AI Contract Negotiations: Why the "Input Data" Definition Decides Who Owns What

In this AI contracting lesson from How to Contract, we focus on how input data" definition controls who owns what in your AI deal. The input definition clause sets who owns the data, what the provider may reuse, what stays confidential, and whether the provider can learn from customer inputs. David Sclar explains why customers push for a broad definition. Inputs reveal a customer's AI strategy, so the definition should reach past raw content to cover prompts, instructions, context files, and fine-tuning data. His punchline: a "don't train on my data" clause loses most of its value if "customer data" is defined narrowly. Laura Belmont lays out the core tension. Providers need room to study real-world usage so they can improve accuracy, fix bugs, monitor safety, and prevent abuse. Customers want their proprietary information and competitive intelligence protected. The panel then reviews a vague AI-drafted sample, "any data or information submitted by the customer while using the product," and shows where it fails. They cover the inclusions and exclusions to add, the data that connectors and agents pull in without anyone "submitting" it, and how to handle prompts, feedback, metadata, and usage patterns. New contract lessons publish every week at howtocontract.com. Segments: 00:00 Why input definitions matter 00:38 Customer ownership and confidentiality 02:02 Provider vs. customer tensions 04:12 Common definition mistakes 05:21 Reviewing a vague clause 05:48 Provider edits and exclusions 07:16 How to justify changes 08:25 Customer-side redlines 10:32 Wrap-up: key takeaways