Provenance in the Age of AI Slop

The internet has never lacked content.

Artificial intelligence has simply changed the economics of producing it.

A competent article, image, summary or opinion can now be generated in seconds. The result may be clear, well structured and entirely plausible. It may also have nobody behind it: no experience, no commitment and no responsibility for whether it is actually true.

This is often called AI slop — content with the shape of expertise and none of the substance.

The term is dismissive, but it points at a real problem. When production becomes almost free, volume stops being evidence of effort or knowledge.

The question changes from “Does this look professional?” to “Where did this come from?”

That is a question of provenance.

Provenance does not prove correctness. A carefully documented idea can still be wrong. But it establishes a chain of responsibility.

Who published it?

When?

How has it changed?

What other work does it connect to?

Is the author willing to stand behind it, revise it and explain their reasoning?

A personal website can offer far more of this than a social feed ever will.

A long-owned domain creates continuity. A public repository provides dated commits. Revision history shows that ideas developed over time rather than materialising from nowhere. Related articles reveal consistency, contradiction and growth.

This is one reason I would rather write the site in source control than inside a conventional content-management database.

The repository becomes part of the publication.

AI assistance does not, in itself, destroy provenance. The important line is not between text typed by hand and text produced with tools.

Writers have always used editors, research assistants, spell-checkers, templates and long arguments over coffee. The relevant questions are whether the named author supplied the ideas, exercised judgement and accepts responsibility for the result.

An AI may help organise a draft, challenge an argument or rescue a limp sentence. It should not become a laundering machine through which nobody is accountable.

Transparency helps, but it has to mean something. A generic badge saying “AI assisted” tells us roughly as much as “contains ingredients”.

More useful signals include:

• the author’s stated editorial principles;
• visible revision history;
• links to source material where factual claims matter;
• retrospective corrections;
• consistency with the author’s wider body of work;
• a willingness to distinguish experience from speculation.

Provenance also offers a defence against impersonation. If someone can generate an article in my apparent style, having an established publication channel makes it easier to show what I actually released — and what I did not.

The same principle applies to organisations. Official repositories, signed releases, documented decision records and traceable data pipelines all become more valuable the moment synthetic output is abundant.

We may need to stop treating a polished surface as a proxy for credibility.

The future internet will contain an extraordinary amount of material that looks as though someone cared.

Provenance helps us find the places where someone actually did.

• #ai
• #writing
• #provenance

@online{byatt2026-provenanceintheageofaislop,
  author  = {Stephen Byatt},
  title   = {Provenance in the Age of AI Slop},
  year    = {2026},
  date    = {2026-07-16},
  url     = {https://learningoutloud.stephenbyatt.com/blog/provenance-in-the-age-of-ai-slop/},
  urldate = {2026-07-16},
  note    = {Learning Out Loud}
}