Modern Creator
Ben AI · YouTube

How to De-Slop Every AI Output Forever (With 1 Skill)

An 11-minute walkthrough of a Claude skill that enforces your quality bar on every AI output before it leaves the building.

Posted
today
Duration
Format
Tutorial
educational
Views
439
31 likes
Big Idea

The argument in one line.

A shared output quality gate -- not better prompts -- is what converts individual AI productivity into institutional value, because AI reinforces the user not the company standard.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You manage a team where multiple people use AI to produce public-facing content and you have caught embarrassing outputs slipping through.
  • You already use Claude skills or a shared second brain and want a last-mile QA step before anything goes out.
  • You are building or consulting on an AI operating system for a business and need the output-checking layer.
  • You notice your own quality bar drops when you are tired, rushed, or frustrated with AI.
SKIP IF…
  • You are a solo creator who reviews every AI output yourself before publishing -- the personal consistency problem is already handled.
  • Your AI output volume is low enough that a careful read catches issues every time.
TL;DR

The full version, fast.

AI slop is not caused by bad prompts -- it is caused by inconsistent quality bars that shift by person, mood, and role. The De-Slop skill addresses this by running every AI output through two layers: a universal check that flags objective red flags (AI vocabulary, em-dashes, unverified claims, readability problems) and a company-specific check personalized to your brand voice, visual standards, and factual baseline. Three sub-agents grade in parallel to reduce confirmation bias, then produce a verdict of good to go, fix some things, or not ready, along with a source-backed table of specific issues. Every run appends to a shared log so teams can track recurring failure patterns across the organization.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →
Chapters

Where the time goes.

00:0000:45

01 · Intro

Hook on AI frustration and the cringe of AI-generated LinkedIn posts. Previews the de-slop skill.

00:4501:52

02 · Why I Built It

Personal pattern: more AI output, more embarrassing work. Companies now banning AI for public-facing content.

01:5203:07

03 · The Problem It Solves for You

Slop is subjective. Your quality bar shifts with fatigue and time pressure. The skill enforces your stated standard even when you cannot be bothered.

03:0705:07

04 · The Problem It Solves for Businesses

Every employee has a different quality bar. Individual AI reinforces them, not the company. References a16z on institutional vs individual AI.

05:0707:30

05 · How the Skill Works

Universal Slop Check + Company Slop Check. Three sub-agents grade simultaneously. Three verdicts: good to go / fix some things / not ready.

07:3009:52

06 · Examples + How to Use

Live demos: sales email (good to go with fixes), LinkedIn post (not ready), newsletter, sales proposal. Skill flags -- human decides.

09:5210:33

07 · De-Slop Log

Every run appends to a shared team log showing what got flagged, by whom, and how often. Pattern detection at org level.

10:3311:22

08 · How to Personalize the Skill

Free download with blanks to fill in. De-Slop Builder in AI Accelerator walks through personalization step by step. CTA to community.

Atomic Insights

Lines worth screenshotting.

  • AI reinforces the user's current state, not the company's standard -- which means more AI usage widens quality variance, not narrows it.
  • Giving every individual a 10x productivity boost with AI does not make the company 10x more valuable if there is no shared quality benchmark.
  • Your personal quality bar is not fixed -- it drops when you are tired, rushed, or frustrated, making automated enforcement the only reliable solution.
  • Slop is subjective: what one person calls embarrassing, another calls acceptable -- so the company must define the standard once and enforce it systematically.
  • Universal slop checks catch objective failures; company slop checks catch brand misalignment -- you need both layers because one without the other misses half the problem.
  • Running checks through multiple sub-agents in parallel reduces AI's tendency to agree with the user who generated the content.
  • Logging every De-Slop run creates an institutional memory of quality failures that reveals which patterns recur most, turning individual mistakes into organizational learning.
  • Adding a quality gate as the final step inside an existing skill automates compliance -- the team never has to remember to run it.
  • The De-Slop skill does not auto-fix everything -- it detects and flags so the human decides which suggestions to apply, keeping judgment in the loop.
  • Speed of production without a shared direction or quality benchmark means more chaos, not more value -- the a16z institutional AI thesis in one sentence.
Takeaway

Why your quality bar needs to live outside your head.

