Using AI for content without producing slop

“Slop” is the word the industry settled on for AI content that is grammatically fine, superficially relevant and completely worthless — because it contains nothing the reader could not have got from the same prompt. Google’s spam policies now explicitly target it at scale. Here is the process we teach for using AI heavily without producing it.

The rule

AI drafts. A person adds what only a person knows. A person decides whether it publishes.

If any of those three is missing, you are producing slop.

The process

1. Brief before prompt

Most bad AI content comes from a bad brief, not a bad model. Before you prompt, write down:

• Who is reading this, and what do they already know?
• What question are they trying to answer?
• What do we know that a generic answer would not include?
• What should they do after reading?

If you cannot answer the third one, do not write the page. That is the test for whether it deserves to exist.

2. Draft with the model — for structure, not facts

Use AI to get a first pass at structure, headings and plain explanation. Do not trust a single fact, figure, date or name it produces. Treat every specific claim as a placeholder to be verified or replaced.

3. Verify every claim you keep

Prices, dates, statistics, quotes, product names. Check each one against a source you can cite. If you cannot verify it, cut it. A confident wrong number is worse than no number.

4. Add what only you know

This is the step that separates a useful page from slop, and it cannot be delegated to the model:

• A real example from your own work, dated
• A screenshot you took
• A number you measured
• An opinion you would defend in the room
• The mistake you made and what you learned

One paragraph of this is worth more than a thousand words of fluent generalities.

5. Edit like a reader

Read it as the person from your brief. Cut every sentence that does not move them towards their answer. AI drafts are reliably 30–40% too long.

6. Sign it and date it

A named author and a “last checked” date. Both are trust signals to readers and to search systems, and both make you accountable for the page — which is the point.

The failure modes

What it looks like	Why it happens	Fix
Fluent, generic, says nothing	No brief; nothing added	Step 1 and step 4
Confident wrong facts	Model output trusted	Step 3
Fifty near-identical pages	Volume mistaken for value	Publish one good page
“In today’s fast-paced digital landscape…”	Unedited model prose	Step 5, ruthlessly
No author, no date	Nobody accountable	Step 6

What this means for a course

We do not teach “prompting”. Anyone can learn that in an afternoon. We teach the brief, the verification, the editing and the judgement — because those are what an employer pays for, and what the model cannot do. That is the AI Content Creation course.