How to edit WordPress posts with Claude, ChatGPT, or Gemini

Claude, ChatGPT, and Gemini are very good at rewriting prose, fixing copy, tightening intros, and grinding through the kind of edits nobody enjoys doing by hand. The problem is they can’t actually reach your WordPress site. There’s no plugin that drops them into the Block Editor, and wp-admin is a hosted panel behind a login — not something they can open. They work on the text and files you give them.

So the fix is the obvious one: bring WordPress to where the AI already works. Sync your posts down to a folder of plain markdown, edit them with Claude, ChatGPT, or Gemini like any other text file, then sync the changes back. That round trip is exactly what Specter handles — a two-way sync between WordPress and a local folder that can edit posts you’ve already published, not only new drafts.

This is the hands-on, tool-specific version. If you want the bigger-picture case for AI plus WordPress, read the overview first.

A note before we start: Specter has no built-in AI. It doesn’t bundle a model or charge you for tokens. You bring Claude, ChatGPT, Gemini, Copilot — whatever you already pay for and already trust. Specter is the bridge; you bring the intelligence.

The workflow, step by step

1. Connect your site. Follow the connect guide — you’ll create an Application Password in wp-admin, paste it into Specter, and pick a local folder.
2. Sync the site down. Specter pulls every post and page as a .md file. Frontmatter at the top preserves the title, slug, status, categories, tags, featured-image URL, and your Yoast or Rank Math meta title and description. The folder can be an Obsidian vault or just a plain directory.
3. Point your AI at the folder. With Claude Code (or the Claude desktop app with file access), give it the folder path and it can read and write your posts directly. Prefer ChatGPT or Gemini? Open a post in any editor and paste it in — or upload the file — then paste the revision back. Either way the AI is now working on your real posts.
4. Give it a specific instruction. Be clear about scope — one post or the whole archive — and about what you do and don’t want touched. Vague prompts produce vague edits.
5. Read the diff. This is the step that keeps your voice intact. Throw out anything that sounds like a robot, keep the lines that earn their place.
6. Run the dry-run preview in Specter. Before anything reaches WordPress, Specter shows you exactly which posts would be created, updated, or flagged as a conflict. After a big AI sweep this is the safety net you’ll be grateful for.
7. Sync back. If the preview looks right, push the edits to WordPress over the REST API. If there’s a conflict — you changed a post in wp-admin and edited it locally — Specter asks you which version wins instead of silently clobbering one.

Prompts that earn their keep

These are the kinds of jobs that are an afternoon by hand and minutes through the folder.

Rewrite the intro paragraph of every post in this folder to lead
with the reader's problem instead of a generic hook. Keep the same
facts, keep my voice, and don't touch anything below the first
heading.

Look through the frontmatter of each .md file. For any post whose
meta_description field is empty, write a ~155-character description
based on the post body. Leave posts that already have one alone.

In this single post, fix passive voice and cut filler words. Show
me the changes — don't rewrite sentences that are already clear.

Start narrow. Run a prompt on one post, check the result, then scale it to the archive once you trust the output. Don’t hand an untested prompt to five hundred posts and hit go.

You review the diff — that’s the whole point

The reason this workflow beats an auto-publishing AI plugin is the human in the loop. Nothing reaches your live site until you’ve looked at it. You read the diff in your editor, you keep the lines you like, you delete the ones that sound generic, and only then do you run the dry-run preview and sync. You’re never one click away from pushing AI slop to readers or to Google.

That matters most for posts that already rank or already get traffic. Most “AI for WordPress” plugins are publish-only — they generate new drafts and can’t safely touch what’s live. Because Specter is two-way, you can hand Claude your three-year-old cornerstone post, improve it, preview the change, and update it in place without breaking the URL, the canonical, or the meta you’ve already invested in.

Where this fits

Editing one post for clarity is the everyday case. The same setup scales straight into bigger jobs — for example, bulk-editing WordPress posts for SEO, where you have an AI generate or rewrite meta descriptions, titles, and internal links across the whole archive in one pass, then preview and sync.

The mental model is worth holding onto: WordPress stays your publishing home, your Mac holds the source of truth as markdown, and the AI tool you already trust does the actual writing work in between. Specter just keeps the two ends honest so the edit you read in your folder is exactly the edit that lands on the site.