WHAT TO LEARN

Personal quality standards are unreliable under pressure -- the only way to hold the line consistently is to encode the standard once and let a system enforce it on every output.

  • AI output quality is not a prompt problem -- it is a standards enforcement problem. Better prompts help, but they do not prevent a tired person from approving a mediocre output at 11pm.
  • Every individual's AI tool reinforces that individual's current judgment, not the organization's established standard. The gap between what an employee thinks is acceptable and what the company requires widens as AI usage scales.
  • The a16z finding that individual 10x productivity has not produced 10x company value is explained by the absence of shared direction and quality benchmarks at the output layer, not the absence of better tools.
  • A quality gate that catches objective failures (AI vocabulary, unverified facts, readability) is different from one that catches brand failures (wrong tone, off-strategy, visual inconsistency) -- you need both layers, and they need different reference inputs.
  • Running quality checks through multiple sub-agents in parallel is a practical workaround for AI's tendency to validate whatever the generating user produced. Structural disagreement produces better verdicts than single-agent review.
  • Logging quality check results at the team level converts individual errors into organizational data -- over time, the log reveals which failure types are systemic versus accidental.
  • Inserting a quality gate as the final automated step inside an existing workflow eliminates reliance on team members remembering to run a separate check. Compliance should be structural, not behavioral.
Glossary

Terms worth knowing.

AI slop
Low-quality, generic, or brand-misaligned content produced with AI assistance, often detectable by characteristic vocabulary, phrasing patterns, and factual vagueness.
De-Slop skill
A Claude-based skill that applies a two-layer quality check to any AI output before it is published, producing a pass/fix/fail verdict with specific improvement suggestions.
Universal Slop Check
The first layer of the De-Slop skill -- checks any content regardless of context for objective red flags like AI writing patterns, em-dashes, unverified factual claims, and readability problems.
Company Slop Check
The second, personalized layer of the De-Slop skill -- checks content against your specific brand voice, visual identity guidelines, company strategy, and factual baseline.
Institutional AI vs Individual AI
A distinction from a16z arguing that individual productivity tools and institutional intelligence are fundamentally different problems -- individual AI makes people faster; institutional AI aligns an organization around shared direction and quality.
AI OS / shared second brain
A centralized knowledge base that all team members' AI tools pull from, ensuring consistent context; distinct from output quality checking, which is what De-Slop adds on top.
Work slop
The organizational problem of AI-generated low-quality work proliferating across a business as AI adoption increases without quality guardrails.
Resources

Things they pointed at.

Quotables

Lines you could clip.

02:55
You define what good looks like once, and the scale holds you to that on every AI output that goes out into the world.
Clean thesis statement, zero setup neededTikTok hook↗ Tweet quote
03:05
Think of it like a spell check, but for slop.
Memorable analogy, universally relatableIG reel cold open↗ Tweet quote
04:09
AI just made every individual 10 x more productive, but no company became 10 x more valuable as a result. Where did the productivity go?
Provocative question that stops the scrollNewsletter pull-quote↗ Tweet quote
03:50
AI just reinforces the user. It doesn't necessarily reinforce the truth or the truth within a company.
Counterintuitive claim, short enough to stand aloneTikTok hook↗ Tweet quote
04:20
Speed of production without a shared direction or quality benchmark means you're not actually adding value to the company but creating more chaos.
Direct consequence framing, quotable for B2B contentLinkedIn pull-quote↗ Tweet quote
The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

