I watched my Google impressions drop 90% in 3 days. I did it on purpose.

Here are four days of Google impressions from late April:

• Apr 23: 150

• Apr 24: 104

• Apr 25: 62

• Apr 26: 12

From 150 to 12 in three days. A 92% drop. And the worst part is I knew it was coming, because I'm the one who caused it, yet I still sat there refreshing Search Console with my stomach in a knot wondering if I'd just nuked my own site.

Why I pulled the trigger

A few days earlier my impressions had been climbing nicely, up into the 150s, which made me feel like my SEO efforts were finally working. Then I actually looked at which queries were driving it.

Most of them were wrong. I'd shipped a set of /operations pages that were really just developer API docs, and Google had decided they were a reasonable answer for generic API-reference and dev-tooling searches. Those people did not want what I'm building. They hit a page that didn't match their intent and bounced. The chart was going up because I was ranking for the wrong audience.

So I de-indexed the whole /operations set. I knew impressions would fall but I did not prepare myself to watch them fall in real time, one day at a time, 150 to 104 to 62 to 12. Knowing a number will drop and watching it drop are very different experiences.

Then six weeks of nothing, slowly

After the crater it just sat there. Single digits to low twenties per day through all of early May: 8, 13, 14, 21. Every morning I'd check and it was still flat. This long quiet stretch was agony. I was not sure if the decision was brave or stupid.

Then in June it finally started climbing again, and this time slowly and honestly:

• Jun 1: 35

• Jun 3: 58

• Jun 6: 86

• Jun 7: 96

• Jun 10: 166

• Jun 11: 167

Roughly six weeks after the crater, I was back to about where I'd been before I torched it. Same altitude, with a completely different foundation.

What's different now

The 150 from April was junk. The 166 from June is made of the right stuff:

1. Long-tail integration searches like fellow todoist integration, ticketmaster connector, trello gmail integration from people trying to wire two specific tools together.

2. Actual intent for what I build like how do i add a human review step and human approval before ai makes payment.

That second bucket is tiny, in the single-digit impressions each, but it's the first time Google has shown my pages to people describing the exact problem I solve. I would take 9 impressions of that over 700 impressions of "wrong product's API docs."

The honest part

The number I keep next to all of this so I don't lie to myself: across four months, March through June, I've gotten about 8 total clicks from Google. Eight.

Impressions aren't traffic. Traffic isn't signups. I'm at the very bottom of that ladder, and a rising impression chart had me feeling further up it than I am. Watching the panic-drop and the slow climb back taught me more than the original line up-and-to-the-right ever did.

What I'd tell past me

1. A rising chart isn't a win until you've checked which queries it's made of. Break impressions down by query before you celebrate.

2. Wrong-intent traffic is worse than no traffic. It pollutes the signal you use to make decisions.

3. You can know a number is going to drop and still panic when it does. Don't reverse a good decision because the graph looks scary for a few weeks.

4. Recovery to the "same" number can mean something completely different underneath. Same height, better base.

Next experiment: I started answering questions on Quora today. At first I just wanted to see what questions people were asking so I could write some targeted content but then I started answering and found it kind of fun. I'm able to reach the human-approval-intent crowd directly instead of waiting for Google to do it. I have no idea if it'll pay off, but I'll post the numbers if I stick with it.

What's the most useless metric you caught yourself optimizing?