Does merging overlapping pages always help SEO?

Not in our data. Across the sites we run PageSeeds on, merges that were re-measured about 30 days later came out 12 worse to 1 better. The redirected pages lost their rankings and the pages we kept did not pick them up.

What did you change after the merges lost traffic?

On 6 October 2026 we stopped the agent from merging or redirecting pages by itself. It now lists merge candidates for a person to decide. We also set a quota of two to three new articles a month, because new publishing had stalled.

Which SEO changes are safe for an agent to make without asking?

Changes that are cheap to undo, such as a title, a paragraph or an internal link. A redirect removes a page that was ranking, and reverting it does not bring the ranking straight back, so that one should wait for a person.

Our Automated Page Merges Lost Traffic 12 of 13 Times

I run PageSeeds on four of my own sites. It's an agent that reads Search Console each week, picks a few changes, and sends them to the repo as a pull request.

One of the things it did by itself was merge pages. If two posts competed for the same search, it kept one, folded the other into it, and redirected the old URL.

About 30 days after every change, it re-measures the page against Search Console and scores the result: improved, neutral or regressed.

For merges, the score across my sites came back 12 regressed to 1 improved.

So on 6 October I turned that part off.

The numbers

These are clicks from Google, last 30 days against the 30 days before, as of 8 October:

| Site | Prior 30 days | Last 30 days | Change | |---|---|---|---| | Expense Sorted | 2,303 | 1,337 | -42% | | Days to Expiry | 981 | 890 | -9% | | BrewedLate | 354 | 335 | -5% | | Learned Late | 15 | 14 | flat |

All four are down. Expense Sorted is the one that hurts.

On that site, the pages that got redirected had been bringing in roughly 200 clicks a week. After the merge, those clicks went away. The pages we kept didn't pick them up.

Two hundred a week is in the same range as the whole drop. I can't prove the merges explain all of it, and I'm not going to pretend I can. But it's the biggest single cause I can point at.

The live numbers, including this decline, are on the results page.

It followed the advice

This is the part I find uncomfortable.

Merging competing pages is standard SEO advice. We wrote a guide on fixing keyword cannibalization that says the same thing: prove two URLs rank for the same query, pick a keeper, redirect the other.

The agent had rules. It needed hard evidence that both pages ranked for the same search, and a minimum length for the page it kept. For high-traffic pages there was a confirmation step, and I'd told the weekly run to answer that for me and log the call.

It still lost 12 times out of 13.

I don't fully know why yet. My two guesses:

The redirected page ranked for a lot of small searches the keeper never covered, and a redirect doesn't carry those over.
"Same query" was too loose a test. Two pages can show up for one search and still be answering different questions.

Both are guesses. I haven't tested either.

The second problem was quieter

While I was looking, I found something else. The sites had almost stopped publishing.

In August I added a planning file that tells the agent what kind of work to do each week: write new articles, improve existing ones, or work on tools. The mode sat on "improve" for weeks. New articles went out about once a month at best.

So the agent was busy every week, rearranging what already existed, and adding nothing.

What changed

Two things, both on 6 October:

The agent no longer merges or redirects anything. It lists the candidates under "needs your decision" and I choose. If I want a merge, I run it by hand.
Two to three new articles a month, as a quota. The weekly run works through it on alternating weeks.

Everything else it still does by itself: title fixes, content fixes, internal links, new articles.

What I'd take from this if you're building an agent

The mistake wasn't letting an agent change the site. It was treating every change as the same kind of risk.

A new title is cheap to undo. If it's worse, you change it back and you've lost a few weeks.

A redirect removes a page that was ranking. You can revert the commit, but the ranking doesn't come straight back.

The agent had a limit on how many things it could do in a week. It had no idea which of those things could be undone. That's the line I should have drawn first: cheap and reversible, go ahead. Hard to reverse, ask.

The other thing is the re-measurement. Without the 30-day check I'd have seen traffic fall and had no idea which of dozens of changes did it. With it, the answer was one line: 12 to 1.

What I don't know yet

Whether traffic comes back. Whether undoing the redirects helps or just churns the pages again. Whether the one merge that worked had something the other twelve didn't.

I'll post the next set of numbers in a month, whichever way they go.

If you want to see what the weekly run looks like on your own site, the read-only reports are free: install the CLI.