AI in Aerospace: Documentation and Fault Diagnosis

Vendor pages document file handling, research access, limits, and accuracy cautions, but provide no aerospace-specific validation for fault diagnosis.

Sources checked 2 Oct 2026

For AI in aerospace documentation and fault diagnosis, OpenAI’s file-upload FAQ, Anthropic’s file-upload guide, and Google’s file-upload guide establish what each product accepts and how it handles sources; they do not establish aerospace accuracy. None says that ChatGPT, Claude, or Gemini is approved for aircraft maintenance or validated against controlled aerospace documents, and Chat Picker has not tested the assistants for this use. Treat outputs as drafts for qualified review against approved source material.

This rewrite removes the previous page’s unsourced statistics and test results.

What the vendors document

ChatGPT. OpenAI’s file uploads FAQ, as read on October 1, 2026, says uploads are available on Free and paid plans, subject to plan limits and account settings. It sets a 512 MB hard limit per file and limits Free users to 3 uploads per day. OpenAI’s Plus page, as read on October 1, 2026, lists Plus at $20 per month, billed monthly. OpenAI says Deep Research can work with uploaded files and the public web and produces reports with citations or source links. Its accuracy note says ChatGPT can produce incorrect or misleading output and may sound confident while wrong, so important information needs verification against reliable sources.
Claude. Anthropic’s upload guide, as read on October 1, 2026, lists PDF, common document and image formats, a 500 MB limit for each chat upload, and a 30 MB limit for each project file. For PDFs of 100 pages or fewer, Claude analyzes text and visual elements; from 101 to 1,000 pages, it processes text only. It cannot interpret embedded images in non-PDF documents. Anthropic’s pricing page, as read on October 1, 2026, lists Pro at $20 per month with monthly billing or $17 per month with annual billing. Research is available on paid plans and uses multiple searches. Anthropic’s accuracy note says not to use Claude as the only source of truth and to inspect cited sources. Its Usage Policy also requires qualified professional review for covered advice or decisions that directly affect people.
Gemini. Google’s upload guide, as read on October 1, 2026, says users must sign in, a prompt can contain up to 10 supported files subject to availability, and each non-video file can be up to 100 MB. Google AI plans, as read on October 1, 2026, list Google AI Pro at $19.99 per month, with 4 times the Free usage limits and 5 TB of storage. Gemini Deep Research can use uploaded files and, when connected, sources such as Gmail and Drive, with Google Search included by default; Google AI Ultra reports may include visuals such as schematics. Google says sources may appear within a Gemini response. Its safety guidelines say Gemini should not produce factually inaccurate output that could cause significant harm and that context, including scientific applications, matters.

Training controls also differ. OpenAI says an opt-out is available on Free, Go, Plus, and Pro; Anthropic says Free, Pro, and Max have an opt-out and Team is not trained on by default. The Gemini plans page does not state a training setting, so that point remains unverified. These pages were read on October 1, 2026.

What the documentation cannot tell you

The vendor pages do not publish aerospace-specific correctness, repeatability, latency, throughput, or maintenance cost-and-benefit results. They also do not show whether connector labels, bulletin revisions, fault-code qualifiers, diagnostic branches, or stop conditions survive document processing intact.

Only your own controlled trial can show how an assistant handles your actual document formats, scan quality, terminology, source access, and review workflow. A displayed citation does not prove that the assistant used all relevant context or interpreted it correctly.

How to check it yourself

Parse a wiring schematic. Give each assistant the same authorized schematic page. Use this prompt: “Using only the attached wiring schematic, list every connector, pin, wire callout, and page-level qualification you can read. Quote each label exactly and mark anything illegible; do not infer a missing label.” Look for unsupported additions and missing qualifications. Record each mismatch and every answer element without a source location.
Interpret an engine bulletin. Upload an authorized engine vibration bulletin. Ask: “Using only the attached bulletin, extract every fault-code condition, likely cause, confirmation step, and task limitation. Quote the bulletin label and page for each item, and write ‘not stated’ when it supplies no next step.” Record invented steps, changed conditions, and code labels that do not match the document.
Build a hydraulic diagnostic tree. Supply a hydraulic diagram and matching bulletin excerpt. Ask: “Build a diagnostic tree from the attached files. For each branch, state the observation, confirmation step, next branch, and stop condition, with a page reference; do not add a step that the files do not support.” Record missing branches, unsupported causes, and advice that bypasses a required confirmation.
Check repeated-query stability. Use the same files and exact prompt in fresh chats with each assistant. Repeat with web search off and then on when available. Look for changes in fault labels, sequence, caveats, and citations. Record every material difference rather than keeping only the answer that appears best.
Measure latency and throughput. Use an identical, fixed upload set and prompt under the plan and network conditions you expect to use. Record upload completion, time to the first usable output, total elapsed time, interruptions, and usage-limit messages. Keep service failures separate from generation time.
Build an MRO cost-benefit worksheet. Give the assistant your actual file inventory, current plan price, usage-counter readings, review time, and rework log. Ask: “Build an MRO cost-benefit worksheet that separates subscription, upload, human review, and rework. Label unverified plan limits or costs as unknown, and do not estimate them.” Record missing cost inputs and any unsupported assumption.

Which rows of the comparison matter

Open the ChatGPT, Claude, and Gemini comparison matrix. Start with the Free plan, Low-cost tier, Main paid plan, Heavy-use plans, and Team plan rows. Then check Context window in the app, How usage limits are described, Training on your chats, and Ads.

Add each vendor’s file formats and upload limits from its linked documentation. A large context figure does not establish that a diagram or bulletin was read correctly. Chat Picker leaves unconfirmed figures blank and marked Not verified; prices are in US dollars unless the cell says otherwise, so confirm current pricing before subscribing.