Generic in, generic out. When you give AI a general prompt you get the average of all averages. The fix is constraints, and then skills.
Everyone is complaining about AI slop. Generic lesson plans. Bland worksheets. Apps that look like every other app. The blame usually lands on the tool.
It is the wrong target.
The average of all averages
When you give AI a general prompt, you get the average of all averages. The model has read an enormous amount of text, and without direction it gives you the safest, most statistically likely answer. Nothing wrong with it. Nothing right with it either. It is the beige paint of content.
Slop is not what AI makes. It is what AI makes when you give it nothing to work with.
Ask for a lesson on photosynthesis for 7th graders and you get the lesson every other person got. Same hook. Same vocabulary list. Same exit ticket. That is not the AI failing. That is the AI doing exactly what you asked, with no information to do anything else.
The same goes for what people build. Tell a tool to make a quiz app and you get the average quiz app. The slop is in the specification, not the software.
Before
Write me a lesson on photosynthesis for 7th graders.
After
Write a 45 minute lesson for 7th graders who can already label a leaf but cannot explain where plant mass comes from. Outcome: students construct an evidence-based explanation. Open with a surprising data set, not a definition. No vocabulary list. No exit ticket that asks for recall. Use plain language and my phrasing for directions.
Constraints are the fix
Slop shrinks when you shrink the space the model can wander in. That means building constraints into every prompt.
Build these into your prompt
- Who the learner is and what they already know
- The exact standard or outcome
- The format, length, and structure you want
- What to avoid, including the cliches you are tired of
- Your voice, your pedagogy, your non-negotiables
Then go one step further: skills
A prompt is a one-time instruction. A skill is a standing set of rules the AI loads every time it does a certain kind of work. It carries your criteria, your examples, your quality bar. Instead of re-explaining what good looks like in every chat, you build it once and the model stays inside the lines.
The takeaway
If your output feels like slop, do not ask whether the AI is bad. Ask what you told it. Vague in, average out. Specific in, useful out.
AI does not remove your expertise from the process. It runs on it. The more of your judgment you build into the constraints, the less slop you get.