AI-Assisted SEO Cleanup: From Legacy WordPress URLs to Measurable Pages

Cleaning up an old site is not glamorous. It is not a redesign. It is not a new brand system. It is usually a pile of forgotten WordPress URLs, old category pages, half-migrated posts, and Search Console rows that quietly tell you where attention is still leaking.

That is exactly why it is a good use case for AI-assisted operations. Not because AI magically knows what to publish, but because the cleanup has a lot of repeatable judgment: group old URLs, decide whether intent still matters, map redirects, refresh titles, and keep a simple action log so the site gets better instead of just different.

The data was small, but it was specific

The Search Console report for this site was not showing massive traffic. That was the point. It was showing a few concrete signals that were easy to act on:

Those numbers are not a market thesis. They are not proof of demand for a full content program. They are a cleanup signal: Google still remembered old URLs, people were still searching adjacent terms, and the current static site needed clearer routing and titles.

What I changed first

The first pass was mechanical but important. I mapped legacy WordPress URLs to current static pages, added 301 redirects, kept the sitemap generated from the current posts folder, and made sure the homepage was pointing readers toward active projects and stronger authority pages.

That turns an old migration from a one-time archive dump into a living system. Instead of asking, "Should I rewrite everything?" the better question became, "Which old URL or query has enough evidence to deserve one focused improvement?"

Where AI helped

AI was most useful as an operations assistant, not an autopublisher. The practical help was turning raw rows into buckets: legacy WordPress cleanup, topic opportunities, redirect candidates, title improvements, and follow-up actions for Search Console.

It also helped draft alternatives quickly. For example: how should a title balance "eudaimonia" with the common misspelling "eudomonia" without making the page look spammy? That is the kind of judgment where AI can generate options, but the final call still needs a human who understands the site and the reader.

The same applies to redirects. AI can propose a mapping, but I still want the redirect file to be readable, small, and manually reviewable. A wrong redirect is worse than no redirect because it tells both people and search engines that the wrong page is the answer.

The rule I would reuse

Do not start with a full relaunch. Start with a cleanup loop:

1. Export the Search Console rows that show current impressions.
2. Group them by intent: old URLs, misspellings, topic clusters, and dead categories.
3. Map each legacy URL to the closest useful current page.
4. Refresh titles and descriptions only where the query data suggests confusion or opportunity.
5. Regenerate the sitemap and resubmit it after the changes are live.
6. Re-run the report after Google has time to crawl the updates.

That loop is boring in the best way. It creates a site that compounds instead of a site that gets redesigned every few years and then neglected.

What still needs a second pass

The next work is to watch whether the redirects and refreshed titles change impressions, clicks, and indexing. If the data improves, the same method can be reused for older travel posts, AI video editing posts, and any legacy GTM content that still has search demand.

This is also the kind of practical workflow I want to keep documenting in the AI community: small, real systems where AI helps with the operational load, but the work is grounded in actual data and visible outcomes.