How to integrate AI into WordPress to generate content automatically
If you look up how to build a WordPress site with AI, you’ll find two different things mixed together. One is building the site itself, with an assistant that suggests a structure, design and pages from a description of your business, like the ones some hosting providers now bundle into their WordPress installer. The other starts once the site is live and you need to fill it with product pages, blog posts and service pages, and keep updating the copy as it goes stale. This article is about the second one, which is where the hours go, year after year.
The practical question for a small business or an online store is which part of that work a model like ChatGPT, Claude or Gemini can do, how its text gets into WordPress without anyone copying and pasting, and what a person still has to do before hitting Publish.
The ways to bring AI into WordPress
Almost everyone starts by opening ChatGPT in another tab, asking for a draft and pasting it into the editor. It works, and if you publish one post a month it may be all you need, but it has a hidden cost. Every time, you have to explain to the model again who you are, what you sell and how you talk, the formatting gets scrambled when you paste, and there’s no record of which prompt produced which text. The other two options exist to take that repetitive work off your plate, each in its own way.
An AI plugin inside the block editor
Once you install an AI plugin, a new button or block shows up in the block editor. From there, whoever runs the store can ask for a description of a new product, shorten a paragraph, suggest a title or write the meta description, without switching windows.
Under the hood, most of them do the same thing. They connect to the API of a model provider such as OpenAI, Anthropic or Google using a key you supply, or they run on the plugin’s own credits, which you pay for separately. The difference matters: in the first case you pay for usage directly to the model provider, and in the second you depend on the pricing and limits the plugin sets.
The upside is that the writer stays in control, sentence by sentence. The downside is exactly the same thing: someone has to be sitting at the editor, so the plugin speeds up writing but doesn’t automate it. For a low-volume site with someone who already spends a bit of time each week on content, it’s usually the option with the best effort-to-results ratio.
An n8n or Make workflow that creates drafts through the REST API
WordPress comes with a built-in REST API, a door through which another program can create and edit posts and pages, and with WooCommerce, products too. Automation tools like n8n and Make have ready-made modules to talk to it, so you can set up a workflow that runs without anyone opening the WordPress dashboard.
Take a store: every time someone adds a row to a Google Sheets spreadsheet with the product name, its dimensions and a few notes, the workflow sends that data to Claude or ChatGPT along with a fixed set of instructions on tone and structure, gets the description back and creates it in WordPress as a draft, with the category already assigned. In the morning, someone opens the list of drafts, reads them, fixes them and publishes.

That “as a draft” is the key decision in the setup. The workflow could publish directly, and technically nothing stops it, but then whatever the model makes up goes live without anyone having read it. Keeping it as a draft leaves the automation doing the heavy lifting, which is the writing, and keeps a person on the part you can’t delegate.
For the workflow to get into WordPress, you need a user with permission to create posts and an application password, which you generate from that user’s profile and can revoke without touching their regular password. Setting up the first workflow takes a few days of testing and tweaking the instructions, and after that it needs maintenance every time something changes in any of the pieces.
How to choose based on volume, budget and who maintains it
The first thing that tips the scales is how many pieces you publish a month. If it’s a handful of blog posts, the plugin inside the editor is enough, and setting up an external workflow would mean paying for infrastructure to do a job that takes an afternoon. If it’s dozens of product pages, or the catalog turns over every season, automation starts paying back the hours it takes to build.
The budget breaks down differently. With the plugin, you pay for its license, if it has one, and for model usage, which grows with how much you generate. With automation, the big cost is the setup and the maintenance hours. Then there’s the question of who maintains it when something breaks. A plugin gets updated by its developer. An n8n or Make workflow has to be understood by someone, on your team or outside it, because sooner or later a step stops working and the draft never shows up. If there’s no one who knows how to open that workflow and see where it stopped, we recommend the more modest option, because a workflow nobody knows how to open ends up being one more problem.
When automating content isn’t worth it for a small business
A dental clinic with a small website and a blog that gets updated whenever someone has a spare moment is a good example. A workflow that generated posts on its own would give it more text than anyone at the clinic can read carefully, in a field where getting a detail wrong about a treatment has consequences.
Before automating, it’s worth running the numbers: how much new content you actually need, which isn’t the same as how much you could publish; how long it takes someone who knows the business to review each generated text, which for health or legal topics gets close to the time it takes to write it from scratch; and how much it costs to build and maintain the workflow. If reviewing eats up almost all the savings, automation just adds one more piece that can break.
