Methodology
How BookWithAI treats fares and reliability
BookWithAI does not put a verdict on a fare. It shows you context you can check, such as the other results in the same search and how a fare you saved has moved, and for US flights an on-time figure built from US Department of Transportation data. This page says how each is produced and where it stops.
Last reviewed against the product's behaviour: October 11, 2026.
Is this fare good? What BookWithAI says, and what it does not
BookWithAI does not label a fare as low, typical, high, good or bad. It does not compare a fare with how the same route has been priced in the past, and it makes no forecast of future prices. A single search cannot support those claims, and the product does not make them.
What it shows instead is context you can check yourself:
- The cheapest of the results in that search. A fare is only counted for "cheapest" if its price is in the currency of the search and is a real price, and on a premium-cabin search (premium economy, business or first) only if the itinerary is confirmed to be in the cabin you asked for. A fare in a different currency, or in a cabin that does not match, can appear in the results but is never called the cheapest.
- The cheapest date among those it checked. When you are flexible, it checks a spread of dates around yours and shows how the price changes. It does not check every possible date.
- How a saved fare has moved. If you save a fare, its price is refreshed periodically when you visit your Deals page (only prices more than about eight hours old, a couple per visit, shown on your next visit), and you can refresh one anytime with Refresh price. BookWithAI shows how the price has moved since you saved it. Nothing is re-checked while you are away and nothing is sent to you.
None of these says a price is a bargain. They narrow the question; the decision, and the final check of the price at the seller, stay with you.
Google's price labels belong to Google
Google Flights shows a label on some fares saying whether the price is low, typical or high for the route, with a typical price range. That judgement is Google's, made with Google's own data and method. BookWithAI does not reproduce it, restate it as its own, or vouch for it. It can be a useful second opinion; treat it as Google's.
BookWithAI's flight search reads Google Flights results through a flight-data provider, and those results can carry Google's price insight for the searched route: a price level, a typical range, the lowest price and the number of history points. BookWithAI does not show that insight on flight cards or call a fare low, typical or high because of it. When it is present on an economy search, BookWithAI can record a summary of it, keyed by route, travel month and currency: the lowest price seen, the low and high ends of the typical range Google gave, and how many times it has been captured.
That record only grows as searches happen, so for most routes it is thin. It is used for one narrow purpose: an answer about the cheapest travel months seen so far on a US route, and only once at least three travel months have been recorded in one currency. Otherwise nothing is said. In that answer, each month's "from" price is the low end of the typical price range Google gave for that month, and it is Google's figure, not a BookWithAI price or a fare you can book. It is a tally of what searches have seen, not a forecast and not a verdict on any fare.
Reliability: where the on-time figure comes from
For US flights, the on-time figure comes from the US Department of Transportation's Bureau of Transportation Statistics (BTS) monthly on-time performance reports, which US airlines file by route. BookWithAI loads these by carrier, origin, destination and month.
- How a flight is counted. A flight counts as on time unless it arrives 15 minutes or more late. Cancelled and diverted flights count as not on time. This is how BookWithAI counts; it is stated here so the number is not mistaken for a different measure.
- What the figure is. It is that airline's on-time share on that route for the most recent month loaded for that airline and route, which can be older than a couple of months. The airline used is the operating airline where we know it, otherwise the airline selling the flight. The figure is for the airline on that city pair, not necessarily for that exact flight or routing: it is not the record of the specific flight you are looking at, and a connecting itinerary can be matched to the airline's figure for the same two cities. A carrier-route-month with fewer than 5 flights is not kept.
- How it is shown. The on-time share is turned into a 0 to 10 score (share times 10) and shown as a plain phrase: "Usually on time" at 80% or better, "Sometimes late" from 60% to under 80%, "Often late" under 60%, with "US data" for this source. The assistant in the chat can also state the percentage and the number of flights behind it. The same score also feeds the travel-risk band, where that is shown, and is one input to how results are ranked.
- Which signal wins. A flight's live status, when live flight status is available for a near-term flight, comes first. It can show "Cancelled today" or "Delayed today", and it can replace the historical figure: for a flight already under way or landed, the card may then read "Not measured for this flight" rather than the historical phrase. Next is BookWithAI's own record of past on-time results for that flight number, used once enough days have been recorded for it. After that comes the BTS route figure above.
- When there is nothing to measure. The flight shows "Not measured for this flight". BookWithAI does not fill the gap with a neutral-looking number.
Honest limits
- Coverage is strongest on US domestic routes. BTS data covers US airline reporting. International routes, foreign carriers and some smaller carriers are covered thinly or not at all. On-time data is limited outside the US.
- The data is old by design. BTS publishes with a lag of about two months, and the month used can be older still, so the figure describes a past month. Weather, schedules and airline operations change.
- Small samples are noisy. One month on a route with few flights can swing a lot. The number of flights behind a figure is part of what it means.
- It is history, not a forecast. A route's past on-time share does not tell you what will happen on your departure.
- Google's typical range is Google's. BookWithAI neither verifies it nor builds a verdict from it.
- The assistant can be wrong. BookWithAI cannot guarantee the lowest fare. Fares change between searches and between a search and the seller, so confirm the price, baggage and conditions with the seller before you pay.
Related reading
See how BookWithAI works, stage by stage, or the guides on how to judge whether a flight will actually be on time and how flight price tracking actually works.
Stop re-running the same search. Save the fare.
Free to try as a guest; after a few messages it asks you to sign in (also free), and guests have a daily limit. Sign in to save a trip or track a fare.