How the free Help Center health check works.
See how CheckMyHelpCenter discovers public articles, flags selected policy conflicts and content overlap, checks freshness, and calculates its health score.
The free audit is designed to produce explainable review signals from public content. It uses bounded discovery and deterministic analysis so every reported issue can be traced to the pages that triggered it.
- Collection
- Bounded HTTPS crawl
- Analysis
- Explainable rules
- Numeric score
- Requires ≥5 pages
- Reliable coverage
- Requires ≥60%
1. Public-page discovery
The audit requires a public HTTPS URL. It follows a bounded set of same-hostname links, reads only pages available without authentication, and never submits forms. The submitted URL and every redirect are rechecked for protocol, hostname, credentials, ports, local-use hostnames, and explicit local or reserved IP addresses.
The application does not independently resolve DNS names before every connection, so these URL checks should not be interpreted as a claim of complete network isolation.
A scan can analyze at most 20 pages. Each response is limited to 1 MB, the total download is limited to 8 MB, and a request can follow at most five redirects. Coverage is reported because an audit cannot describe pages it could not reach.
2. Policy conflict signals
The current free audit extracts selected policy values such as free-shipping thresholds and return windows. A high-risk signal requires different values on different pages with a matching policy type and scope.
Scope matters. A rule for sale items can coexist with a different rule for full-price items, so a person should review every flag in context.
3. Near-duplicate content signals
The current detector is deterministic rather than embedding-based. It normalizes text, removes common words, builds word features and five-character shingles, then compares Jaccard and containment overlap. A pair is flagged at a 72 percent score, or when title containment is at least 75 percent and body-token containment is at least 55 percent.
High overlap indicates that two pages may be near-duplicates. It does not prove semantic equivalence and is never an automatic instruction to merge them.
4. Freshness and topic coverage
An article is treated as a freshness review candidate only when the page exposes a usable update date at least 365 days old. A missing date is reported as unverifiable, not as proof that the article is stale.
Public topic coverage checks common ecommerce themes. They are not ticket-based knowledge-gap mining and do not reduce the displayed content-health score.
5. Directional health score
The displayed score begins at 100. It subtracts 14 points per potential conflict, up to 42; 6 points per duplicate pair, up to 18; and 4 points per freshness signal, up to 16. A numeric score is withheld when fewer than five pages are analyzed or when successful coverage is below 60 percent.
The score describes the scanned public content. It does not measure an AI agent's production accuracy, private knowledge, ticket resolution rate, or customer satisfaction.
Questions about this audit.
Is every flagged item a confirmed error?
No. Findings are automated review signals. Different statements may both be valid when their region, product, timing, or eligibility conditions differ.
Why might the report omit a numeric score?
A number is withheld when fewer than five pages are analyzed or when the scan covers less than 60 percent of discovered pages.
Does topic coverage affect the health score?
No. Public topic coverage is shown as context and is not treated as ticket-based knowledge-gap analysis.