analogystory
00:00We all know the frustration of going back and forth with AI only to end up with a completely useless output or the feeling of cringe when we see someone else or even worse, a team member put out a clearly AI generated LinkedIn post, email, comment, or anything else. And, of course, good prompts, skills, and context can improve AI outputs.
00:19But even with that, AI drifts and it never actually guarantees high quality outputs. And across an entire business, most people won't actually apply many of these best practices resulting in lots of AI slop going out into the world.
00:32So in this video, I wanna show you my new favorite skill that these slops every AI output before it goes out the door, show you how it can prevent you and your business from putting out slop, and show you how to adapt it for yourself or your business easily. Now you'll be able to download the scale for free in the first link in the description below, but before showing you the scale with some examples and exactly how it works, let me quickly go over why I actually built this and why I think it's so powerful.
00:56Now, built this because I kept seeing a recurring pattern for myself across my own business and kept hearing it more and more from some of my AI agency clients, and I've been reading it online more and more too, which is something along the lines of, yes, me and my business are probably getting more done with AI, but I'm also producing and seeing more low quality work across my business, whether that's in marketing, internal presentations, customer support replies, or any really asset inside of a business.
01:25I noticed this for myself. Sometimes I look at past things I put out, and I go like, Ben, this is embarrassing, but I also notice it sometimes with things my team members put out.
01:34And, of course, if you look at LinkedIn these days, you might wonder if there are any people actually writing stuff themselves anymore. And this is, of course, increasingly becoming a problem across businesses, which is something they call work slop. And there are even companies that go as far as banning employees from using AI for anything public facing just to protect their brand.
01:52So I started to think about how I could potentially solve this by firstly thinking about what slop actually is. And what I realized is that the term slop is actually subjective. What I might consider slop, one of my team members or someone else might not consider slop.
02:06And it has a lot to do, of course, with your taste and your specific quality bar. If your quality bar lies here, you would consider this I AI output, uh, slop. But if person b's quality bar lies here, you might say this is perfectly fine.
02:20And if person's a quality bar lies here, you might think this is a great output. And honestly, I think this quality bar with more and more people using AI for everything is gonna really set you apart from other people. But what I've also noticed is that my own personal quality bar changes moment to moment where I I don't actually enforce it consistently.
02:38Because when I get tired, for example, when I get frustrated with AI, when I have no time or I have to just get something out quickly, a lot of my principles go out of the door. So my quality bar drops, and I might just put something out into the world that doesn't actually meet my standard.
02:52And this is the first reason I built this scale to enforce my own standard even when I can't be bothered to because you define what good looks like once, and the scale holds you to that on every AI output that goes out into the world. Think of it like a a spell check, but for slop. And then the second reason I built this is, of course, one of the biggest problems for companies is that everyone, every employee in the business have a has a different quality bar in their head, and each individual AI just tends to agree with everything the user says without having any any notion or idea of what the actual quality bar is for the company.
03:25And besides that, AI just reinforces the user. It doesn't necessarily reinforce the truth or the truth within a company. And with more and more employees using AI, it becomes impossible for a company to actually see all of it, let alone adapt it.
03:38There's an excellent article by a 16 z on institutional AI versus individual AI that talks about exactly this problem. The article starts off with a question. AI just made every individual 10 x more productive, but no company became 10 x more valuable as a result.
03:53Where did the productivity go? And the main point he makes is that giving every individual a productivity boost with AI doesn't necessarily add up to a better company because speed of production without a shared direction or quality benchmark, uh, means you're not actually adding value to the company but creating more chaos.
04:10And in order to actually do that, the way a company works and the technology itself need to be redesigned to actually make it useful for an institution, not just individual. Now I thought it's a great article.
04:20It goes a lot more in-depth. Definitely recommend reading through it if you have time. I'll make sure to link in the description below too.
04:24So that's the second reason I built this, to run that same slop check across my entire business. So every AI output, no matter who generated it in my team, actually is aligned with my company's truth and meets my company's quality bar. That's why I'm such a proponent of the AI OS or shared second brain across a business because, of course, then everyone's AI tools and AI agents can actually pull from the company's shared strategy and knowledge docs and, of course, align them a bit more.
04:51But just providing these models with, um, some context doesn't actually guarantee that the outputs they generate are actually well aligned and meet the quality bar of a company. So that's the reason I built this. So every AI output, no matter who generated it in my team, actually meets the company's bar before it goes out.
05:07Now before showing you some examples, let me quickly go over how this actually works. Now the idea again of this d slop scale is, of course, to use it on any AI output before you or any of your team members puts it out into the world, uh, whether that's an internal or external use case to make sure it passes that quality bar, and if not, suggest changes.
05:26Now the way it does that is by first going through a universal slop check, which basically checks universally agreed upon slop patterns like m dashes, AI writing patterns, AI words.
05:39It fact checks the the AI output. It looks at contradictions and unreadable writing. And then it runs the second layer, which is the company slop check, which is the subjective layer.
