AI in WordPress and CMS: a guide for B2B and e-commerce

In short: AI in WordPress and other CMS platforms is mainly useful for four jobs: drafting content, translating, handling basic SEO and answering customers. The official WordPress AI plugin is free, but every request to the model is billed by the provider. It works when a person reviews every text, the data is clean and recurring costs stay under control. Once you need complex processes, pair the CMS with external automation.

Key takeaways

The CMS manages content; AI speeds up producing and maintaining it, but it doesn't decide what goes live.
Since WordPress 7.0 (May 2026), connecting to AI models is part of the CMS core; ready-made features come from a separate plugin, and almost all of them are still experimental.
Google doesn't penalise AI-written content, but it does penalise pages mass-produced without value.
Personalisation based on visitor behaviour needs consent to profiling cookies in the EU.
When AI has to work with your ERP, CRM and catalogue, you need a dedicated automation layer.

Why AI in your CMS matters now

Your CMS is where everything a customer reads gets published: product pages, service pages, articles, documentation. According to W3Techs, as of November 2026 WordPress runs 39.3% of all websites and 58.1% of those whose CMS is known; Shopify comes next with 5.3% of all websites. Whatever changes in these tools affects almost every company with a website.

WordPress 7.0, released on 20 May 2026, brought an AI client into core that lets the site talk to generative models. For an online store or a B2B company the question is no longer whether to use AI in the CMS, but for which tasks and with what safeguards.

What is a CMS, and what does AI change?

A CMS (content management system) is software that lets you create, edit and publish website content without writing code: copy, images, pages, products.

AI in a CMS means the features that use language or vision models to suggest content, metadata, translations or replies right inside the editor. It changes the pace of work: a blank page becomes a draft to review. It doesn't change accountability: whoever publishes is responsible for what the site says.

A headless CMS, by contrast, only manages content and delivers it through an API to a separate front end, such as a React app or a product configurator: the usual route once a website turns into an application.

Which AI features actually help an online store or a B2B company?

The ones that remove repetitive work without removing control. The table sums them up with an example for each type of business.

Feature	E-commerce example	B2B example	Human oversight
Content drafts	Product descriptions from spec sheets and attributes	Industry application pages from engineers' notes	High: data, claims and tone
Translation	Catalogue in German and French	Datasheets for overseas distributors	High: technical terminology
Basic SEO	Titles, meta descriptions and alt text	Heading structure and FAQs on service pages	Medium: spot checks
Classification	Consistent categories and attributes across thousands of SKUs	Taxonomy for case studies and documentation	Medium: rules set in advance
Content updates	Product pages with outdated data, photos or compatibility	Service pages and documentation after a range change	High: check the new data
Recommendations	Related and complementary products	Related articles and documents by industry	Medium: rules and consent
Customer support	Answers on shipping, returns and stock	First-pass qualification of quote requests	High: hand-off to a person

A hypothetical scenario: a B2B components distributor with 1,500 product pages in Italian only. AI produces the first English version of the pages and the image alt text within days; an engineer reviews the terminology and a spot check covers the metadata. Skip that review and you risk publishing the wrong units of measure in a catalogue buyers use to place orders. On translation, see how to choose AI translation software.

How does AI work in WordPress today?

AI in WordPress now operates on three levels. Knowing which one you're using tells you where your data goes and who controls it.

Core. WordPress 6.9 introduced the Abilities API, a central registry where core, plugins and themes can declare what the site can do, with a schema and explicit permissions, so automations and AI agents can find and use those capabilities. WordPress 7.0 added the AI client and a Connectors screen for model provider credentials.
Plugins. The official AI plugin, built by the WordPress AI team, adds the ready-to-use features (details in the next section); on top of that come third-party plugins for SEO, translation and chatbots.
External agents. The MCP Adapter, one of the WordPress AI team's projects, translates these abilities into Model Context Protocol format, so an AI assistant can create a draft or update content within the permissions you grant it.

What does the official WordPress AI plugin do?

The plugin is simply called "AI". It's published on WordPress.org and isn't part of core: you install it, and once activated it switches nothing on until you choose what to enable. To use it you need a connector plugin (for example for Anthropic, Google or OpenAI) and an API key from the provider, entered on the Connectors screen.

