AI Without the Hype.
Chapter 16 of 17
Part Six · Keep It Honest · Chapter 16

Making sure
it is not slop

By the end of this chapter you can catch and fix the generic, templated quality that is the single most common complaint about AI output, in both writing and design, using a set of passes we built and give away.
AVATAR OPENER · ~90s
Watch: the same output, before and after the anti-slop passes
HeyGen avatar · generated, consistent presenter

Ask people what is wrong with AI output and you get one answer more than any other: it is generic. Bland writing that could have been about anyone. Designs that look like every other template. The word for it is slop, and producing it at scale is the default failure mode of every tool in this course. This chapter is the fix, and it is a discernment skill, the ability to look at output and see that it is average, before you ship it as if it were yours.

Here is why slop happens, because understanding it is half the cure. A model, asked plainly, gives the average answer to the average question, because the average is what it was trained to produce. That is not a flaw you prompt away with one clever instruction. It is the gravity of the thing, and the only way to beat gravity is a deliberate pass afterward that drags the output off the average and toward something that is actually yours.

Generic is the gravity of AI output. Escaping it is a pass you run, not a prompt you hope with. Slop is not the model being bad. It is the model being average, which is its default. Beating it takes a pass on purpose, not a better single prompt.
TEXT SLOP AND HOW TO STRIP IT

Written slop has a fingerprint, and once you can see it you cannot unsee it. The giveaway phrases, the tidy lists of three, the "it is not just X, it is Y" shape, the breathless adjectives that describe nothing, and, the tell this whole brand refuses, the em-dash sprinkled everywhere as a rhythm crutch. None of it is wrong exactly. It is just what the average sounds like, which means it sounds like no one.

The fix is a set of editing passes, each hunting one kind of slop, run after the draft exists. We have built these and give them away: a humanizer that strips the AI tells and generic phrasing, a copy-editing pass of sequential sweeps that each fix one thing, and the voice audit that flags any sentence that could have been written about anyone. You do not run them all every time, but knowing they exist and running the right one is what separates shipped-average from shipped-yours.

Strip the AI tells
The humanizer pass hunts the fingerprint: generic phrasing, the tidy triples, the "not just X but Y" shape, the em-dashes. Cut them out.
Sweep one thing at a time
The copy-editing pass makes several separate sweeps, each fixing a single kind of problem, because one pass trying to fix everything fixes nothing well.
Flag the anonymous
The voice audit marks any sentence that could have been written about anyone. If a stranger could not tell it was you, it is slop, however polished.
Put back what is yours
Stripping is only half. Then you add the concrete detail, the real example, the odd true phrasing only you would use. That is what makes it yours.

Notice the last card, because it is the half people forget. Removing the slop leaves you with clean, correct, still-generic text. The output is not yours until you put yourself back in: the specific detail only you know, the real example, the slightly odd phrasing a committee would have smoothed out. Stripping the tells and adding the truth are two moves, and you need both.

The anti-slop pass on text
Here is a piece of AI-drafted text I do not want to ship as generic slop: [paste the draft] Run it in two passes. Pass one, strip the slop: cut the generic phrasing, the tidy lists of three, the "it is not just X, it is Y" shape, the empty adjectives, and every em-dash. Flag any sentence that could have been written about anyone, not specifically me or this thing. Pass two, put me back in: show me where a concrete detail, a real example, or a blunter, truer line would replace something average. Ask me for the specifics you need rather than inventing them. Do not make it longer or smoother. Make it more specific and more mine.
Try it in Claude
DESIGN SLOP AND HOW TO BEAT IT

Design slop is the same disease in a different medium. The default AI layout is competent and utterly forgettable: the same centred hero, the same three feature cards, the same rounded everything, the look that says a machine made safe choices. It is not broken. It is just the visual average, and the average is invisible.

The cure is the same shape as for text, and we build design skills that give it away too: front-end design guidance, web-design guidelines, and a library of real, distinctive design systems to learn from rather than defaulting past. The single most useful instruction inside them is this: take one real aesthetic risk you can justify. Not ten, not zero. One considered choice that a template would never have made, that you can explain, that makes the thing look like a decision instead of a default.

VS

The discipline is precisely one risk, justified. Ten risks is chaos, another kind of slop. Zero risks is the forgettable default. One choice you can defend, with everything else kept clean and calm around it, is how a design stops looking machine-made. And it ties back to the beginning of the course: you bring the taste and the point of view, the tool brings the execution. Design slop is what you get when you let the tool bring the taste too.

