Diagnosis · product description / ecommerce copy
Why your product description sounds like ChatGPT
Every marketplace listing reaches for the same handful of adjectives, and shoppers stopped trusting adjectives around the time ad copy did.
Product descriptions sound like ChatGPT because the model fills empty specifics with hype: game-changers, seamless upgrades, promises to elevate your day. Buyers already discount that language after a decade of ad copy. The real cost is unverifiability: a measurement, a material, a comparison against a named alternative reads as credible no matter who drafted it, while an adjective with nothing under it reads as generated, whatever words carry it.
Why this happens to product description / ecommerce copy in particular
A product description prompt is usually the thinnest brief a model gets: a product name, maybe three bullet features, and an instruction to make it exciting or persuasive. With no voice, no comparison, and no customer detail to work from, the model reaches for the only material it has, adjectives that apply to almost anything. Marketplace conventions make it worse: a short word limit rewards density over specifics, so a seller pastes the model's output whole rather than trimming it. And the incentive runs the wrong way. A generic superlative never gets fact-checked by a return request the way a specific number does, so there is no immediate cost to leaving the hype in, only a slow one: buyers who have read the same five adjectives on fifty other listings and stopped reading them at all.
The tells that show up here
Not every AI tell appears everywhere. These are the ones that cluster on this surface, and what it is about the format that produces them.
- game-changer
- A feature-thin brief pushes the model toward the one claim that needs no evidence: that this changes everything. It shows up in listings for phone cases as readily as for software, because it never has to be specific to be used.
- groundbreaking
- Sellers ask for copy that sounds exciting, and the model reaches for significance words instead of details, since it has no comparison point available to make the product sound genuinely new.
- not-only-but
- Bundled feature lists get compressed into one clause to hit a tight word count, joining two claims with a connector that was never doing real work, which reads as padding once you notice the seam.
- bold-header-list
- Marketplace listings train the model on their own convention: a bolded spec label followed by a colon. It repeats that format even in the short, single-paragraph description where a shopper wants one paragraph, not a spec sheet.
Before and after
A generic AI draft on the left. On the right, the real output of running that draft through underslop, checked so it adds no name or number the draft did not already carry. The voice is The Operator, concrete and unhurried
Introducing the must-have accessory of the season! This groundbreaking backpack isn't just a bag, it's a lifestyle upgrade. Not only does it feature a padded laptop sleeve, but it also boasts a sleek, water-resistant design that seamlessly elevates your everyday commute. It's the ultimate game-changer for anyone on the go, whether you're crushing it at the office or exploring the great outdoors. Order today and experience the difference for yourself!
A padded laptop sleeve and water-resistant shell. That's what this backpack offers. It works for the office and for the trail. Order it if that sounds useful.
How to fix it yourself
- Circle every adjective with no unit behind it Go through the description and mark words that describe a feeling rather than a fact, like premium or seamless. Anything you cannot attach a number, material, or comparison to is a candidate to cut.
- Replace each hype word with the spec it stands in for For every adjective you circled, ask what measurable claim it was standing in for. A vague durability claim might really mean ripstop nylon or a two-year warranty. If there is no real answer, the sentence has nothing under it and should go.
- Read the description as a return request Ask which sentence a buyer could hold you to after the product arrives. Keep the ones that survive that test. Cut or rewrite the ones that only survive as long as no one checks them.
- Run the trimmed draft through underslop for the register, not the facts Once the unverifiable claims are gone, paste what's left into underslop and set a plain, confident voice. It will not add features back in; it only fixes how the remaining claims read.
Questions
How to make it not sound like ChatGPT?
Cut every claim a shopper cannot check once the product arrives; that is most of what reads as generated. Keep the measurements, materials, and named comparisons, and state them once, plainly. The register fixes itself once the unverifiable adjectives are gone, because there is nothing left to inflate.
How to get AI to stop writing like AI?
Give it less room to guess. A prompt with a name, a spec, and a comparison produces copy with less hype than a prompt asking for something exciting, because the model fills whatever gap you leave with adjectives instead of facts. Feed it the facts and it has less reason to reach for hype.
What does ai sound like in writing?
In product copy specifically, it sounds like a claim nobody could act on: a seamless integration, an everyday elevated, a total transformation. None of those describe a checkable fact about the object. That absence, not any particular word, is the real signal.
Does removing the hype language hurt sales copy?
No. Buyers already discount superlative language, so cutting it barely touches persuasion, since that language was not persuading anyone to begin with. What is left, the actual measurement or material, is the part a buyer weighs when deciding whether to buy.
This page explains the cause. underslop is the edit: paste up to 500 words on the free tier, and it marks the tells it could not remove. Your text is held for 0 seconds.