Use AI to edit your WordPress posts (the big-picture workflow)
There’s a strange mismatch at the heart of editing WordPress with AI. The models are very good at exactly the work you want done — tightening intros, rewriting metadata, normalizing headings, refreshing dated facts across hundreds of posts. The CMS is a hosted admin panel that none of them can actually open. So you end up doing one of two unsatisfying things: copy and paste between wp-admin and a chat window all afternoon, or install yet another plugin that auto-generates content with no preview, no diff, and no way back.
This guide is the strategic version of the picture — why the workflow looks the way it does, and what to insist on when you pick tools. If you want the hands-on step-by-step, here is the practical guide to editing with Claude, ChatGPT, or Gemini.
The actual problem
WordPress, even hosted on your own server, behaves like a hosted app once you’re inside it. Posts live in a database. The Block Editor saves serialized blocks back through the admin. The REST API exists, but it’s a network surface, not a folder. None of your AI tools can browse it, search across posts, or edit five hundred of them in a single pass.
Meanwhile, every AI you’d actually want to use — Claude, ChatGPT, Gemini, Copilot, whatever’s next — is excellent at one thing: reading and writing text files. Hand it a directory of .md files and it can sweep across them, follow a careful instruction, and produce a diff. That’s the gap. The content lives in a system the AI can’t reach, in a format the AI doesn’t naturally edit.
The bridge: a folder of markdown
The fix is unglamorous and durable: take the part of WordPress the AI can use — the posts themselves — and mirror it as plain markdown files on your Mac. Frontmatter at the top of each file holds the metadata: title, slug, status, categories, tags, featured-image URL, Yoast or Rank Math title and description. Body below in markdown.
Now the AI has something it can work with. So can a five-line Python script. So can a search-and-replace. So can git diff. Your content is text, the way text was supposed to be.
Specter is what we make to do that round trip — a native macOS app that does two-way sync between WordPress and a local folder. Two-way is the important word. Most “AI for WordPress” tools are publish-only — they can create new drafts but can’t touch posts that are already live. That makes them useless for the work that actually matters, which is improving the URLs that already rank.
If you haven’t already, the connect guide walks through Application Passwords and getting your first sync running.
Why this beats one-shot AI plugins
The “AI for WordPress” plugin shelf is full of one-shot tools: paste a prompt, get a draft, publish. They’re seductive because they’re fast. They’re also dangerous because there’s no review step that matches the size of the edit. A plugin that rewrites one post is fine. A plugin that rewrites a hundred has no way to show you what it’s about to do.
A folder-and-sync workflow inverts that. The edit happens locally, where you can read the diff in any editor you like. The push to WordPress is a separate step with a preview. You see exactly which posts will change before any byte leaves your Mac.
It also avoids model lock-in. You aren’t tied to whatever model the plugin author wrapped six months ago. Use Claude today, swap to Gemini tomorrow, run a deterministic script the day after. The folder doesn’t care.
What this unlocks
Once your posts are a folder of markdown, the jobs that used to be impossible become routine:
- Refresh old posts. Filter by published date in frontmatter, hand the AI everything older than 18 months, refresh facts and the date. The old-post refresh guide walks through it.
- Bulk SEO sweeps. Generate missing meta descriptions, rewrite SEO titles, standardize categories. The bulk SEO guide covers concrete jobs.
- Voice and tone passes. Run a consistent edit across an archive that grew over years and a dozen contributors.
- Internal linking. Have an AI scan each post for topics you’ve covered elsewhere and weave in links.
- Fact-checking. Use a model with web access to flag claims that look stale across the archive.
None of these require a special plugin. They require your posts to be in a format the AI can read.
The honest limitations
Worth being upfront. AI editing is not magic and the workflow has edges:
- You still have to review the diff. Skip this and you’ll ship slop. The point of editing in a folder is that reviewing is easy, not that it’s optional.
- Page-builder blocks (Elementor, Divi) round-trip as opaque HTML. Standard Gutenberg blocks are clean markdown. If your site is built on a heavy page builder, AI edits to those sections are best avoided. See how Specter handles the Block Editor for the detail.
- Specter doesn’t bundle AI. It’s a bridge. You bring Claude, ChatGPT, or whichever model you already trust. That’s a feature — no lock-in, no per-token markup — but it does mean the intelligence is on you.
The mental model
WordPress stays your publishing home. Your Mac holds the source of truth as markdown. The AI tool you already pay for does the writing work in the folder, where text-on-text is the natural case. The sync layer keeps the two ends honest so what you read in the file is what lands on the site.
The whole reason to bother with this setup is leverage without losing control. You get the speed of a script and the reach of an AI while keeping the review step a human-sized one.