14 of 20 network sites deindexed in one week — the footprint I think did it

Post-mortem, written mostly so I stop going over it in my head. Twenty sites, built between 2021 and 2023, aged domains, all pointing at two properties. Between the 6th and the 13th of May, fourteen of them went from indexed to nothing. Not ranking drops — gone from the index entirely, no manual action in the console on any of them.

I've spent a fortnight comparing the fourteen against the six survivors. Here's what I found, in the order I'd fix it.

The nameserver pair

Eleven of the fourteen sat on the same two nameservers. Not the same host — I'd been careful about hosts — but I'd routed everything through one DNS provider under one account because it was convenient to manage. All six survivors were on registrar defaults.

This is the one I'd fix first if I were doing it again, and it's embarrassing because it's on every checklist written since about 2016 and I still did it.

The build window

Thirteen of the fourteen were built inside a nine-week stretch in early 2022. Same theme family, same plugin set, same seven-page structure, and — this is the part I hadn't thought about — the same about-page voice, because I wrote them all myself in the same fortnight and I have habits.

The six that survived were the ones I built later, slower, when I'd got bored of the template.

The outbound pattern

Every one of the fourteen linked out to both money sites. Four of the six survivors linked to only one, and two linked to neither yet — they were still warming.

I don't have a way to separate this from the build-window finding, because the same sites are in both buckets. Flagging that rather than pretending the analysis is clean.

What I don't think it was

Content. It's the first thing everyone says and it doesn't fit: the fourteen and the six had content from the same two writers, at the same quality, and if anything the survivors are thinner.

Hosting. Six providers across twenty sites, and the deindexed fourteen were spread across all six.

What I'm doing

Rebuilding six of them on separate DNS, six months apart, single outbound each. Writing off eight. Rough cost of the lesson is a bit over four thousand and about two years.

Summary: my infrastructure separation was good and my behavioural separation was terrible, and I'd assumed the first one was the hard part.

Good write-up and I think the conclusion is wrong, or at least unsupported.

What happened in the first half of May is that an entire tier went. I know four people who lost sites in that window and not one of them shared a nameserver with you. March took the obvious stuff and then there was a second sweep, and if you were a small site whose reason for existing was to point at something, the odds were bad regardless of your DNS hygiene.

You've got twenty data points, no control, and a fortnight of looking for a pattern with the outcome already known. You will always find one. I've written this exact post myself — mine was in 2017 and I was certain it was the WHOIS, and the guy next to me was certain it was his footer.

The six that survived are interesting. The fourteen that didn't are a coin that came up tails.

The nameserver finding matches what I've seen and I'd fix it whether or not it caused this.

One account, one API key, twenty zones is a single object that describes your whole network, and you don't get to know who reads it. Same reason people get burned by putting everything behind one analytics property.

FWIW the fix is cheap and boring: registrar defaults, different registrars, and never touch DNS from the same session.

I take the point about looking for a pattern with the answer known — that's exactly what I did and I should have said so in the post.

But "a tier went" isn't a competing explanation, it's a description. Something selected fourteen of my twenty and left six, and the six are not a random six: they're the later, slower, less-templated ones. If the sweep were indifferent to my sites' properties I'd expect the survivors to look like the casualties, and they don't.

That's not proof. It is more than a coin.

The later ones were also the ones with the least link equity pointed at them and the shortest history in the index, which on a sweep aimed at established link-tier sites would produce precisely the survivor set you've got. Two explanations, same data, n=20.

We're not going to resolve this. Rebuild them the way you said either way — it costs you nothing if I'm right and it saves you if you are.

Three buyers wrote to me in the same fortnight about domains going dark, all of them small networks, none of them connected to each other.

Doesn't settle your argument, adds a bit of weight to old_growth's side of it. Whatever it was, it wasn't only you.

Did the six survivors have any traffic of their own? Sorry if that's obvious, it's the only difference I can think of that nobody's mentioned.