Why does it matter more than it used to?

Because replacing a draft went from taking an afternoon to taking seconds. The version you liked can now be overwritten before you have registered what was good about it, and nothing in your own memory has kept up with the pace.

How does it help with questions about AI use?

It shows the work happening over time, which is checkable by both sides. That is a far better answer to an accusation of machine writing than any detector score, because it is evidence rather than an estimate nobody can contest.

Chat windows, downloaded files and anything edited outside a shared system. Those are exactly where a lot of drafting with these tools happens, so the work most likely to raise a question is often the work with no record behind it.

Glossary

Version history

A record of a document's earlier states, which matters more once new drafts arrive fast enough to bury the one before.

In plain terms

A record of what a document looked like before. It used to be insurance against losing work. Now that a replacement draft takes seconds, it is more often the only trace of how something came to be written the way it was.

Why it matters

Because the question people are asked about a document has changed. It is no longer only what does it say, but increasingly how was this made, and version history is the only thing in most organisations that can answer the second.

How it works

Its original job was recovery: something went wrong and you wanted last week's copy. That use has not gone away and is no longer the main one, because drafts now arrive fast enough that losing a good one is a routine rather than an incident.

The rate of change is what altered its value. When a rewrite took an afternoon, the previous version was still in somebody's mind; when it takes ten seconds, the version you liked can be replaced before you have noticed what was good about it.

It is now the only record of process most work has. Drafts, revisions and the sequence between them show that a person shaped the material, which is the evidence a dispute about authorship needs and which nothing about the finished document supplies.

That makes it the honest answer to an accusation of machine writing. A detector produces a score nobody can argue with; a revision history shows the work happening over time, which is checkable by both sides and does not require anybody to trust a tool.

It also protects against a specific new failure: regeneration replacing something better. The discarded version is frequently stronger in one respect, and without a history the only copy of that paragraph was the one you just overwrote.

Where it exists at all varies more than people assume. A shared document keeps it, a chat window may not, a file downloaded and edited locally has none, and the work most likely to need a record is often the work done in the place least likely to keep one.

Two questions a document gets asked

The practical version of this is narrower than it sounds and worth doing. It is not a records programme: it is knowing which of the places your people draft in keep a history and which do not, and moving anything that might be questioned into the first group before it is questioned rather than afterwards. A piece of writing that began in a chat window, was pasted into a document and then edited has a record of the last part only, and the last part is the part that looks least like human work because the interesting decisions were taken earlier. Establishing that habit costs nothing and pays in the one situation where nothing else will help: somebody asks how a document came to exist, and the honest answer is available, dated and checkable rather than a recollection of a conversation that a tool no longer holds.

Seen in the wild

A regenerated draft replacing a paragraph that was better, with no way back.
ChatGPT
A shared document whose revision record shows a person shaping the work over days.
Notion AI
A generated deck edited heavily, where only the final state was ever saved.
Gamma

Common misconceptions

People assume

It is a backup.

In fact

Recovery was its original purpose and is now the smaller half. What it uniquely provides is a record of how something was made, which is what an authorship question or a process question actually needs.

People assume

Everything we work in keeps one.

In fact

Shared documents usually do, chat windows often do not, and anything downloaded and edited locally has none at all. The gap tends to sit exactly where drafting with these tools happens most.

Questions

Why does it matter more than it used to?: Because replacing a draft went from taking an afternoon to taking seconds. The version you liked can now be overwritten before you have registered what was good about it, and nothing in your own memory has kept up with the pace.
How does it help with questions about AI use?: It shows the work happening over time, which is checkable by both sides. That is a far better answer to an accusation of machine writing than any detector score, because it is evidence rather than an estimate nobody can contest.
Where are the gaps?: Chat windows, downloaded files and anything edited outside a shared system. Those are exactly where a lot of drafting with these tools happens, so the work most likely to raise a question is often the work with no record behind it.