How search works

Every AirportFusion search runs a five-stage pipeline. Understanding it helps you get better results.

Stage 1 — Geocoding your addresses

Your two inputs are converted to coordinates using OpenStreetMap's Nominatim geocoder (or another configured provider). If external geocoding is unavailable, we fall back to matching city and airport names in our own database, so search keeps working. You can also bypass geocoding entirely by typing coordinates such as 40.71, -74.00.

Stage 2 — Finding airports in each zone

We look up every active airport with scheduled service inside the radius you chose around each point, using great-circle (haversine) distance. Airports are sorted nearest-first and capped at the 25 closest per zone — enough to cover any metropolitan region plus its secondary airports.

This is the core idea of AirportFusion: the "best" airport is often not the nearest one. A slightly farther airport frequently has the only direct route, or a dramatically cheaper one.

Stage 3 — Matching direct routes

We then intersect the two airport sets against our route network — built from the OpenFlights public dataset — keeping only non-stop routes that start in your origin zone and land in your destination zone. Why direct only? Because connections multiply total travel time and uncertainty, and comparing a 2-leg itinerary door-to-door against a non-stop one is rarely a fair fight. If no direct route exists, we say so honestly rather than padding results.

Stage 4 — Building door-to-door routes and scores

For each direct route we assemble the full journey:

Ground legs — for each side we estimate car, taxi, rideshare, train, metro (short distances) and bus options, from distance-based models. The fastest sensible option is used for the totals.
Flight leg — duration is estimated from great-circle distance at typical cruise speeds plus taxi/climb time; price is estimated from a distance-based model with a stable per-airline variation (so the same search always gives the same estimates).
Score (0–10) — a weighted blend: 40% total time, 40% total cost, 20% simplicity (fewer transfers, shorter ground legs). Scores are normalised within your result set, so a 9.2 means "excellent among these options", not a universal grade.

Routes are ranked by score, best first, capped at 50 options.

Stage 5 — The AI advisor (optional)

When enabled, a compact summary of the top routes (airports, times, costs, scores — no personal data) is sent to an AI model that acts as a travel expert. It picks the route it would book, explains the trade-offs in plain language, and adds one practical tip. The AI never sees data that is not already on your screen, and its advice never overrides the transparent score ranking — it is a second opinion, not a black box.

What this means for your searches

Widen the radius when results are thin — the search is designed for it.
Watch the ground legs — a "cheap" flight from a far airport can lose on the total.
Treat numbers as estimates — see "Understanding estimates" for exactly how they are computed.