05:49Does this actually match my or my company's, uh, standards on the tone of voice? Does it match our brand identity if it's a visual? Is this aligned with our company strategy and our company facts?
06:00And does it actually do the job it's supposed to do? And this is, of course, the part that has to be personalized for you or your business. Now the scale will basically do this through the following process.
06:09First, it checks what kind of output this actually is. Is it a marketing post? Is it a customer support reply or an internal document?
06:15Based on that, it def defines which checks it has to do. I have checks for universal slop detection and company specific slop detection, and each of the checks basically has a reference file, uh, with instructions and guidelines that the output should follow.
06:29For example, the AI writing tells doc, looks at common AI content patterns like language, grammar, vocabulary, and style patterns to detect AI writing.
06:40The factual accuracy does a systematic claim check by checking against authoritative sources on the Internet. And for the company specific ones, for example, my voice doc, it has guideline lines on how my brand's voice should sound and not sound.
06:56For example, no Guru energy or no bold claim or big statements with no backing or no new one. Now you can read all of them if you download the scale, you can adapt and personalize it easily, which I'll show you how to do in a second.
07:08Then it spins up three sub agents that grade the output against the universal and company specific criteria. We do this through sub agents to make it more unbiased. And then lastly, it gives you a clear verdict if this is good enough or not, gives you suggestions on how you might wanna change it, and mentions the sources it found it in.
07:26It can be one of three verdicts, good to go, good to go, but might wanna fix some things or not ready. For example, here I ran one of my other skills to do a follow-up email, uh, after a sales call that basically goes through the last sales call transcript and write a a follow-up email, and then I asked it to run the d slop scale on the email copy.
07:44So in this case, the verdict was good to go with a few things you might wanna fix, and it gives a table with an overview of what might be off and a suggestion on how to improve. So it tells you for each of the checks what might be off in this one, for example, the source of the reference file that backs up that claim, and a suggestion for how to improve it.
08:01So in this case, it detected a couple of signs of AI writing. The fact check actually detected that something in the email copy wasn't completely aligned with what actually was said in the call, and it found some small readability and voice issues. Issues.
08:13Now, the point is they're not always gonna be perfectly relevant. That's why you can easily decide which of these suggestions you actually want to apply, and you can then tell Claude right away which fixes you wanna apply, and it can rewrite the copy actually based on a good source of truth.
08:27Here, for example, I ran it after using my LinkedIn skill to write a LinkedIn post, and in this case, it came back with a verdict not ready because the post breaks some hard rules and has many issues. But it can be run on anything. Here, ran it on a newsletter.
08:39Here, I ran it on a sales proposal where it actually also checks the visual guidelines, but I think you get the idea. And the point of this scale is not to expect it to fix everything perfectly instantly. That's not really the goal and the in the scope of this scale because, of course, there can be hundreds of different AI outputs or different assets that it can judge, and it doesn't have enough context around each of the specific use cases, which can literally be hundreds.
09:02But that's not the point of this scale. It just detects things that could that are potentially misaligned with the brand, and if they're completely misaligned, it flags it and tells you this is not ready. You can then decide based on what those suggestions are, what changes you actually wanna apply.
09:16And if you use this across a company or in a team, the nice thing is that they will be a very nice easy helper for all of the team members to improve their outputs fast, but it also makes your team more aware of what's actually important in your business. Basically, it makes them more aware of the quality bar. So I basically added this as a rule for myself and for all my team members to always run this before putting any AI output into the world.
09:40And it can also easily be, uh, added as a last step inside of skills, for example. Uh, so for example, my LinkedIn post writer skill, it could be the last step inside of the skill, and then there's actually one more step to the skill, which is at the end, it will add the run into the d slop log. Basically, it logs every time it runs in order to start detecting common patterns across the business in terms of what's going wrong with these AI outputs, and it logs how often people use it.
10:07In my case, it saves in in my second brain or my AIOS, but it can also save it inside of the skill. And if you already have a shared second brain set up for your company, I think this could be a great add on to combine it with this skill to not only provide everyone's AI tools with better context, but to also make sure that AI outputs are actually checked against the company's benchmark because that's what, of course, the second brain setup can't do or can't guarantee.
10:32Now you can download the skill for free in the first link in the description below. I'll add some blanks where, uh, you need to personalize it for yourself, and then you can ask Claude, uh, to personalize and fill out some of these for yourself, like voice, etcetera. Or if you're interested, we've also built a d slot builder skill that basically walks you through step by step through the process of personalizing this skill for yourself, which is available in my AI accelerator community, uh, in the second link in the description below together with all the other skills and plugins we're building out.
10:59Inside of my accelerator, we also have full courses on how to set up your own second brain, your OS, cloud courses, unlimited one on one, uh, live tech calls, and multiple weekly q and a. So so if you're interested in that, you can check it out in the second link in the description below. Now that's it for this video.
11:13Thank you so much for watching. If you got any value out of it, I highly appreciate, uh, like and subscribe. It really does help me.
11:18If And you wanna learn more about clots and clot scales, you can check out the video here.
The Hook

