Ask any AI for ideas and you get the same predictable list everyone else gets. It is not your prompting. It is how models work, and there is a system that beats it. This guide gives you the anti-slop method step by step, plus one copy-paste prompt that runs the whole thing for you.
You open Claude, ask for ten content ideas, and get back a list you could have written in your sleep. A talking-head testimonial. A before and after. Three things you wish you knew. Something that opens with "POV:". None of it is wrong, exactly. It is just the same output every other founder in your space is getting from the same tools on the same day.
Here is the part almost nobody explains: examples in a prompt become the ceiling, not the floor. The moment you feed a model a list of formats to consider, every idea it hands back clusters around that list. Most people accidentally give the AI its own slop as a starting point, then wonder why the results feel generic. The fix is not a better example. It is refusing to give one.
This guide is the anti-slop system for ideas. You generate with zero examples, make the model attack its own output against a slop floor, force it to reframe anything that fails, and keep only the angles that clear the bar on three dimensions. At the bottom you get a single prompt that runs all of it in one paste, so you never have to think about the mechanics again.
Do not hand the AI a menu of formats. The list you give it is the box it will think inside, so keep the box empty. Ask for specific, non-obvious angles on your topic and give it only three things: the topic, who the ideas are for, and the instruction to be specific. Nothing else. No "consider a listicle or a story time," no genre suggestions, no reference posts to match.
Generate 10 candidate angles on this topic: [your topic]. These are for [your audience or voice]. Do not give me formats or genres. Do not suggest a type of post. Each angle should be specific and non-obvious, the kind of take that only makes sense coming from someone who actually knows this subject. Return them as a plain numbered list.
Now turn the model on its own output. Give it the slop floor, a checklist of tired patterns, and have it tag every candidate that matches. This is where the generic ideas get named out loud instead of quietly making it into your content calendar.
Anything that trips the checklist gets tagged slop and sent back. Nothing tagged slop is allowed to survive as written.
For every idea tagged slop, make the model run it through five transforms. These are operations, not templates to copy. The goal is to break the pattern and produce real raw material, not to fill in a nicer-looking blank.
For each angle you tagged slop, generate 3 reframes using these transforms. Apply the operation, do not copy an example.
1. Inversion: flip the assumption. If the idea says "here is what to do," argue why the thing everyone is told is wrong, and why the wrong way actually works.
2. Contrarian: take the unpopular position. If everyone says this is always good, ask when it is actually bad and who it fails.
3. Metaphor: connect the topic to an unrelated domain, such as architecture, cooking, weather, or sports, and let the analogy carry the idea.
4. Specificity drilling: instead of generalizing, drill to one ultra-specific case the audience will instantly recognize as themselves.
5. Direct eye-contact: write as if speaking to one specific person you have in mind, not broadcasting to everyone.
Every surviving candidate and reframe gets scored on three dimensions from 1 to 10. Slop resistance: how unlike generic AI output is it. Fit: how well does it match the audience and voice you gave it. Specificity: is there a real, particular observation in here, or is it vague. An angle only makes the cut if all three land at 6 or higher. A 9 on specificity that scores a 4 on fit is out, because a brilliant idea for the wrong person is still the wrong idea.
One more rule, and it is the one people skip: name the emotion each surviving angle is built to evoke. Pick one of recognition, curiosity, soft validation, mild surprise, or shared complaint. If you cannot name the feeling, the angle is not ready. Keep the top three by their weakest score, not their loudest one.
The method hands you the angle, not the finished post. That is by design. Once you have your top three, you pick one and build the real thing: the script, the hook, the caption, the carousel. What you are protecting is the idea underneath, so that everything you make on top of it starts from something that was never generic to begin with.
Once you understand the four moves, you never have to run them by hand again. Paste this in, fill the two brackets, and the model does the generate, attack, reframe, and score loop in a single pass, then hands you the top three angles with the emotion each one is built to evoke.
You are an anti-slop idea engine. I am going to give you a topic. Refuse to hand me generic AI ideas. Work in four steps and show your work.
TOPIC: [your topic]
CONTEXT: [your audience, voice, or constraints]
Step 1. Generate 10 candidate angles. Use only the topic and context above plus the instruction to be specific and non-obvious. Do not give yourself any formats, genres, or example posts to work from.
Step 2. Attack each candidate against this slop floor and tag any match as slop: talking-head testimonial, before and after, "3 things I wish I knew" listicle, opens with "POV:" or "Imagine if" or "Did you know", could come from any brand in this space, leans on one overused trope, pitches the product before giving value, or claims a result with no specificity.
Step 3. For every slop-tagged candidate, generate 3 reframes using these transforms as operations, not examples: inversion, contrarian, metaphor or cross-domain, specificity drilling, and direct eye-contact.
Step 4. Score every surviving candidate and reframe from 1 to 10 on slop resistance, fit, and specificity. Only keep angles where all three are 6 or higher. For each one, name the dominant emotion it evokes: recognition, curiosity, soft validation, mild surprise, or shared complaint. If you cannot name the emotion, drop the angle.
Return the top 3 angles ranked by their lowest of the three scores. For each: the pitch in one or two sentences, one line on why it is not generic, the emotion, and the three scores. Then list the rejected candidates with the reason each was tagged slop.
Generate with nothing, make the model turn on its own output, force the reframes, and keep only what clears all three bars. That is the entire system. The next time your AI hands you a list you have seen a hundred times, you will know exactly why, and you will have the prompt that refuses to accept it.
Once your ideas stop being generic, the next move is building a real content and offer system on top of them. That is exactly what we build inside Her AI Systems™.
Because examples in a prompt become the ceiling, not the floor. The moment you feed a model a list of formats to consider, every idea it returns clusters around that list. Most people accidentally hand the AI its own slop and then wonder why the output is generic.
Any capable model works. The method is about how you prompt, not which tool you use. Claude, ChatGPT, or Gemini will all run the generate, self-attack, reframe, and score loop when you paste the prompt in.
The slop floor is a checklist of tired patterns that mark an idea as generic: testimonials, before and afters, listicles, openers like "POV:" or "Imagine if", angles any brand in your space could post, pitching the product before giving value, and vague claims with no specificity. If an idea matches, it gets tagged slop and sent back for a reframe.
No. The single prompt near the bottom of this guide runs the whole method in one paste: it generates with full latitude, attacks its own output, forces reframes, scores what survives, and hands you the top three angles with the emotion each one is built to evoke.
Free Download
Keep the anti-slop prompt handy for the next time your AI gets lazy.
Follow along for the next guide
New free guide every Tuesday. Find them first on Instagram, TikTok, or by subscribing to the email list.
Want more?
One new free guide. Every Tuesday. No spam, no fluff, just the systems and prompts I'm building for clients, simplified for you.
Subscribe