What is AI slop, and how do we stop our team sending it?

AI slop is writing a model produced and nobody edited. It is padded, vague and the same shape every time, and clients spot it within a line or two. Your team stops sending it when Claude starts from your house style instead of a blank page, and when people are trained to edit before they send.

Below: the tells to look for, three rewrites side by side, why it happens, and the fixes that last.

What does AI slop look like?

You will know most of these from your own inbox.

• An opener that says nothing. “I hope this email finds you well.” “In today’s fast-paced world.” The reader skips it, and the point arrives a paragraph late.
• The same dead words. Leverage, unlock, seamless, robust, empower, holistic, cutting-edge. Nobody uses them out loud. A model uses them constantly.
• The fake reveal. “It’s not about the technology, it’s about the people.” It sets up a claim nobody made so it can knock it down. Say the second half and stop.
• Telling instead of showing. “This is a crucial point.” “The implications are significant.” If it matters, the reader will see why from the facts.
• Claims nobody can check. “Studies show.” “Experts agree.” Name the source, or cut the claim.
• Everything in threes. Three adjectives, three bullets, three short punchy sentences in a row. Real thinking comes in odd numbers and uneven lengths.
• The recap ending. A closing paragraph that repeats the email the reader has just read.
• The portability test fails. If a sentence could go in a competitor’s email unchanged, it is saying nothing about you.

What does the fix look like?

Three rewrites, in the way I write. They are illustrations to show the difference, not client stories, and the prices in the last one are our published ones.

1. A follow-up email after a sales call

The rewrite names the job, the person, the time it costs and the next step. Every line is about this client and nobody else.

2. A LinkedIn post

The first version has an opinion about AI in general. The second has a method, a time of day and a way to check it worked.

3. A paragraph from a proposal

A buyer can check every claim in the second version, and the finance director can see the total without a call.

Why does AI slop happen?

Because the model is working from nothing. Someone opens a blank chat, types “write a follow-up email to a prospect”, and gets the average of every follow-up email it has seen. It knows nothing about the client, the call, your prices or how your business sounds, so it fills the space with the safest words it has. Then the draft goes out unread, because nobody showed that person what to cut.

It is the same root cause as Claude licences going unused: the tool was handed out without a set-up or any training on the work people do.

How do we stop our team sending it?

1. Give Claude your house style once

Write down how your business sounds, the words you never use and a few examples of your best writing, and put it in the shared set-up so every draft starts there. When the whole team works from one Claude set-up, the whole business drafts from the same rules, instead of each person correcting the model on their own.

2. Train people to edit before they send

The first draft is raw material. Before anything goes to a client, three questions catch most of it:

• Would I say this out loud to this client?
• Can I back up every claim with something I could show them?
• Could this sentence go in a competitor’s email unchanged?

Pin those three questions where people draft. Our Claude training days run on each person’s real work, which is the best place to practise the habit.

3. Keep a list of words you never use

We keep one, and it runs to more than a hundred words and phrases. You do not need ours. Start with ten that make you wince when a supplier sends them, and add to it every time one gets through.

4. A person signs off anything client-facing

The plugin we set up for teams drafts and never sends, so a person reads every client email before it leaves. That rule exists for safety, and it catches slop too.

Is there a free tool that removes AI slop?

Yes. Peter Yang’s no-ai-slop is a free, open-source skill for Claude, released under the MIT licence. Give it a draft and it edits out the patterns above, or flags them without rewriting if you only want a check. We keep a copy alongside our own rules. It works best with your house style in the set-up and a person who reads the result before it goes.

Can you help our team write without the slop?

Yes. Our Claude training for teams starts by putting everyone on one shared Claude set-up, then trains each room of up to 10 on its own work, in person or remote, with a return visit at day 30. Prices are on our pricing page.

Read next: Claude licences going unused · How to train your team on Claude · Book a call