Source material
- Industry term
Source material is the set of records used as input or evidence for something made from them, such as a summary, an analysis, a decision, or an AI's answer.
An AI building an answer from its sources treats them as true, so careful writing and faithful copying carry their mistakes, gaps, and age into the result unchanged.
Where you’ll meet it
Your assistant drafted a one-page health summary for your dad's first visit with a new doctor. It reads well, but one file it used was a medication list from before his hospital stay last spring, when two of his prescriptions changed.
What to do
Give the assistant the current medication list, have it redo that section, and check the doses against his pill bottles. Before the visit, also ask it to list every other file it used with its date, and replace any older than the hospital stay.
Easy to mix up
| Source material | Source trail |
|---|---|
| The records themselves, such as files and notes | The record of which claim came from which source |
Check yourself
Your assistant drafts a quarterly sales report from the team's shared spreadsheet. A second AI tool, handed the same spreadsheet, checks the report and confirms every figure. What has that check shown?
Go deeperEach copy can lose detail, and originals can't always be kept
In research and records work, a source is primary, such as a recording, or secondary, such as someone's notes on it. AI workflows add steps in between, such as a transcript of the recording or a summary of the transcript. Each step can drop or change detail, so a claim that carries weight is checked against the earliest version available.
Keeping every source forever helps with audits and corrections, but it can run against privacy promises, what people consented to, or legal limits on how long data is kept. A common compromise keeps the exact passages a claim relied on, or a hash, a short fingerprint that shows whether a file has changed, in place of the whole file.