AI Skills for eCommerce: custom assistants in Claude, ChatGPT and Gemini

I will start with something that has surely happened to you: you ask an AI chat to write a product description, a few seconds pass and it creates a decent text. But you do it again a few days later, and the text does not  sound like your brand at all.

This is not a problem with AI as a tool; it has more to do with how you use it. You start from scratch in every conversation, and that is a mistake: you have to give it context again and explain who you are, what you sell and how you speak. And even then… either the result is generic, or you have skipped one of the steps and what it generates is not consistent with the other posts in your store.

The good news is that you can save all of this in a permanent assistant. You just need to use the right feature for the platform you work with (skills in Claude, GPTs in ChatGPT, Gems in Gemini), and you can set it up in no time, without having to study any complicated documentation.

Most eCommerce managers use AI as a machine for one-off answers. They open the chat, ask for something, copy the result and close it. The next day, if the same need comes up… they start all over again. It works, but it’s like having a new employee every morning who needs to be told again how things are done here.

Advanced users of the three biggest AI platforms save that context once and for all. They give it their rules, products and writing style, and from then on it knows who it’s talking to and who it’s talking for. It’s the difference between assigning a task to a stranger and asking someone who has been working in your store for a year.

The idea is to look at and explain how skills work on the market’s leading platforms, because you’ll quickly see that it’s just a matter of names: the philosophy behind them is the same.

In Claude, they are called skills. They are packages of instructions and resources that the model loads only when they are needed for the task you give it, without you having to activate them manually. Anthropic launched them in late 2025 and published the format as an open standard, which means the same skill could be used in more than one tool, and not just in Claude.

ChatGPT offers custom GPTs: tailored versions of ChatGPT where you add instructions, files with your information (what they call knowledge) and, if you like, connections to other tools.

In Gemini, they are called Gems, and their advantage is how well they work with the rest of the Google ecosystem. If your store relies on Gmail, Drive and spreadsheets, a Gem can read and work with those files directly, without you copying and pasting anything from one place to another.

For our practical purposes, all three do the same thing: remember your context so you do not have to repeat it and can reuse it. So the answer to “which one should I choose?” is easy: the one you already use day to day. It is not worth switching platforms just for this, because the real work is not in the tool, but in what you use to configure it.

I imagine you would like an example, so let’s use GPTs from OpenAI as our reference. You will see that creating a reusable resource is very simple: all you have to do is fill in a form.

Go to ChatGPT and, in the left-hand column, expand the “more” option in the menu and click GPT, the one with the 3D cube icon.

If you are only going to create one assistant, make it this one. The most common complaint about AI text is that it sounds like AI: flat, interchangeable, with that brochure-like feel that could work for any store. It can be fixed, but not by writing “make it sound more human” in every prompt and waiting to see what happens. That gives you no control.

The trick is in the examples. When you configure the assistant, do not just tell it “friendly and professional tone”, an empty formula that says nothing. Instead, paste in 3 of your own texts that sound exactly how you want, and, to fine-tune it completely, 3 more that sound exactly like what you don’t want it to use. With that contrast, it stops averaging things out and starts imitating the way you write.

But let’s get even more specific. Add a list of negative keywords: words you never want to see in your texts. Also, fill in the information you always repeat again:

An example of what changes. Without context, you ask for a backpack description and it gives you: “A versatile solution that combines style and functionality for your everyday life.”

With the voice built in and the audience detail that your customers are urban cyclists, the same backpack becomes: “Truly waterproof, with a padded laptop compartment and straps that don’t dig in while you pedal.”

No one would have written the first one; the second sounds like your store. From then on, every product listing, email and ad comes out in your “language”, with much less to correct. You increase productivity and save time and tokens.

With the voice ready, these are the tasks that really save you hours. They work the same way in Claude, ChatGPT or Gemini; basically, only the button’s name changes.

Product listings at scale. The star of the show. Give it the raw specifications from the manufacturer and it turns them into a description in your tone, written for the person buying, not the person making it. If you need to add fifty new products, this turns a task that would take a whole day into a short review session.

Category SEO and metadata. Meta titles and meta descriptions (the text that appears on Google) for category pages are among those things almost nobody writes by hand and almost everyone leaves on autopilot. An assistant that already knows your categories can generate them in batches, optimized and with good judgment.

Customer service and reviews. Draft replies to common questions and reviews, in your tone and without sounding copied and pasted. Not so it can reply on its own, be careful with that, but so the draft is ready for you to approve in 10 seconds.

Ads and newsletters. Five versions of an ad to test which converts best, or an outline for the week’s newsletter based on four loose ideas. AI is at its best when generating options, but you choose the one that works.

Analysis of your numbers. Gemini and Claude handle large spreadsheets well. Give them the month’s sales spreadsheet and ask which products have fallen, which are trending up or where your margin is slipping away. Instead of opening a file you have to analyze, you get answers quickly.

Selling internationally. If you are thinking about selling abroad, an assistant with a glossary of your brand can translate and adapt product listings into another language without losing your tone or expressions. It’s nothing like running the text through an automatic translator and hoping for the best.

An assistant is only as good as the context you give it. If you give it four vague instructions, it returns vague text, and no platform can fix that. Quality doesn’t depend on which one you choose; it depends on the time you spend at the beginning explaining your store properly. You invest that time once, and get it back every week.

Another basic point is manual review. Always do it before publishing; always means always. These assistants sometimes make things up with complete confidence: a delivery time that is not yours, a product feature they have pulled out of thin air… In marketing copy, the risk is relatively low, but a false detail in a product listing could cost you a return or a bad review.

Use it to get 90% of the work done; you provide the final 10% that makes it credible.

A note about stores of ready-made assistants, the GPT Store and the skill directories that are appearing: they are fine for generic tasks, such as “description generator”. But for anything that depends on your brand, a borrowed assistant will never sound like you, and you already know how important differentiation is in this channel.

I’d say it is absolutely worth setting up your own; make the most of the fact that you now know how.

Do not try to set up all 7 assistants at once. Choose the task that takes up most of your time this week, which is probably product listings, and spend half an hour creating that one assistant in the tool you already have open. Give it your examples, rules and data, test it with three real products, and adjust it until it sounds like you.

Once that one works, set up the next. In a month, you will have a small team that knows your store inside out and does not go home in the evening. More or less what you have been asking AI for all this time, without knowing you could save and reuse it.