Google Flights Scraper & Fare Intelligence

Extract public Google Flights itineraries and round-trip fares. No API key, login, paid proxy, or external paid service is required. The Actor returns a clean Dataset, a machine-readable OUTPUT summary, and a responsive REPORT.html for fare comparison.

What it delivers

Each valid result includes:

• origin and destination IATA codes;
• requested departure and return dates;
• airline, departure and arrival time, duration, and stops;
• round-trip price and explicitly observed currency;
• flight number and cabin class only when Google exposes them;
• a Google Flights URL;
• a deterministic Low, Typical, High, or N/A value signal;
• evidence warnings when a field cannot be confirmed.

Quick start

{
  "origin": "GUA",
  "destination": "MIA",
  "adults": 1,
  "max_results": 10,
  "currency": "USD"
}

Dates are intentionally omitted from the sample. At runtime, departure defaults to today plus 14 days and return defaults to seven days after departure. This keeps the public sample evergreen and ADWIC-safe.

Other route examples:

{
  "origin": "JFK",
  "destination": "LAX",
  "departure_date": "2026-10-20",
  "return_date": "2026-10-27",
  "adults": 2,
  "max_results": 20,
  "currency": "USD",
  "cabin_class": "economy"
}

Use future dates when providing dates explicitly. Supported cabin requests are economy, premium_economy, business, and first.

Outputs

• Dataset: one typed record per validated, deduplicated fare.
• OUTPUT: status, route, dates, result count, comparable sample, median/range, currency, acquisition diagnostics, and warnings.
• REPORT.html: responsive route summary and fare table with airline, times, duration, stops, price, Value Signal, Google Flights link, methodology, and warnings.

An empty Google response or isolated HTTP failure completes as a clean EMPTY result with diagnostics. It does not create fabricated fares.

Fare intelligence

Fare intelligence runs only when at least three delivered fares have one common, explicitly observed currency.

• Low: price is at or below 85% of the median.
• Typical: price is between the Low and High thresholds.
• High: price is at or above 115% of the median.
• N/A: insufficient or mixed-currency evidence.

The calculation is deterministic and reproducible. It is a comparison of observed results, not a prediction and not a guarantee that a fare is bookable.

Currency and optional evidence

Currency comes from visible Google evidence such as “US dollars” or “euros”. The Actor never infers USD from an airport or country. When currency cannot be confirmed, the record uses null and emits currency_evidence_unavailable.

flight_number and cabin_class also remain null unless the result exposes those values explicitly. The generated Google Flights URL is a search URL, not a fabricated airline booking link.

Acquisition and privacy

The Actor uses Python's standard HTTP client for bounded requests to public Google Flights pages. It has two isolated HTTP URL strategies, a 25-second timeout per strategy, and no browser runtime in the candidate. It does not bypass CAPTCHA, use residential proxies, call paid APIs, or require user credentials or private secrets.

Google can change public markup, localize responses, throttle traffic, or update fares between requests. Verify final availability, baggage rules, itinerary details, and price before booking.

Billing contract

The prepared Pay-Per-Event event is flight-result. A result is billable only after a valid deduplicated record with explicit currency evidence has actually been delivered. Errors, timeouts, blocked responses, discarded candidates, duplicates, empty results, currency-unconfirmed records, and undelivered records are not billable.

The Pay-Per-Event price is USD 0.003 per valid flight-result delivered.

Local QA

Run the deterministic suite from this directory:

& "C:\path\to\python.exe" -m unittest discover -s tests -v

Fixtures are reduced evidence captures from the 2026-09-12 technical gate for GUA→MIA, JFK→LAX, and GUA→MAD. They contain public semantic flight-card text, not competitor code.