Every WordPress plugin directory search for “AI” now returns hundreds of results, and most of them promise the same thing: write your content, optimise your SEO, build your pages. Some of that is genuinely useful. Most of it adds a dependency, an API bill and a new place for something to go wrong.
This is where we have seen AI earn its place on a WordPress site, and where it tends to cost more than it returns.
The short version
AI is good at the work around your content: first drafts, alt text, summaries, tagging, moderation, translation. It is much weaker at the work that decides whether a page performs, which is knowing your customer and saying something true that nobody else is saying.
Use it to remove the boring half of publishing. Do not use it to replace the thinking half.
Where it genuinely helps
Alt text and image housekeeping
Most media libraries are full of images with no alt text, because writing it is tedious and nobody owns the job. Generating a first pass and having an editor correct it is a clear win: it is faster than writing from scratch, and the failure mode is a slightly generic description rather than a broken page.
First drafts and outlines
Turning a rough set of notes into a structured draft saves real time, particularly for the pages nobody wants to write: policy pages, category descriptions, product specifications. The draft is a starting point, not a publishable article. If it goes live unedited, readers notice, and so does your credibility.
Internal linking suggestions
On a site with a few hundred posts, nobody remembers everything that has been published. Tools that suggest related pages from your own content are useful because the suggestions are grounded in something real: your own library. Review them before accepting, because relevance and usefulness are not the same thing.
Comment and form moderation
Classifying incoming messages as spam, support, or sales is a narrow task with a clear answer, which is exactly the kind of work a model does well. It does not need to be clever, only consistent.
Where it usually disappoints
AI page builders produce layouts that look plausible in a demo and generic on a real site. They tend to generate markup the theme was never designed for, which then has to be styled or thrown away. If a design system already exists, generated layouts fight it rather than use it.
Bulk content generation is the other one. Publishing fifty AI-written posts is easy; the difficult part was never the writing. Thin pages that answer a question nobody asked still perform the way thin pages have always performed, and now you have fifty of them to maintain.
SEO score plugins that use AI to grade your writing are worth treating carefully. A score is a proxy, and optimising a proxy usually produces text written for the scorer rather than the reader.
The plugin problem
Every AI feature you add through a plugin brings three things with it: an outbound API call on a request your visitors may be waiting for, an API key stored somewhere in your site, and a subscription that renews whether or not anyone uses the feature.
Before installing one, it is worth asking:
- Does this run in the admin only, or on the front end where it can slow pages down?
- Where is the API key stored, and who on the team can see it?
- What happens to the site if the service is down, or the plan lapses?
- Is the content it produces stored in the database, or regenerated every time?
The features that survive on client sites are almost always the ones that run in the editor, write their result into the database, and leave the front end untouched.
Privacy and what leaves your site
If a plugin sends post content, customer enquiries or form submissions to a third party, that is a processing decision, not just a technical one. It belongs in your privacy policy, and if you handle personal data for clients in Europe or the UK, in your processing records too.
Check whether the provider trains on submitted data by default, and whether the plugin lets you turn that off. Several do; several do not mention it at all.
A setup that tends to work
- Drafting and alt text inside the editor, with a human approving everything before publish.
- One provider, one API key, stored in the site configuration rather than pasted into a settings screen.
- Nothing generated on the front end at request time.
- A note in the privacy policy covering what is sent and to whom.
That is a short list on purpose. The sites where AI has clearly helped are the ones that added two or three narrow tools, not the ones that installed an AI suite.
What we do on client projects
We use it the same way: to speed up the parts of a build that are repetitive, and never to decide what a page should say. If you are considering adding AI features to a WordPress site and want an honest view of whether they will pay for themselves, tell us what you are trying to remove from your week and we will tell you if it is worth automating.
