You Don’t Need a Fancy AI Setup to Stay Current With WordPress

Most AI tools for WordPress try to replace the developer.

I built one that requires the developer.

Here’s why, and how to make it your own.

WP Trend Watcher graphic with magnifying glass and graph

The Problem

WordPress moves fast. Between Make Core posts, Developer Blog updates, ACF releases, and ecosystem news, staying current is a part-time job.

I’m a freelance WordPress developer. I want to know what’s changing. I do not want to read every post.

AI summaries exist. But most AI WordPress tools have two problems:

  1. They try to automate opinions. I want the summary, not the hot take.
  2. They obscure their costs. I want to know what the thing actually costs to run.

So I built something different.

What WP Trend Watcher Actually Does

It’s a small TypeScript project. Two commands:

pnpm collect     # pulls recent articles from WordPress sources
pnpm summarize   # generates a draft weekly report using a local AI model

That’s it. No dashboard. No automation. No publishing pipeline.

The output is a Markdown file. I read it, check the sources, add my own notes, and decide what matters.

The AI does the collection and summarization. I own the conclusions.

Who This Is For

Freelance and agency WordPress developers who want to stay current without spending hours reading every source.

If you run a plugin company, contribute to core, or need deep technical analysis of every commit, this probably isn’t enough. But if you want a weekly pulse check on the WordPress ecosystem, it works.

Setup in 5 Minutes

You need Node.js and pnpm. That’s it for collection. For summarization, you need Ollama, which runs a local AI model on your machine.

# Clone and install
git clone https://github.com/colorful-tones/wp-trend-watcher.git
cd wp-trend-watcher
pnpm install

# Pull a local model (one time, ~2GB download)
ollama pull llama3.2:3b

# Copy the env file (defaults work out of the box)
cp .env.example .env

# Collect articles from the last 7 days
pnpm collect -- --days 7

# Generate a draft report
pnpm summarize

After pnpm summarize, you’ll find a draft report at reports/YYYY-MM-DD.md.

The Report Format

Every report has four sections:

Weekly Summary — what happened this week across the tracked sources.

Emerging Trends — topics showing up across multiple sources.

Developer Implications — why any of this matters for your actual work.

What I’m Watching — my personal observations. This is the human-authored section. It’s where I add what I think is interesting, uncertain, or worth following.

There’s also a Build Notes section that lists exactly what the run cost, how many articles were analyzed, and how long review took.

Making It Your Own

The defaults are reasonable. The real value is customizing it for what YOU care about.

Add Your Own Sources

Edit src/sources.ts. Add any WordPress-related site with an RSS feed:

{
  id: "my-plugin-blog",
  name: "My Plugin Blog",
  url: "https://myplugin.com/blog/",
  feedUrl: "https://myplugin.com/blog/feed/",
  tier: 1,
}

Then run pnpm collect again. The new source appears in your reports.

Swap the AI Model

The default is llama3.2:3b, a ~2GB local model. It’s good enough for per-article summaries but sometimes misses articles in the synthesis step.

For better coverage, use a larger model:

# ~5GB, covers 5-7 of 7 articles consistently
WP_TREND_OLLAMA_MODEL=llama3.1:8b pnpm summarize

# ~9GB, covers all 7 with better nuance
WP_TREND_OLLAMA_MODEL=qwen3:14b pnpm summarize

All local. All $0 per run.

When cloud providers are added, you’ll be able to swap to GPT-4o mini for about $0.01 per report. The architecture already supports it.

Change How Far Back You Collect

Default is 7 days. For a monthly recap or a specific event week:

pnpm collect -- --days 14
pnpm collect -- --days 30

What It Costs

The first report I generated:

  • 7 articles analyzed
  • 4 sources reviewed
  • Local model: llama3.2:3b
  • Tokens: ~6,400 prompt + ~1,100 completion
  • Cloud cost: $0.00
  • Human review time: ~15 minutes

If you run this weekly, your monthly cost is somewhere between $0 and the electricity to keep your laptop on.

If you eventually use a cloud model like GPT-4o mini, expect about $0.04/month for weekly reports. That’s the high end.

I track this because most AI projects hide their real operating cost. I want the opposite.

Why Human Review Matters

The AI draft is a starting point. It misses things. It sometimes connects dots that shouldn’t be connected. It writes in AI-voice unless you prompt it carefully.

The review step is where the report becomes useful:

  1. Read the summaries. Do they match the source articles?
  2. Check the trends. Are they real or artifacts of the prompt?
  3. Add your observations. What do YOU think is important?
  4. Remove anything weak, vague, or unsupported.

This takes about 15 minutes. Less time than reading all the source articles yourself.

What’s Next

Phase 1 is about proving the workflow. One human-reviewed weekly report, shared publicly.

If the format proves useful, later phases might add:

  • Custom source configuration via a YAML file
  • Better report formatting
  • Monthly summaries
  • Historical trend comparisons

None of this is built yet. The project ships small, proves itself, then grows.

Try It

The repo is public. The code is a few hundred lines of TypeScript. The prompts are readable strings, not hidden behind a service.

If you try it and something breaks, open an issue. If you add a source that works well, send a PR. If you hate the report format, tell me.

This is a workflow, not a product. It works for me. Maybe it works for you too.

Repo: https://github.com/colorful-tones/wp-trend-watcher

First report: reports/2026-06-12.md