in

AI in my WordPress work: what actually helps

Illustration of an AI drafting a reply and a person approving it

At WordCamp Rajasthan, almost every hallway conversation ended up on AI. People were using it to write code faster, plan projects and create content. What I liked was that nobody talked about replacing people. They talked about getting more done.

I’ve written a few AI plugins for WordPress over the last two years. Here’s where I’ve found AI really helps, and where I keep it away from decisions.

The rule I build with: the AI suggests, a human decides

My AI Comment Moderator catches spam and abuse before anyone sees it, and drafts replies to good comments. It doesn’t publish those replies. You approve them. AI Agent for Website answers visitors from your own content, not from whatever the model happens to know.

Both follow the same idea. The AI does the repetitive part, and the decision stays with the site owner. If your plugin lets an AI publish, delete or email on its own, think again.

Don’t marry one provider

Models change every few months, and so do prices. I wrote ai-engine, a small PHP library, so my plugins don’t care whether they talk to Gemini, Llama or Groq:

use AIEngine\AIEngine;

$ai = AIEngine::create( 'groq', $api_key );
$summary = $ai->generateContent( 'Summarise this support ticket in two lines: ' . $ticket );

Changing provider is one line. Put a layer like this between your plugin and any AI API. You’ll thank yourself later.

Where it helps me every day

  • Reading unfamiliar code. “What does this hook chain actually do?” is a great question for an AI, as long as you check the answer against the source.
  • Boilerplate. Settings pages, REST routes, block.json files. Boring code I’d type the same way anyway.
  • First drafts of docs and readme files. I always rewrite them. A draft is still faster than a blank page.
  • Tests. Ask for edge cases you didn’t think of. Half of them will be useless, and one will find a bug.

Where I don’t use it

  • Security decisions. Sanitising, escaping and capability checks. I write these myself and review them twice.
  • Publishing content without reading it. I tried AI-written blog posts on this site. They read like everyone else’s, and I’ve rewritten or removed them.
  • Anything with personal data, unless the user has clearly agreed to it being sent to a third party.

That second point taught me the most. A few posts on this blog were written mostly by AI, and you could tell. They were correct, polished and forgettable. What people come back for is the part only you can write: what happened, what went wrong, what you’d do differently.

Use AI to get to the interesting part faster. Then write the interesting part yourself.


Rajan's Assistant
Rajan's Assistant
Online now

Start a conversation

Please introduce yourself so we can assist you better.

How was your experience?

Please rate your conversation

Schedule a Meeting?

Connecting to Agent...

Please wait while we connect you to a live agent...

Connected!

You are now connected with a live agent.

Agents Unavailable

Our agents are currently offline. Please try again later or leave a message.

AI Agent for Website, an open-source plugin by Rajan