THE HABIT: BEFORE AND AFTER

The way to build this into how you work is to always look at the before and after. Run the pass, and hold the average version next to the specific one. Seeing the two side by side trains your discernment faster than any rule, because you feel the difference between forgettable and yours, and after a while you start catching the slop in the draft before you even run the pass.

Before and after, in one line each
Before: "Our platform delivers powerful, seamless solutions that help you unlock your full potential and achieve more." After: "We fix film cameras. You post them a broken one, we send it back working, and we tell you honestly if yours is not worth the repair."

That is the whole chapter in two sentences. The first could be any company selling anything, which means it sells nothing. The second could only be one shop, which is exactly why it lands. The passes exist to get you from the first to the second, reliably, on purpose, every time you are about to ship something a machine drafted.

NOW YOU TRY · EVALUATE
Run one real output through the passes

Take one real thing you drafted or designed with AI recently, a piece of copy, a page, a layout. Run the matching anti-slop pass: for text, strip the tells then put yourself back in; for design, name the one justified risk you would take and what you would keep clean around it. Keep the before and after side by side. The win is a specific, yours version next to the generic one, and the trained eye that comes from seeing the gap.

Right if you have a before and after where the after could only be about you or your thing, and you can point to the specific slop you cut and the specific truth you added.
Show the worked solution
The drill works when the after is unmistakably yours and you can say exactly why. Say the draft is an about-page paragraph for your business, and it reads: "We are a passionate team dedicated to delivering exceptional results and building lasting relationships with our clients through innovative solutions." Every word is fine and the whole thing is slop, because it could be any company on earth. Run the first pass and you cut it all: "passionate," "dedicated," "exceptional," "innovative solutions," the empty relationship talk, and the rhythm that exists only to sound professional. Now you have almost nothing left, which is correct, because there was almost nothing there. Then the second pass, the one that matters: you put yourself back in with what is actually true. "There are two of us. We have done this for eleven years. We turn work around in a week, we tell you when a job is not worth doing, and we have never once used the word synergy." Suddenly a stranger can picture the actual business, because you replaced the average with the specific. Hold the two versions side by side and the lesson lands in your body, not just your head: the polished generic one impressed no one, the plain specific one is the one a real customer trusts. Do this a dozen times and you stop needing the pass, because you start hearing the slop as you draft it, which is the discernment the whole chapter was building.
WATCH FOR
You try to prompt away slop in one shot. Generic is the model’s default gravity, not a one-prompt bug. Draft, then run a deliberate stripping pass afterward.
You strip the slop and ship the clean-but-generic result. Removing tells is only half. Put yourself back in with concrete detail and real examples, or it is still average, just tidier.
You let the tool make the taste decisions. You bring the point of view, it brings execution. Design slop is what happens when you outsource the judgment along with the work.
You take ten aesthetic risks, or none. One justified risk with everything else kept clean is the move. Zero is forgettable; ten is a different kind of slop.
WHAT YOU LEARNED
The takeaways
  • Slop, the generic templated quality, is the default output of every AI tool, because the model produces the average and the average sounds like no one.
  • You beat it with a deliberate pass after drafting, not a cleverer single prompt. We build and give away the passes: a humanizer, a copy-editing sweep set, and a voice audit.
  • Text has two moves: strip the tells (generic phrasing, tidy triples, empty adjectives, em-dashes), then put yourself back in with concrete, true, specific detail.
  • Design slop is the same disease: beat it by taking exactly one justified aesthetic risk and keeping everything else clean, so it looks like a decision, not a default.
  • Train the eye by comparing before and after every time, until you catch the slop in the draft and stop shipping the average as if it were yours.
Your project · strip the slop

Run one real AI output through the matching anti-slop pass this week, and keep the before and after side by side. You now have the discernment that separates shipped-average from shipped-yours. One chapter left, the honest one: where all of this breaks at scale, and what stays yours no matter how much power you have.

The tool will always hand you the average first, because the average is what it knows. Slop is not it failing, it is it defaulting. The work, the only work it cannot do, is dragging the output off the average and putting the specific, true, unmistakably-you thing back in. That refusal to ship the generic is taste, and taste is the part that stays yours.