ChatGPT for your product descriptions — the limits of copy-paste

"I already do it with ChatGPT, for free." That is the objection we hear most often — and it is a fair one. Yes, ChatGPT (or Claude, or Gemini) writes decent product descriptions when you give it a good prompt. The real question is not "does it work?" but "does it hold at catalogue scale?" Let us compare honestly, use case by use case.

This article covers product descriptions only. For the broader choice between a copywriter, a general-purpose chat and a specialised tool — turnaround, price, quality, across all of a store's content — the underlying comparison is in AI, copywriter or ChatGPT.

What ChatGPT does very well

Let us start by giving it its due. For a one-off description, the general-purpose chat is unbeatable on flexibility:

• You paste the specifications, you get fluent text in thirty seconds.
• You iterate live: "shorter", "more technical", "add an FAQ".
• It costs nothing, or almost nothing.

If you have twelve products and some time, a well-driven general-purpose chat is enough. The problem does not appear at twelve products — it appears at two hundred.

Where copy-paste breaks

Context is lost with every description

ChatGPT does not know your catalogue. With every product you start over: copy the name, the specifications, the price, restate the tone, repeat the constraints. Count five to ten minutes of handling per description, review excluded — that is 30 to 60 hours for 400 references, in copy-paste alone. The promise of automation dissolves into logistics.

Consistency drifts

With no memory of your catalogue, every generation reinvents the style. Description 40 is casual, description 41 is formal; one uses emoji, the other headings in capitals. Across a whole catalogue that drift shows — and fixing consistency costs more than the writing did.

The SEO tags fall through the cracks

A product page that ranks is not just a description: it is a 70-character title tag and a 160-character meta description, set on a target keyword, with a preview of what Google will display. The general-purpose chat can produce them if you ask — but who checks the lengths, description after description? In practice, nobody: the metas stay empty and the click is lost.

Paraphrase is not differentiation

The common reflex — "rewrite this supplier description" — produces a paraphrase: same information, different words. But your underlying problem, duplicate content, is informational as much as textual. Without new data (your structured attributes, your angle, your photo), the text stays interchangeable.

And publishing stays manual

The last mile is the dumbest one: copying the result into the PrestaShop back office, field by field, product by product, language by language. That is where "I'll rewrite the whole catalogue with ChatGPT" projects die — at product 37, on a Tuesday evening.

What a tool plugged into the catalogue changes

A specialised tool like Publium does not use better AI — it uses the same family of models, wired differently:

• It knows your products. The catalogue is synchronised (name, attributes, categories, price, photo): no copy-paste, the context is already there.
• It structures the SEO output. Description plus title tag plus meta description, in the recommended formats, with a Google result preview and checks on keyword presence — for every product page, without thinking about it.
• It looks at the photo. For products with no data, generation from the cover image describes material, shape and use — something no text prompt can honestly invent.
• It publishes. One click, and the page is updated in PrestaShop through the module's API, in the right language, after your review.
• It holds consistency. Same tone, same formats, same structure across the whole catalogue.

The difference is not the quality of a single description — on one description, a good prompter matches it. The difference is what is still true at description number 200: time per item, consistency, tags, publishing.

The comparison in figures

For a 200-product catalogue, observed orders of magnitude:

• ChatGPT alone: 5 to 10 min of handling per description plus review plus copying into the back office ≈ 40 to 60 hours. Near-zero software cost, high human cost, fragile consistency.
• Freelance copywriter: £80 to £150 per worked description ≈ £16,000 to £30,000. Excellent quality, turnaround in months.
• A tool plugged into the catalogue: batch generation, around 5 min of review per item ≈ 15 to 20 human hours plus a subscription. Consistency and tags guaranteed by construction.

We put the three approaches side by side, criterion by criterion, in our full comparison of generic AI, copywriter and specialised tool if you want the detail.

Our honest advice

Use both. The general-purpose chat stays excellent for exploring an angle, unsticking a text, brainstorming arguments — keep it. But to bring a whole catalogue up to standard — differentiated descriptions, tags in the right formats, direct publishing — specialised tooling is not a luxury, it is what separates "I rewrote ten descriptions over a weekend" from "my catalogue is up to date".

The simplest thing is to judge on the evidence: ask for a demo, we generate a description on your worst supplier text, and you compare it with your best ChatGPT prompt. The verdict takes fifteen minutes.