What an AI SEO agent actually changed on 6 WordPress sites in its first week

Most SEO tools tell you what is wrong. Very few of them tell you what they actually did about it. So here is the plain record of TinyRank's first week doing real work on real client sites.

Between 28 October and 2 November 2026 the agent ran on 6 WordPress sites. All of them belong to clients of my own agency, so I could watch every change and check it on the live site afterwards. In those six days it made 105 fixes. Every one of them is logged with the old value and the new value, and most of them can be undone with one click.

Below is what those 105 fixes were, in order of how often they came up.

The 105 fixes, by type

• Insecure http links on https pages: 34. Links inside page content that still pointed at http:// versions of pages, usually left over from a migration years ago. Each one was rewritten to https.
• Duplicate content: 21. Pages or posts that repeated another page's title, description or body closely enough that Google would struggle to pick one. Mostly duplicate titles and descriptions across similar service pages.
• Broken links: 14. Internal links to pages that no longer exist. Each one was pointed at the right live page, or removed if there was nothing sensible to point at.
• Titles too long: 10. Page titles over the length Google shows in results, so the end was being cut off. Rewritten shorter, keeping the words that matter.
• Missing image alt text: 10. Photos with no description. The agent wrote a short description of what is in each one.
• Thin content: 8. Service or location pages with a heading and one or two sentences. Flagged and worked on with the agency's approval, because rewriting a page is not a small change.
• Missing meta description: 3. Pages with no description at all, so Google was inventing one. Written from what the page is actually about.

That adds up to 100. The remaining 5 were small one-offs: a stray noindex on a page that should be indexed, a heading structure with several H1s, and similar.

What surprised me

I expected missing alt text and meta descriptions to be the biggest group. They were not. The single biggest problem across these six sites was old http links sitting inside page content.

This makes sense when you think about how small business sites get built. Someone sets up WordPress on http, writes 40 pages, then a year later the host turns on SSL and adds a redirect. The redirect makes the site look fine in a browser. But every internal link in the content still says http, so every click goes through a redirect, and Google's crawler sees a site that cannot quite make up its mind. Nobody goes back and fixes 34 links by hand. It is boring, and it never reaches the top of anyone's list.

An agent does not find it boring. It found all 34 on the first audit and fixed them over the first few days, 7 changes a day per site.

What it did not do

I think this part matters as much as the fixes.

It did not delete anything. Not a page, not a post, not an image. When it found the 21 duplicates, it did not remove one copy. It rewrote titles and descriptions so each page had its own, and where two pages were genuinely the same it flagged that for a person to decide.

It did not rewrite the 8 thin pages on its own. Those went to the dashboard as recommendations with a draft, and the agency decides whether to publish. Writing a new service page for a client is a judgement call about the business, not a technical fix.

It did not touch users, passwords, theme code or plugin code. It is not allowed to.

The daily rhythm

Each site gets audited once, then re-audited every 30 days. Every morning at 9am in the site's own time zone, the agent picks the most valuable open problems and fixes up to 7 of them. I wrote about why we chose 7 separately, but the short version is that 7 is enough to clear a typical backlog in a few weeks and small enough that an agency owner can read the whole day's log over a coffee.

So across six sites and six days, 105 fixes works out to a little under 3 per site per day. Some days a site hit 7. Some days it had nothing left in its queue and did nothing, which is the right answer.

How I checked

For each site I opened the change log and picked changes at random, then opened the live page. The old value in the log matched what Google's cache had shown. The new value matched what was live. On two sites I undid a change from the dashboard to make sure undo worked, then let the agent redo it the next day.

I did not see a single change that broke a page. One alt text description was more generic than I would have written. I left it, because a plain description beats an empty one.

What this does and does not prove

It proves that an agent can make a hundred small, correct, reversible fixes on WordPress sites in a week without anyone sitting over it, and that the record of those fixes is clear enough for an agency to show its client.

It does not prove anything about rankings or traffic yet. Six days is far too short to see that, and I am not going to make that claim until I can show it. What I can show today is the work itself, which is what most agencies were paying someone to do by hand and mostly not getting.

If you run an agency and want to see this on one of your own client sites, TinyRank is $99 per site per month with no setup fee and no contract. Add one site, read the log at the end of the week, and decide from there.