The bait, then the rug-pull.

Every person who uses AI regularly has published something they are not proud of -- not because the prompt was bad, but because they were tired, rushed, or simply could not be bothered to hold the line that day. Ben AI built the De-Slop skill to make that irrelevant: define your standard once, and a two-layer Claude quality gate enforces it on every output, whether you are in the room or not.

Frameworks

Named ideas worth stealing.

05:07model

The De-Slop Two-Layer Check

  1. Universal Slop Check (AI tells, factual accuracy, consistency, artifacts, readability)
  2. Company Slop Check (voice, visual, company fit, completeness)

Every AI output passes through a universal objective layer first, then a personalized company-standard layer. Three sub-agents run checks in parallel. Output is a three-state verdict plus a source-backed issues table.

Steal forLast step inside any existing Claude skill or AI workflow to enforce quality before publishing
03:07concept

Individual AI vs Institutional AI (a16z)

Individual AI makes each person faster but does not add up to company-level value if there is no shared direction or quality benchmark. The De-Slop skill is the institutional layer that individual tools lack.

Steal forFraming for any pitch about company-wide AI quality or AI OS adoption
02:40concept

The Fluctuating Quality Bar

Personal quality standards are not fixed -- they drop with fatigue, frustration, and time pressure. Written standards enforced by automation are the only reliable substitute for consistent human judgment.

Steal forJustifying any automated review or checklist system to clients skeptical of process overhead
CTA Breakdown

How they asked for the click.

VERBAL ASK
10:33product
You can download the skill for free in the first link in the description below.

Dual CTA: free skill download (link 1) and paid AI Accelerator community (link 2). Well-executed value ladder -- free tool demonstrates the concept, paid community provides personalization support and all other skills.

FROM THE DESCRIPTION
Storyboard

Visual structure at a glance.

hook -- frustration cap
hookhook -- frustration cap00:00
why built -- AI LinkedIn cringe
promisewhy built -- AI LinkedIn cringe00:45
what is AI slop diagram
valuewhat is AI slop diagram01:52
company quality bar split
valuecompany quality bar split03:07
a16z institutional AI article
valuea16z institutional AI article04:00
De-Slop skill open in Claude
valueDe-Slop skill open in Claude05:07
checks table -- which fires when
valuechecks table -- which fires when06:32
sales email -- good to go result
valuesales email -- good to go result07:30
LinkedIn post -- not ready
valueLinkedIn post -- not ready08:40
De-Slop log
valueDe-Slop log09:52
personalization walkthrough
ctapersonalization walkthrough10:33
AI Accelerator community CTA
ctaAI Accelerator community CTA11:00
Frame Gallery

Visual moments.

Watch next

More from this channel + related breakdowns.

Chat about this