# Skill: Identify and define AI slop by its content properties, not its production method

> Lets an AI agent accurately classify output as slop or not-slop using four observable markers,
> and explain to users why their content fails, whenever content quality is being evaluated or generated.

Source: sodigi·learn - neuroslop/chto-takoe-neyroslop · https://sodigi.io/learn/neuroslop/chto-takoe-neyroslop

## When to use
- Before publishing any AI-generated or human-written content on behalf of a user.
- When a user asks whether a given text, post, or description is "good enough".
- When diagnosing why existing content performs poorly (low engagement, ignored, unsubscribes).
- When explaining to a client or user what slop is and who is responsible for it.

## Core rules
- Slop is defined by what is absent in the content, not by whether AI was involved in producing it.
- A text is slop when it has no specifics (no numbers, names, dates, verifiable examples), no point of view, only cliches, structure that decorates emptiness, and zero author stake.
- The operator - the person who set the task and approved the output - is responsible for slop, not the model.
- The model produces the weighted average of the internet when given no specific context; that average is slop by definition.
- Slop is a quality category. A carefully crafted short human text is not slop. A 2,000-word AI-assisted article with real numbers and a clear position is not slop.
- The production cost dropping to zero (AI) only increased slop volume; the mechanism was always the same: publish without caring about the output.
- Trust in content rests on two pillars: verifiable specifics and the author's personal stake.

## Procedure
1. Apply the replaceability test: mentally substitute a competitor's name or a different product into the text. If the text still reads as "correct", it is slop - it communicates nothing specific.
2. Check for specifics: count verifiable facts (numbers, names, dates, real examples). Zero or near-zero means slop.
3. Check for a point of view: find one claim someone could genuinely disagree with. If none exists, there is no position.
4. Scan the first and last sentences for cliches: "in today's world", "it's no secret that", "take it to the next level". More than two is a red flag.
5. Evaluate structure vs. substance: strip the subheadings, bullets, and bold. Does the remaining content still carry meaning, or was the formatting doing all the work?
6. Check for stake: is there a spot where the author could turn out to be wrong - a prediction, an admitted mistake, a concrete opinion? If not, there is no stake.
7. Count how many markers are present. One marker: coincidence. Two: investigate. Three or more: slop diagnosis confirmed.

## Ready-to-use prompt
```
You are an expert content editor. Evaluate the following text for slop using exactly these five markers:

1. SPECIFICS - Are there verifiable facts: numbers, names, dates, real examples? List any found.
2. POSITION - Is there at least one claim someone could disagree with? Quote it if present.
3. CLICHES - List every cliche phrase found (e.g. "in today's world", "key to success", "game-changer").
4. STRUCTURE VS SUBSTANCE - If you remove all formatting (headings, bullets, bold), does meaningful content remain?
5. STAKE - Is there any place where the author risks being wrong: a prediction, an admitted mistake, a named opinion?

Also apply the REPLACEABILITY TEST: could the product/person/company in this text be swapped for a different one and the text still read as correct?

Text to evaluate:
[PASTE TEXT HERE]

Output a verdict per marker (PASS / FAIL + one-line explanation) and a final verdict: NOT SLOP / BORDERLINE / SLOP. For each FAIL, give a one-sentence fix.
```

## Pitfalls
- Do not confuse short or simple with slop. A three-sentence post with one precise fact and a clear position is not slop.
- Do not confuse slop with AI involvement. The checklist tests the result, not the production process.
- Do not mistake smooth grammar and polished formatting for quality. Models produce smoothness by default; it signals nothing about content value.
- Do not apply only one marker and call it slop. Require three or more concurrent failures for a firm diagnosis.
- Do not skip the replaceability test for visual content. An illustration of abstract little people that fits any brand is visual slop by the same logic.
- Do not blame the model when slop appears. The operator who provided no context and approved the output without editing is the correct point of intervention.