As of 10 November 2026, the only feature marked stable is image generation and editing. Everything else is labelled an "experiment", meaning it's still in development, and WordPress itself recommends trying experiments on a staging site before using them on a live one. The experiments most useful for a business website are these:

Content: titles, excerpts, summaries, rewriting blocks, and translating paragraphs and headings into another language.
SEO and accessibility: meta descriptions (which plug into several SEO plugins), permalink slugs and image alt text.
Organisation: tag and category suggestions, plus block-by-block editorial notes on accessibility, readability, grammar and SEO.
Control: a log of AI requests, administrator approval before other plugins can use the connectors, and encryption of API keys stored in the database.

Two details matter more than the rest. Some experiments run on their own: comment moderation, for instance, checks every new comment, including visitors' comments, so it sends third-party text to the provider. And while the plugin is free, every request is billed by the AI provider to the account whose key is entered.

What does AI offer on WooCommerce and Shopify?

On WooCommerce, the e-commerce platform built on WordPress, there's a dedicated extension, WooCommerce AI. It adds an assistant to the admin that answers questions about your store, works on unpublished product drafts and proposes changes for you to review before applying them, with undo for supported operations. It requires connecting your store and user account to WordPress.com.

On Shopify, the Sidekick assistant is included with your plan (features and limits vary by plan) and works from the admin on product descriptions, email, store analysis and images. It respects admin permissions: each staff member only sees the data and features they're authorised to use.

The difference is the model. On WordPress and WooCommerce you choose the AI provider and manage keys, costs and plugins; on a SaaS platform like Shopify you choose less and manage less: convenient at first, limiting when you need custom integrations. For a broader view of how online trade between businesses is changing, read the piece on the future of B2B e-commerce; more examples are in the blog's websites and e-commerce section.

Can AI personalise your website for each visitor?

Yes, and it's one of the most quoted promises: recommendation algorithms that suggest relevant products or articles based on what the visitor has already viewed. In an online store these are related or complementary products; on a B2B site, the case studies and documentation for the visitor's industry.

The question to settle first is which data it uses. Under the Italian data protection authority's cookie guidelines (2021), profiling cookies and other tracking tools used to tailor the service to a user's preferences, beyond what is strictly needed to provide it, require prior consent. If the visitor declines, personalisation based on their behaviour is off the table.

That's why it pays to separate two levels:

Contextual recommendations: based on the content of the page being viewed (same category, same attributes, same industry), with no need to follow the person around.
Behavioural recommendations: based on browsing or purchase history, and usable only for visitors who have consented or for registered customers, within the limits of your privacy notice.

Either way, the effect on sales and leads has to be measured with a test, not taken for granted. In B2B, personalising by account or industry ties in with account-based marketing.

How much does AI in WordPress cost?

It depends on how much you use it, because almost every cost is variable. There are five lines to budget for:

Plugins and licences: the official AI plugin is free; many third-party plugins only offer AI features on paid plans or with monthly credits.
Model usage: every request is billed by the AI provider according to use; experiments that run on their own, such as comment moderation, generate requests even when nobody clicks anything.
Review: the hours spent checking copy, translations and data before publication.
Set-up: rules, glossaries, prompts, staging tests and, where needed, integration with your ERP or CRM.
Maintenance: plugin updates, security checks and regular quality reviews.

The right way to assess them is against the hours saved and the errors avoided, using the right KPIs for AI value.

Is AI in your CMS good or bad for SEO?

It depends on how you use it. Google says it rewards content quality, not how content is produced: AI-assisted copy can rank well if it's helpful and original, but AI gives no advantage in itself. So there's no AI that "optimises keywords and structure on its own" and lifts rankings without human input.

The line is drawn in Google's spam policies: scaled content abuse means publishing many pages to manipulate rankings rather than help users, generative tools included. For an online store, that rules out hundreds of auto-written category-by-city pages; for a B2B company, generic articles with no data or point of view.

Where AI genuinely helps SEO is maintenance: missing meta descriptions and alt text, consistent slugs, tidy taxonomies, notes on readability and heading structure. Topic ideas suggested by a model, on the other hand, need checking against real search data, such as Search Console's. For how search itself is changing, read the piece on AI-powered search engines.

Do you have to disclose that content was written with AI?

It depends on the type of content. There are three rules to know:

Chatbots. Article 50 of the EU AI Act, applicable from 2 August 2026, requires telling people when they're interacting with an AI system, unless it's obvious: a chatbot has to introduce itself as one.
Public-interest text. The same article requires disclosing that text has been generated or manipulated with AI when it's published to inform the public on matters of public interest. The obligation doesn't apply if the text has gone through human review or editorial control and a person or company holds editorial responsibility for it.
SEO. Google suggests adding an AI or automation disclosure where readers might reasonably wonder how the content was created, and advises against listing AI as the author.

