You have four hundred products and you copied the descriptions from the supplier's catalog. Just like those of your twelve competitors, who copied the same catalog. Google knows it, and that's why none of you appear.
Artificial intelligence solves the problem of volume, but misused it creates a worse one: four hundred descriptions that sound exactly alike, full of "ideal solution for" and "don't miss this opportunity." That doesn't convert and doesn't rank either.
This article is about the method to prevent that from happening.
The starting error: asking for the description directly
The flow that almost everyone tries first is this: "write a product description for running shoes." And the result is correct, publishable, and absolutely interchangeable with any other.
The problem is not the model, it's the input. You've given it three words and expect a specific text. The AI fills the void with what is statistically most probable, which is exactly the average of everything written about running shoes. The average doesn't sell.
The rule is uncomfortable but simple: the quality of the sheet depends on the amount of specific data you give it, not on the prompt. A sophisticated prompt with poor data yields poor text.
Step 1: assemble the technical sheet before writing anything
Before opening any AI tool, gather by product:
- Objective data: measurements, weight, materials, composition, compatibilities, box contents.
- Who it's for: the specific customer profile, not "everyone".
- What problem it solves: the real reason why someone buys it.
- How it differs from the previous model or the one next to it in your own catalog.
- What it IS NOT: what it's not for. This is the field that most people omit and the one that most differentiates the final text.
- Real objections: what people ask before buying, taken from your emails and your customer service.
That last point is gold and your competition doesn't have it, because they are your conversations with your customers. If you sell mattresses and the most common question you get is whether it works for a trundle bed, that answer on the product sheet is worth more than three paragraphs of adjectives.
You can set all this up in a spreadsheet with one column per field. If you have many products, export the catalog from Products > All Products and work on that CSV.
Step 2: set the voice only once
If you generate each product sheet in a new conversation without further context, you will have four hundred different voices. Define yours once and reuse it in all prompts.
A useful voice block includes: who you are addressing and how you speak to them (formal or informal), what length you want, what words are forbidden, and a sample of one of your descriptions that already works.
Example of a reusable block:
"You write for a Spanish online climbing gear store. Customer: people who already climb, not absolute beginners; they know what a harness and a belayer are. Tone: direct and technical, informal (tú), without commercial exaggerations. Length: 120-180 words. Forbidden to use: revolutionary, incredible, ideal solution, don't miss out, in today's world, immerse yourself. Never invent specifications that I haven't given you. If a relevant piece of data is missing, write [MISSING: data name] in its place."
That last instruction is the most important of all. Without it, the model will fill in the blanks by inventing data, and you will publish false specifications in your store. With it, it marks where you need to complete it yourself.
Step 3: the generation prompt
With the voice set and the technical sheet in front, the prompt per product is short:
"[voice block] — Generate the description of this product with this structure: (1) an opening sentence that says what it is and for whom, without adjectives; (2) two or three short paragraphs about real use, not about features; (3) a list of technical specifications exactly as I give them to you; (4) a line about what it is NOT for. Product data: [paste technical sheet here]."
Look at point 2: real use, not features. "Vibram sole" is a feature. "Grips on wet rock, which is where soft soles slip" is real use. AI does that translation well if you ask it explicitly and give it the starting data.
And point 4, the "what it's not for," is counterintuitive but works: it builds trust, reduces returns, and, incidentally, is content that no competitor who copies the supplier's catalog will have.
Step 4: short description and long description are different
WooCommerce has two fields and many people put the same thing in both or leave one empty.
The short description appears next to the price and the buy button. It's what someone who is deciding reads. It should answer in two or three lines: what it is, for whom, and why this one and not another.
The long description goes in the tab below. It's read by someone who is already quite convinced and is looking for details: specifications, measurements, compatibilities, care.
Generate each one with its own prompt. Asking for "a description" and then splitting it gives a worse result than asking for both separately, because they fulfill different functions.
Step 5: quality control, which is where you win
This step is not skipped. Before publishing anything, review:
- Invented data. Look for any number, measurement, or material that wasn't in your technical sheet. If something appears that you didn't provide, delete it. It's the most serious and most frequent error.
- Repetition between products. Take five product sheets of the same type and read them consecutively. If they start the same or share phrases, adjust the prompt to force variety.
- Filler words. Even if you prohibit them, some slip through. Look for "ideal," "perfect for," "not only... but also" and rewrite them.
- Claims that commit you. Special care if you sell cosmetics, supplements, or any regulated product: an AI doesn't know what you can legally promise and what you cannot.
A quick way to do point 2 is to ask the AI itself: paste ten already generated descriptions and ask it to point out the phrases that repeat among them. It detects patterns that you miss due to saturation.
What AI won't fix in your store
It's worth saying clearly, because a lot of hype is sold with this.
Excellent descriptions don't compensate for bad photos. Nor do they compensate for shipping costs that appear by surprise in the last step, or a store that takes eight seconds to load, or a payment gateway that doesn't offer the method your customer uses. If your conversion is low, the text is probably not the first problem.
And about SEO: unique descriptions are a necessary condition for positioning product sheets, but not enough. A well-written product sheet in a store without authority will still not appear. It's one piece of the puzzle, not a magic lever.
If you're setting up the store now, our guide on how to create an online store with WooCommerce covers the basics, and the one for payment gateways in WooCommerce solves the part that takes away the most sales when it's poorly chosen.
Frequently Asked Questions
¿Penaliza Google el contenido generado con IA?
Google has explicitly stated that it values useful content regardless of how it was produced. What it penalizes is low-quality content made in bulk to manipulate results. A product sheet generated with AI from real data, reviewed, and with information that helps in purchasing, does not fall into that category. Four hundred cloned, unreviewed product sheets, yes.
How many products can I do at once?
Technically many; in practice, as many as you can review. The bottleneck is quality control, not generation. A realistic pace is in batches of twenty or thirty: you generate, review, adjust the prompt with what you've learned, and continue. The next batch comes out better than the previous one.
Is an AI plugin integrated into WooCommerce worth it?
It depends on the volume. With fewer than fifty products, manual work with a spreadsheet and any AI is more than enough and gives you more control. With hundreds of references that change often, a plugin that reads product data directly saves a lot of copying. We review several options in our selection of best AI plugins for WordPress.
And the descriptions I already have published?
Don't rewrite them all at once. Start with the products that sell the most and those that receive visits but don't convert: that's where a better product sheet makes a difference in euros. The rest can wait.
Summary
The complete method fits into five lines: gather specific data for each product, set the voice only once, generate the short and long descriptions separately, review for invented data and repetitions, and publish in small batches.
The difference between a store with product sheets that convert and one with automatic filler is not in the tool they use. It's in the work of gathering real information before asking for anything, and in reviewing afterwards. That's still manual, and it's exactly what your competition won't do.
If you also sell outside of Spain, the next step is to translate that catalog without it breaking along the way, which is much more complex than it seems. And if you're missing utilities for day-to-day tasks, take a look at our free tools library.
