PriceCensus

Methodology

How we track AI pricing changes, score confidence, and publish data.

What we track

PriceCensus monitors publicly available pricing pages for AI tools including foundation model APIs, coding assistants, image generation tools, writing aids, and productivity apps. We currently track 6 vendors across 4 categories.

For each vendor we track:

  • Plan names and tiers
  • Prices per billing period (monthly, annual, usage-based)
  • Free tier availability and limits
  • Quotas, rate limits, and credit allocations
  • Trial information
  • Feature availability changes

How we detect changes

Our scraper runs every 6 hours and fetches each vendor's pricing page. We compute a content hash of the page — if unchanged since the last snapshot, we skip processing to avoid false positives.

When content changes, we run our extraction pipeline:

  1. Fetch — HTTP scraping (static) or headless browser (dynamic JS pages)
  2. Extract — AI-powered extraction converts raw HTML into structured plan data
  3. Diff — New plan data is compared against the current known state
  4. Classify — Change type and confidence score are assigned
  5. Review — Every detected change enters a human review queue before it can be published

Confidence scoring

Every detected change receives a confidence score from 0 to 100%:

90–100%
Fast-track review. High-signal changes are easier to verify, but still require approval before publication.
70–89%
Queued for review within 24h. Ambiguous signals that need verification.
0–69%
Manual review required before publishing. Conflicting signals or major page structure changes receive the highest scrutiny.

Factors that reduce confidence include: ambiguous currency or unit formatting, significant page structure changes from the previous snapshot, conflicting price signals on the same page, and AI extraction uncertainty flags.

Correction policy

We aim to be accurate, not just fast. If you spot an error in our data:

  • Prices we publish that are outdated will be updated on the next scrape cycle
  • Material errors (wrong vendor, wrong direction) are corrected and the original event is marked
  • We don't silently delete corrections — each fix is timestamped

Data freshness

Our pipeline checks each vendor every 6 hours. The last checked timestamp on each vendor page shows the most recent successful check. Published changes appear on the site only after a reviewer approves the evidence for the detected change.

PriceCensus is an independent tracker. We are not affiliated with, endorsed by, or compensated by any of the vendors we track. All data is sourced from publicly available pricing pages. Prices may have changed since our last check — always verify at the vendor's official site before making purchasing decisions.