AI Website Change Intelligence

Monitor public pages without drowning in cookie banners, timestamps, duplicate text, and raw line diffs. This Actor deterministically normalizes each page, saves a stateful baseline, classifies meaningful content and structural changes, and returns importance, confidence, alerts, and before/after evidence. The MVP does not call an external AI model and requires no API key.

What you get

• Meaningful content-added, content-removed, and content-modified events.
• Heading, link, and form-structure changes.
• Pricing, product, availability, legal, contact, or general classification.
• A 0–100 importance score, confidence score, severity, bounded alerts, and evidence.
• Default suppression for scripts, styles, cookie UI, duplicate lines, generated timestamps, hashes, and copyright-year churn.
• A persistent baseline in your named key-value store for scheduled runs.

Input options

Use any combination of:

• targets[].url for a public HTTP or HTTPS page.
• targets[].currentContent plus optional previousContent for private, local, or already-fetched HTML/text.
• datasetId for a dataset containing the same target fields.
• No input for the built-in before/after product-page sample.

Per-page ignoreSelectors and noisePatterns handle known dynamic widgets. sensitivity controls how readily removed and added lines are paired as one modification. includeEvidenceText: false removes full normalized text while retaining per-change snippets.

Stateful behavior

The first successful URL check creates a baseline. A later check returns unchanged or changed. Supplying previousContent performs an immediate comparison and takes precedence over stored state. resetBaseline replaces state without turning the replacement into an alert. State is updated only after a successful output record and successful pay-per-event charge.

Charging and failure safety

A one-time $0.0004 Actor-start event covers measured platform startup and URL-safety validation. The primary page-change-analyzed event is charged only after a page is fetched or accepted inline, normalized, compared, and prepared as a useful result. Invalid URLs, blocked private-network targets, HTTP errors, oversized responses, empty pages, and other failed pages return dataset diagnostics with no value-event charge. The OUTPUT summary is written separately and uncharged. The Actor checks the run's PPE budget before work and stops cleanly when the event limit is reached.

Network and privacy safety

HTTP fetching is bounded by time, response size, content type, and five redirects. Every initial URL and redirect is revalidated. Localhost plus private, loopback, link-local, carrier-grade NAT, benchmarking, multicast, reserved IPv4, unique-local IPv6, and link-local IPv6 destinations are rejected. URL mode reads a public page; inline mode lets you avoid network access. Normalized evidence and snapshots are stored in your Apify storage—use a private run and disable evidence text when content is sensitive.

Output

Each dataset row contains status, page identity, source, fingerprint, change, evidence, warnings, error, pricing model, and valueEvent. change.details contains typed before/after evidence. change.alerts contains only changes scoring at least 75. OUTPUT contains uncharged run counts and budget status.

Limitations

This HTTP-first MVP does not execute client-side JavaScript, log in, bypass bot defenses, or interpret images. Deterministic semantic heuristics are transparent and inexpensive but may miss paraphrases or personalized content. Verify evidence before consequential action. Only monitor pages you are allowed to access and follow applicable terms, privacy rules, and laws.

Local verification

npm install
npm run typecheck
npm test
npm run build
npm run benchmark
npm run launch:verify