What Happens When You Let an AI Run Your Blog’s Back Office?

I have been blogging here since 2009. Seventeen years, more than 120 posts, two chronic illnesses, a lot of running — and, if I am honest, years of quiet neglect in the back office. Posts went up. Maintenance did not. Sound familiar?

This fall I tried something different. I gave an AI assistant — Muse, which I call Archer — an Editor-level login to this WordPress site and put it to work on the jobs I had been avoiding. Not writing my posts. Administering them. There is a difference, and the difference is the whole point of this article.

What can an AI actually administer on a WordPress blog?

More than I expected. Working through a browser, the same way I would, Archer can:

• Sign in to WordPress (as an Editor, not an admin — that limit is deliberate) and edit posts and pages
• Read and update SEO fields — titles and meta descriptions in SEOPress
• Fix image alt text, titles, and captions
• Read Google Search Console directly: clicks, impressions, positions, and which pages Google has and has not indexed
• Submit indexing requests for pages Google has missed
• Build the working documents the job needs — audit spreadsheets, follow-up sheets, link inventories in Google Docs and Sheets
• Run on a schedule, checking things daily and reporting back only when something matters

It does not get my admin password, my publishing rights, or the final say on anything. We will come back to that, because it matters.

So what did the audit actually find?

Numbers first. I always trust numbers more than vibes.

Archer audited all 122 posts. Here is what 17 years of good intentions looked like:

• 103 of 122 posts had zero internal links pointing to them from anywhere else on the site
• 109 posts had not been modified since before 2024
• 51 posts were under 600 words
• Over three months, 96% of my Google clicks — 409 of 425 — came from a single post

One post was carrying the entire site’s search traffic, and a hundred others were effectively invisible because nothing linked to them. I wrote them. Google just had no reason to care about them.

Can an AI fix years of SEO neglect?

Not in a weekend, and anyone who promises you that is selling something. What it can do is the slow, boring, measurable work, at a scale one person will never get around to:

A pilot before a rollout. Before changing anything site-wide, we picked three posts, improved their titles, meta descriptions, and internal links, and agreed to judge the results after 28 days of Search Console data. Test small, measure, then decide. That is not an AI habit — that is just good process, and the AI followed it.

A full copy edit. Archer reviewed all 124 posts and 26 pages, line by line, and applied 913 fixes — typos, Canadian spelling, punctuation — automatically. Anything that needed a human decision went onto a follow-up sheet instead: 847 rows. A second pass, where I approved whole categories of safe fixes, cleared 217 of those rows. The remaining 630 are mine to judge, one coffee at a time. The AI did the finding and the fixing. The judgment stayed with me.

Dead links, resurrected. An old plugin had died somewhere along the way, leaving 28 Amazon product shortcodes across 7 posts and a page rendering as blank space — readers saw nothing where the product boxes used to be. Archer found every one, identified what each product was, and rebuilt them with working links. The links themselves had to come from my own Amazon account (only I can generate those), so the workflow was: it showed me the exact change, I supplied the link, it applied and verified it live. Every fixed page also got its proper disclosure line — 21 published items now carry “As an Amazon Associate I earn from qualifying purchases.”

Consistency I never got around to. Every post where I write about living with diabetes or Crohn’s disease now ends with the same disclaimer — my own words, about my own lived experience, pointing readers to their own healthcare team. That is 32 posts, updated in a day. By hand, that job would still be on my someday list.

The indexing triage. Search Console said 3,130 URLs on my site were not indexed, which sounds like a catastrophe until someone reads all of them. Archer did. Roughly 3,050 turned out to be junk — plugin files, feeds, tag archives, share-button variants — that should never be indexed at all. The real number was 74 of my own posts and pages missing from Google, including my 2,500-word overview of diabetes and Crohn’s disease. Those are now triaged into priority lists and the first indexing requests are in. What most of them needed even more than a request, though, was a link — 14 of the 20 highest-priority pages had no links pointing to them from any indexed page at all.

Posts that point to each other. A blog is not 124 separate articles — or it should not be. Archer rebuilt the site’s entire internal link structure from all 150 published posts and pages, so we could see, post by post, exactly what links to what. Then it started weaving. “Related reading” lines now sit at the end of related posts, so a reader who finishes one piece has somewhere natural to go next — and Google has a path to follow. My Libre questions-and-answers post points to my World Diabetes Day post. The sensor falling-off post is linked in from its follow-ups. And every one of the 20 priority pages above now carries at least one link from a page Google already indexes — for 14 of them, the first such link they have ever had. Cross-post integration sounds like jargon until you watch a decade of orphaned posts get adopted.

How much time does this actually save?

The honest answer is that the saving is not in any single task. It is in the tasks that were never going to happen. I was never going to read 150 pages looking for typos. I was never going to export and classify 3,130 URLs from Search Console by hand. I was never going to rebuild my site’s entire internal link structure in a spreadsheet to see which pages were orphaned. Those jobs took the AI hours of background work and me minutes of review — approving a batch here, supplying a link there, answering a question when it found something odd.

What is it not allowed to do?

This part I set up before any of the work started, and I would not run it any other way:

• Nothing publishes without my explicit approval. Drafts and edits I have approved, yes. The publish button is mine.
• It works as an Editor, never as admin. Some jobs it simply cannot do — my privacy policy page and a typo in my own author bio (“ilnesses” — still there) need an admin, so they sit on a list for me. That is a feature, not a bug.
• Passwords never travel through chat. Logins go through a secure vault, and my email password gets used once, for one job, when I say so.
• It verifies, and so do I. It once reported a typo on my About page that turned out to live in dormant code that never displays — the live page was fine. I checked, we laughed, it cleaned up the code anyway. Trust, but verify, in both directions.

In summary

An AI agent will not care about your blog the way you do, and it should not write in your voice on the topics that are yours — my illness, my family, my finish lines. What it can be is the editor and administrator I never had time to hire: tireless at audits, patient with spreadsheets, honest about what it finds, and fast at the repairs once I have made the decisions. Seventeen years of neglect is not undone. But for the first time, it is itemized, prioritized, and shrinking.

If you run a blog that has outgrown your evenings, it is worth trying. I use Muse — the assistant this article calls Archer. If you would like to try it yourself, you can join here and enter my invite code CTADX7 in Settings within 48 hours of joining — we will both get 1 billion Muse tokens. Fair warning: you may end up, like me, with a spreadsheet of 630 small decisions waiting for you. But at least now I know exactly what they are.