How to Fix Claude Content That Reads Like AI Slop

You asked Claude for a caption and got back something that reads like every other post on your feed. You also don't have the energy today to rewrite it into something that sounds like you.

This one is for the founder who edits everyone else's content in their head and still can't catch their own on a low day. I run a business on ADHD and multiple chronic illnesses, and use Claude for content because that’s one of the most exhausting tasks I have. I spent months working out why the output kept landing so damn generic, and what actually changes it. The real cause is the setup feeding Claude, and the fix catches the tells for you instead of relying on a hand-edit you may not have the focus for.

Claude content sounds like generic AI for one reason: the setup feeding it is generic, so the raw draft arrives with the usual tells baked in. The model produces the slop, and that part has nothing to do with you. What keeps the tells in your published work is a separate problem. The usual advice is to catch and rewrite them by hand, and that pass depends on focus you can't reliably count on. Build one reusable scrubbing step, and the tells get caught whether or not you have a good brain day.

Why does my Claude content sound like generic AI?

Claude content sounds generic because the setup feeding it is generic, so the model produces the raw draft with the usual AI tells already baked in.

A language model pulls toward the middle of everything it was trained on. Without specific instruction about how you sound, it hands you competent, forgettable copy. Your saved context and project files are generic, so the output is too.

The draft reads like AI because the model built it from a generic setup. Whether that slop reaches your published work is a separate question, and that is where capacity comes in.

What are the AI tells that make Claude sound generic?

Common AI tells include em dashes, false-contrast phrasing, inflated vocabulary, and an eager helper tone that reads like a customer service script.

The tells to watch for in a raw Claude draft:

• Em dashes scattered through every paragraph, often two or three per section.

• False-contrast lines built on "it's not about X, it's about Y."

• Three-beat phrasing like "clear, confident, and consistent."

• Inflated vocabulary: unlock, elevate, leverage, seamless, transform.

• Helper tone: "great question," "happy to help," "let's dive in."

• Filler hedges: "it's worth noting," "in today's landscape."

The difference on a single line: a raw Claude draft might hand you "In today's landscape, Claude helps you unlock seamless workflows and elevate your business." Scrubbed, the same idea reads "Claude takes the repetitive parts of your workflow off your plate so you can focus your low energy where you need it the most."

When you're depleted, you stop seeing these tells.

Why doesn't editing it yourself fix the problem?

Self-editing fails on low-capacity days because catching AI tells takes fresh focus, and the manual editing pass is the first thing to get dropped when energy is low.

The standard advice is to give Claude examples of your writing and then proofread everything. Both only work when your capacity is steady. For a variable-capacity founder, the proofreading pass is a tax you can't reliably pay. On a good day you catch the em dashes and rewrite the stiff lines. On a flare day you paste and post, because making one more decision is past what you've got.

This has nothing to do with willpower. The editing pass needs the one resource you have least of, so it falls apart on the exact days you lean on Claude hardest. Anything that lives in "remember to scan for X" gets skipped right when it matters.

Does better prompting fix AI slop?

Better prompting lowers how much slop shows up, but it can't cover the capacity half, because you still have to catch what slips through on low-energy days.

Prompting and good project setup reduce the slop at the source. They don't remove the need to catch what still gets through. And building a strong prompt is itself a good-brain-day task, so on the days you most need clean output, you're least able to engineer your way there in the moment. A saved scrub does the good-day work for you on the bad days.

How do you get Claude to sound like you instead of generic AI?

Get Claude to sound like you by saving the voice checks as a reusable skill it runs on every output, so the scrub happens automatically and stops depending on your focus.

The move is to take the checks out of your head and hand them to Claude as a saved instruction set. You write the tells down once and load them into a skill, and Claude runs that scrub on everything it gives you before it reaches you. The draft arrives already checked against em dashes, false contrasts, inflated vocabulary, and helper tone.

Your good-brain-day quality becomes the everyday default, because the check no longer depends on whether today is a good brain day.

What is a Claude skill, and how does it catch AI slop?

A Claude skill is a saved set of instructions Claude applies on its own, so a voice scrub built as a skill runs on every draft without you re-explaining it each time.

Think of a skill as a filter every draft passes through before it lands in front of you. Build one around your own voice rules and Claude checks its output against them automatically. No re-pasting your preferences into a fresh chat, or relying on catching the slop by eye when your energy is tanked and your attention-span is practically non-existent.

This is the exact thing I built for my own content and packaged as Stoppy Sloppy, a free Claude skill that scrubs output against the same patterns I check my own writing against. Install it once and your first-pass drafts come back shippable, so you skip the rewrite step and keep the energy you needed for everything else that day.

What can you do about it on a low-capacity day?

If you have five minutes and not much brain, remove the editing pass from your plate: install a scrubbing skill so the check runs without you.

Grab a skill built for this and let it run on your next draft. You don't have to audit your setup or write anything first.

When you have more in the tank, take it further. Write out your own running list of tells, the specific words and phrases that make you cringe in your own voice, and feed them in so the scrub sharpens to you. The version that runs on your list catches more than any generic pass.

What is the fastest fix for Claude content that sounds like AI?

The fastest fix is a saved scrubbing skill that runs your voice checks on every draft, so the slop gets caught on good days and bad days without you doing the pass by hand.

Your Claude content reads generic because the model builds it from a generic setup. That is on the model. The tells only stay in your published work because the standard fix, hand-proofreading, needs focus you can't count on. Moving the check into a skill does the catching for you, so it keeps working on the days you can't.

Stoppy Sloppy is the free Claude skill that runs that scrub for you. Grab it here 👇🏻