The fine print.
A receipt for what the browser downloaded. Not a verdict on the people who built the website.
What goes on the receipt?
Fresh scans open a public page in an isolated Chromium browser. Transfer totals add the encoded response body size and response header size reported by Playwright for completed HTTP(S) requests. Request upload bytes, TLS overhead, WebSocket messages and unfinished transfers are not included. KB and MB use decimal units: 1 KB = 1,000 bytes.
HTML, CSS, JavaScript, images and fonts are grouped using MIME type and browser resource type. Everything else is “Other”. The total equals the sum of the six groups.
What counts as a request?
Requests initiated in the clean browser context are counted, including redirects, failed attempts and requests blocked by the scan policy. Byte totals only cover completed responses. “Domains contacted” counts distinct hosts that returned a response. Third-party classification compares registrable domains against the final page’s registrable domain, including private suffix rules.
The lab setup
The default profile is a mobile browser at 390 × 844 pixels, device pixel ratio 3, en-GB language and UTC time zone. Each fresh scan has a cold browser cache and no user session. The initial connection is unthrottled. The region is shown on every receipt; on a local installation it is your computer’s internet connection.
Navigation waits for DOMContentLoaded for up to 9 seconds. Observation then continues for up to 2.5 seconds, ending after at least 1 second if the network has been quiet for 750 milliseconds. Long-lived pages can therefore have unfinished requests. Only GET and HEAD are permitted. Service workers, downloads, pop-ups and WebSockets are blocked. These choices can change a site’s behaviour.
Performance without pretending
LCP is the last Largest Contentful Paint candidate seen during the observation window. CLS is the largest layout-shift session-window score observed without recent input. Neither is a complete Lighthouse run or CrUX field metric. No interaction is performed, so INP is deliberately not reported. A missing value is “not measured”, never a made-up zero.
Probable, not proven
Technology detection uses a limited, documented collection of asset paths, response headers and runtime markers. False positives and omissions are possible. Hover over a detected technology for the signal that triggered it. We do not claim to discover the entire stack.
AI readiness: show the evidence
The page-signals-1 rubric measures 24 observable checks in six areas: AI citability (22%), content and credibility signals (22%), access and rendering (25%), structured data (13%), brand identity (10%) and platform readiness (8%). Each area gives equal value to its checks; its score is the percentage detected. We round area scores, then combine them using these weights. This is WebReceipt’s own heuristic, not an industry standard or a platform’s ranking formula.
We inspect visible text, original server HTML up to 2 MB, metadata, attribution, dates, labelled source links, about/contact routes and bounded schema.org JSON-LD. A heading, author or date is a signal, not proof of expertise, accuracy or trust. Some link heuristics use English labels. JSON-LD syntax/types are checked; full vocabulary validation and microdata/RDFa are not. Non-empty source text passes the server-text check. A server/rendered text-length ratio of at least 50% passes the rendering-dependence check; that ratio does not prove equivalent content.
A separate robots.txt read has a three-second and 512 KB limit and stays behind the same public-network protections. It is excluded from page transfer and request totals. Rules are checked against the final page URL for OAI-SearchBot and Googlebot; GPTBot training preferences are reported separately without affecting the score. We also inspect Google-relevant noindex and snippet controls. Permissions do not prove that real crawler IPs pass a CDN or firewall. A missing robots file (404/410) has no observed file restrictions; failures or HTML fallback responses remain unknown. Unknown checks show a minimum-to-maximum score range.
We do not query AI answers or verify search index inclusion, citations, brand reputation, factual accuracy or E-E-A-T. Platform readiness covers metadata and page configuration. These checks do not promise AI exposure. See the Google AI search guidance, OpenAI crawler documentation and robots protocol.
The drawer, not another fresh scan
Results under 6 hours old are reused. Results 6–24 hours old are reused unless you request a refresh. Older results trigger a new scan when the URL is submitted again. Existing public receipt URLs remain historical snapshots. Cached receipt views do not consume your fresh-scan quota. When the scanner version changes, submitting the website starts a scan with the new checks rather than reusing an older version.
What we do not sell as science
No overengineering scores, invented percentiles or model-generated judgements. Jokes come from deterministic payload rules. There is no benchmark comparison until a verified, named dataset is configured.
Traffic is a separate measurement
A scan does not reveal another website’s visitor counts. The owner dashboard and media kit measure WebReceipt’s own consented browser traffic. Placement impressions use viewport visibility and time, not page requests alone; test campaigns and empty inventory never count as paid delivery.
Paid coupons are paid coupons
All sponsor placements are labelled Sponsored. The campaign active when a new receipt is printed is preserved with that receipt and its share image. Subsequent campaign changes do not rewrite history. External reposts cannot be reliably counted as ad impressions.