In some industries AI doesn’t deliver much because of the nature of the content. A firm that writes about specific regulatory changes, a manufacturer with highly technical products, or a neighborhood business whose value lies in what happens on its street need information the model doesn’t have, and without it you get copy that’s a carbon copy of the competition’s.
It also isn’t worth it when the site’s problem lies elsewhere. If it gets no traffic because of a technical issue or because nobody links to it, publishing more content won’t fix it: automating just speeds up the production of pages nobody reads.
With most small businesses, we start the same way. A plugin for a few weeks, keeping track of how long it really takes to review each text, and with that number in hand we decide whether the volume justifies building something more.
AI plugins and tools for WordPress, by what they solve
Search for “AI” in the official WordPress plugin directory and you’ll get a huge number of results, many of which do the same thing with a different interface. To keep your bearings, it helps to group them by what they solve:
- Writing assistants inside the editor. Jetpack AI Assistant, by Automattic, adds a block that generates and rewrites text right in the editor. AI Engine, by Meow Apps, lets you connect your own key from several model providers and, besides writing, can be used to set up a chatbot on your site.
- SEO plugins with AI features. Rank Math, with its Content AI module, and All in One SEO, with its content generator, focus on titles, meta descriptions and keyword-related suggestions. They mainly make sense if you already use one of them.
- External automation tools. n8n, Make and Zapier aren’t installed in WordPress; they connect from the outside through the REST API, and they’re what make the draft workflow from the previous section possible.
Before installing any of them, take the time to read its plugin page carefully. Check whether it works with your own API key or with plugin credits, which models it supports, when it was last updated and whether it lists compatibility with your version of WordPress. An abandoned AI plugin stops working when the provider changes its API, and it stays installed on your site.
There’s one thing almost nobody checks at first: everything you type into the plugin is sent to the model provider for processing, so don’t paste in customers’ personal data, quotes with names on them or information you wouldn’t email to a third party. Each provider’s data usage terms are on its website.
What to check before publishing AI-generated text
Reviewing works better with a fixed checklist than with a quick read to see how it sounds.
Facts, figures and proper names
A language model will write that a product has a three-year warranty with the same confidence whether or not that detail was in what you gave it. The model completes the sentence with whatever usually appears in similar texts, which is why its mistakes sound believable.
What you need to check one by one:
- prices, dimensions and delivery times
- warranties and return policies
- names of regulations, agencies and brands
- statistics and links
The rule of thumb: any number or proper name that doesn’t come from your own data gets deleted or checked against its source, and if you can’t find the source, it gets deleted.
Brand voice and filler phrases
Generated text tends to open by restating the title in other words, to close with a paragraph summarizing what’s already been said, and to use three adjectives where one would do. If your store talks to customers casually and the text comes back stiff and formal, or the other way around, it shows in the very first line.
The shape gives it away too. Every paragraph the same length, every section with the same number of bullet points, openers like “it’s important to keep in mind that” that add nothing. What works is cutting ruthlessly: if a sentence doesn’t tell the reader something new, it goes.
If the same mistakes keep showing up text after text, the fix is in the instructions the model gets. Write down what you correct for a couple of weeks and turn it into rules within those instructions.
What business data AI needs so it doesn’t write generic copy
Ask a model for a description of an oak table without giving it anything else and it’ll hand you the same one it gives the store next door. The model knows how to write; what it doesn’t know is anything about your business, and the only way it can know is if you give it that information with every request.

The first source is your catalog. If your store runs on WooCommerce, you can export it to a spreadsheet with attributes, dimensions, materials and prices, and that file is the source the model should pull its data from instead of making it up. For a service business, the equivalent is a sheet for each service covering what it includes, what it doesn’t, who it’s for and how long it takes.
The second is the questions your customers actually ask. The ones that come in by email, over WhatsApp, at the counter or on the phone, written in their own words. A list of those questions makes the text answer doubts real people have, instead of the ones the model imagines.
The third is your own cases. A project that went well, a common objection and how you answer it, told without customer names. It’s the one thing your competitors can’t copy.
All of this goes into a single standing document that the plugin or the n8n workflow sends to the model with every request. When you change a price, you update that document and new texts come out with the right figure.
If you’d like to know which option fits your site’s volume and who could maintain it, tell us about your case and we’ll look at it with you.