verdicts and AI risk indicators carry zero technical-score weight.
It is a summary of observable signals, not a certification, guarantee or accusation. A valid TLS certificate protects a connection; it does not prove that the organization behind a website is legitimate.
AI website assessment
After deterministic analysis completes, Depais may prepare a separate AI analysis of the website's purpose, audience, content quality and transparency. It uses bounded title, metadata, headings, visible text snippets and selected deterministic findings. Raw HTML is not supplied, and the assessment has no browser, URL-fetching tools or direct access to the website.
The AI content score evaluates purpose clarity, audience clarity, content quality, transparency and information structure. When validated output has at least 60% confidence, the displayed Depais composite uses 75% technical score and 25% AI content score. Otherwise the composite remains technical-only. Community ratings are never part of either formula.
Classification
Depais uses a maintained two-level taxonomy. An assessment can suggest one primary category at 60% confidence or higher and up to three distinct secondary categories at 70% confidence or higher. Insufficient evidence is recorded as Other and Unclear. Administrator or manually locked classifications always take precedence and cannot be overwritten by AI output.
AI risk indicators
A validated assessment may publish a small set of neutral, evidence-cited risk indicators at 70% confidence or higher. These indicators are separate from every score. A high-severity presentation requires corroboration from a stored deterministic security, malware or external scanner finding; without corroboration it is labeled an unverified AI indicator.
Risk indicators do not establish malware, fraud, criminal conduct, legality, trustworthiness or safety. Unsupported wording, invalid evidence references and output that violates the strict schema are discarded.
Assessment technical details
The assessment request uses a fixed system instruction, untrusted-data delimiters, a strict JSON schema, bounded evidence and output limits. It cannot call tools. Published API provenance includes the configured provider, model revision, runtime and extractor, prompt, schema, taxonomy, risk and scoring versions.
What reviews measure
Community ratings reflect users' stated experiences. They are not used to calculate technical confidence. Depais does not copy ratings from other review platforms.
Freshness and limitations
Websites change continuously. Every report identifies when evidence was observed. A site may block automated requests, return different content by location or change immediately after a scan. Missing evidence is shown as unavailable rather than guessed.
Corrections
Anyone can report an inaccuracy. Verified representatives can provide corrections and official context, but ownership verification does not change ratings or bypass moderation.