Artificial intelligence is beginning to enter the flight-planning workflow that pilots already trust for charts, weather, NOTAMs and route information. The promise is straightforward, it reduces the time spent searching dense aviation material and helps pilots identify the decision that matters before the engine starts.
The risk is just as straightforward. Aviation is full of information that is current only for a few hours, valid only in a particular place, or written with legal precision. An AI answer that sounds clear but omits one temporary restriction, runway closure or weather qualifier can turn convenience into a hazard.
Jeppesen ForeFlight is developing aviation-specific AI capabilities as an enhancement to its existing platform rather than a replacement for it. That distinction matters. Pilots do not need a chatbot to 'fly' the aircraft. They need a better way to interrogate the data they already must review and verify.
The useful cases are obvious. A pilot could ask for a plain-language summary of destination weather, identify NOTAMs that affect a planned arrival, compare alternates, search maintenance records, check fuel implications of a route change, or find the operational consequence of an airport restriction. Instead of reading through pages of material, the pilot may receive a focused briefing: the runway is closed at the expected arrival time, the approach lighting is unavailable, the crosswind forecast exceeds a personal or company limit, the alternate has a fuel constraint.
That can reduce workload. It can also improve the quality of the pre-flight conversation between pilots, dispatchers and maintenance teams. But aviation’s safety system rests on a principle AI must not blur. The official source is the authority, not the summary.
A NOTAM is often difficult to read because it is compressed, technical and written for a specific operational purpose. An AI system may make it easier to understand. It must not decide that a restriction is irrelevant simply because it appears unlikely to affect the planned flight. A temporary restricted area, military activity, navigation-aid outage or airport operating limitation may matter only under a narrow set of conditions. Those are precisely the details a confident summary can miss.
The most important design question is traceability. Every AI-generated recommendation should link directly to the source data, show its issue time and validity period, and make uncertainty visible. A pilot should be able to ask: 'Where did this come from'? and reach the original weather product, NOTAM, chart, airport notice or aircraft record immediately.
The danger is not merely an obvious hallucination. Most pilots will distrust a wildly inaccurate answer. The more dangerous failure is a plausible answer that is 95 percent correct, written with calm confidence, and missing the one fact that changes the decision.
AI also cannot replace operational judgment. It may identify deteriorating weather, but it cannot know a pilot’s recent experience in icing conditions, fatigue level, personal comfort with a short wet runway or the subtle operational culture of a particular flight department. It can calculate fuel. It cannot own the decision to continue, divert or turn back.
That is why training will matter more than the interface. Pilots will need to learn when to use AI for search and synthesis, when to return to raw source material, how to recognise uncertain output, and how to resist automation bias. The right mindset is not 'the AI cleared the flight'. It is 'the AI helped me ask a better question'.
Aviation has always adopted automation cautiously because the technology changes the work before it changes the aircraft. AI flight planning should follow the same path. It can make preparation faster, clearer and more disciplined. But the pilot must remain the final filter not because AI is useless, but because the sky does not accept a confident answer as proof.