AI automation for WordPress and WooCommerce

If your shop has hundreds or thousands of products whose descriptions, meta descriptions, image alt texts and translations someone is supposed to write by hand, that work usually stays half done. The supplier sends a product name, an EAN code and a few lines of specifications, not sales copy in three languages.

We connect a language model (OpenAI API or a comparable provider) to WordPress and WooCommerce to produce first drafts of the repetitive content and data work: product descriptions from supplier data, first-pass translations, mapping supplier categories to your shop categories, alt texts, summaries and tags. Every result waits for a human review and goes live only after approval. In the ProSwim e-shop we mapped supplier categories to shop categories with AI and corrected the result by hand; in the Nordenbeauty e-shop an AI step keeps short interface texts in sync across languages.

What the work includes

• Choosing the model and provider. OpenAI API or a comparable provider, chosen by the languages you need (smaller languages vary in quality between models), price and where the data is processed.
• Prompt design with your tone rules. We write a system prompt that carries your writing style, banned phrases, length limits and example texts. We test it on real products and adjust until the output is usable.
• Batch processing with WP-CLI or Action Scheduler. The job runs in the background in batches, survives interruptions and resumes where it stopped without processing the same product twice.
• A review step. AI output is saved as a draft or into a separate field, never straight into the public text. A list view lets you approve, edit or reject rows, and only approved text reaches the page.
• Cost caps and logging. Every request is logged with its token count and cost. We set a daily or monthly ceiling that stops the job when reached, so a faulty loop cannot run up the bill.
• Personal data stays out of prompts. Order, customer and review data is not sent to the model; where a use case needs it, we anonymise before sending. We document the data flow so you can add it to your privacy notice.
• An admin screen or WP-CLI command you can rerun. For example “describe all products with an empty description” or “translate new products into Finnish”.
• A support assistant built on your content (optional). A chat window that answers only from your site’s pages, FAQ and delivery terms, and hands over to a person when it cannot find an answer instead of making one up.

Typical situations

A supplier feed delivers three thousand products with a name and specifications but no description or meta description, and nobody finds them in Google. The shop runs in three languages, the translator covers the front page and categories, and product pages stay in English. The supplier’s category tree does not match the shop’s and someone has to sort thousands of products into the right categories. A blog archive has no summaries or tags. Customer support answers the same delivery, returns and sizing questions every day, although the answers are already on the site.

What we need from you

• WordPress administrator access and, where possible, SSH or WP-CLI access to the server.
• A provider account and API key in your company’s name, so that billing and the data processing agreement are yours. We can help set it up.
• Tone rules and five to ten example texts you consider good.
• A glossary: product names, brands and terms that are not translated or are translated in a fixed way.
• A decision on who in your team reviews the results and how often.

What is not included

• Importing the supplier feed and syncing stock: that is a separate service, product feed import. The AI step works on data already in the shop.
• Human translation and editing. AI gives the first version; an editor is responsible for the final quality. See website translation.
• Product photos, dimensions and specifications the supplier does not have. The model does not know them and we do not let it invent them.
• A chat tool with human operators: see online chat.
• Workflows between applications with Make, Zapier or n8n: see automation.
• The provider’s API costs, which are billed directly to your account. We estimate them from the volume before starting.

Who it suits

It suits an e-shop or content-heavy site with hundreds of units of repetitive text and data work and a person who will review the output. With a couple of dozen products it is cheaper to write them by hand. If the goal is for your site to appear in ChatGPT and other AI search answers, that is a separate topic: AI visibility. If the workflow needs logic across several systems or its own data model, the work grows into custom plugin development, which we estimate after the review.