For a company blog, the safest approach is documented human review and a real, recognisable author on every article.

What are the limits and risks of AI in a CMS?

The risks are real, and nearly all are manageable if you know them before switching a plugin on.

Privacy. Every request to an external model sends text to a provider. If customer data, visitor comments or confidential price lists end up in a prompt, you need a legal basis, a contract with the provider and an updated privacy notice under the GDPR.
Prompt injection. OWASP defines prompt injection as input that alters a model's behaviour in unintended ways, including input coming from web pages or files. Give AI tools minimum permissions and require human approval for actions that change the site.
Vendor dependence. If prompts, rules and workflows are tied to a single model or platform, switching is expensive. In WordPress, connectors let you choose and replace the provider; on a SaaS platform you take the AI it builds in.
Less human control. The more AI publishes without intermediate steps, the higher the risk of errors in data, tone and commercial promises: features that apply changes without a preview should be restricted.
Plugin sprawl. Every plugin adds code, updates and potential vulnerabilities; more in the article on AI and cybersecurity.
Flat content. Without your own data, real cases and a recognisable voice, AI writes correct copy that sounds like your competitors'.

For the rules that govern AI inside a company (who decides, who checks, how choices are explained), see the article on trustworthy AI and governance.

When should you move to a different stack?

When the problem is no longer writing content but getting the website to talk to the rest of the business. WordPress with a few AI plugins is enough for a blog, a showcase catalogue or a store with simple processes. These are the signs it's time for a different architecture:

AI needs to read or write data in your ERP, PIM or CRM;
the same content has to feed the website, marketplaces, apps and sales documents;
AI plugins overlap or slow the site down;
you need workflows with approvals, logs and error recovery, for example on prices and stock.

The usual answer is to leave content management to the CMS and move the AI logic into an automation layer (tools such as n8n or Make, backed by a database like Supabase), or to switch to a headless CMS with a custom front end. The options are covered in the article on business process automation; how I design them is in the solutions section.

How to apply it in your company: first steps

Take inventory. List the content you produce each month and the hours it takes today: that's your baseline.
Pick one use case. Start with the most repetitive, lowest-risk task, such as alt text and meta descriptions.
Assign roles. Who configures plugins and keys, who writes the rules, who approves before publication.
Write the rules before the prompt. Glossary, tone, data that never leaves the site, what AI may not change.
Test on staging, then on a sample. Run it for 30 days on fifty product pages or ten service pages, tracking time saved, errors corrected and request costs.
Decide whether to scale or change architecture. If the bottleneck is data in other systems, look at an automation layer.

Mistakes to avoid

Publishing generated copy without review, especially on prices, technical specs and terms of sale.
Enabling features that run on their own on the live site without checking what they send to the provider and how many requests they generate.
Launching a chatbot with no limits on what it can say and no hand-off to a person.
Using behaviour-based recommendations without consent to profiling cookies.
Installing an AI plugin for every feature, with API keys scattered around and no cost control.

Frequently asked questions (FAQ)

What is a CMS, in simple terms?

A CMS is the software you use to manage a website's content without writing code: pages, articles, images and products. The most widely used are WordPress, Shopify and Wix.

Does WordPress have built-in AI?

Yes. Since WordPress 7.0 (May 2026), core includes an AI client and a screen for managing model providers. Ready-made features such as titles, excerpts, translation and alt text come from the official AI plugin, which you install separately and which is still largely experimental.

Is AI in WordPress free?

The official AI plugin is free, but to use it you need an account and an API key from a provider such as Anthropic, Google or OpenAI, which bills every request according to use. Many third-party plugins only offer AI features on paid plans.

Does AI-written content hurt SEO?

Not if it's helpful and original: Google assesses quality, not how the text was produced. The problem starts when you publish lots of automated pages with no value for users, which Google treats as spam.

Can I use an AI chatbot on my company website?

Yes, but it must disclose that it's an AI system, as Article 50 of the EU AI Act requires from 2 August 2026. Keep it to defined topics, provide a hand-off to a person and handle data under the GDPR.

Is AI in a CMS worth it for small and mid-sized companies?

Yes, especially for metadata, translations and first drafts, where costs are low and time savings are measurable. The limit isn't company size but data quality and the time set aside for review.

When is WordPress no longer enough?

When AI has to work with data outside the website, such as your ERP, PIM or CRM, or when the same content feeds several channels. The CMS stays for content and the logic moves to an automation layer or a headless architecture.

Sources

If you want to know which AI features make sense in your CMS and where you need a different architecture, book a free assessment.