The Meter Was Broken the Whole Time

On April 3, 2026, Google quietly disclosed that a logging error had been inflating impression data in Search Console since May 13, 2025. Eleven months. The disclosure was roughly 47 words long, buried in a documentation update that nobody checks unless something already looks wrong.

Every impression number. Every CTR calculation. Every visibility report a site owner stared at during a morning coffee, trying to decide whether their content strategy was working. For almost a year, those numbers were inflated. Not by a little. Enough that Passionfruit Labs, after studying 17 properties, concluded year-over-year comparisons are “not meaningful.”

I want to sit with that phrase. The entire premise of tracking SEO performance is comparison over time. If year-over-year data isn’t meaningful, what exactly has anyone been reporting for the last eleven months?

The clicks were accurate. Google was very clear about that. Only impressions were affected. Which means only the single most important metric for understanding whether Google is showing your content to real people was compromised. Just that.

The Compound Bug

This wasn’t one problem. It stacked. Five documented data discontinuities in one tool across eleven months, each one making the picture less reliable than the last.

May 2025: the impression logging bug begins. Nobody is told. June 2025: Google merges AI Mode clicks and impressions into the “Web” search type with no way to separate them. Your impression data is now both inflated by a bug and blended with AI Mode traffic you can’t isolate. September 2025: Google deprecates the &num=100 parameter researchers used to study long-tail SERP data. October 2025: a data freeze affects reporting. February and March 2026: two full days of bulk export data vanish entirely. Google confirms they will not be recovered.

Five breaks. One tool. One year.

And it kept getting worse. On April 16, 2026, Google disclosed a new logging error on top of the old one. Impressions and clicks for Job listing and Job details search types stopped reporting. Two data integrity failures running simultaneously in the same tool. The Performance report delay, which normally cycles every four to five hours, stretched past 26 hours. It had been 13.5 hours two days before that. Then 17.5. Then 26. The gap was widening, not closing.

The Part Nobody Talks About

Google Ads impression data works perfectly. It has always worked perfectly.

Every impression tracked. Every click attributed. Every conversion recorded. Every dollar accounted for. The system that bills advertisers has never had an eleven-month logging error. Has never blended data sources without a filter. Has never lost two days of export data that won’t be recovered.

The free tool broke. The paid tool didn’t.

I want to be careful here because I genuinely don’t know whether the asymmetry is intentional. The structural argument is strong and I find it persuasive roughly 60% of the time. Ads data has contractual and legal obligations attached to it. Billing data has to be accurate or Google faces lawsuits. Organic data has no such obligation. It’s provided as a courtesy. Different incentive structures produce different reliability outcomes. That’s not conspiracy. That’s capitalism.

But the effect is the same regardless of intent. The tool that costs money works. The tool that’s free doesn’t. And the free tool is what millions of site owners use to decide whether their organic strategy is worth continuing or whether they should switch to paid. When the free tool shows declining impressions, some of those owners conclude organic isn’t working and move their budget to ads. The broken free tool becomes a funnel to the working paid tool. Whether anyone designed it that way is something I can’t answer. That the outcome exists is something I can observe.

What This Means If You Run a Site

If your site launched after May 13, 2025, every impression number you’ve ever seen in Google Search Console was generated during the bug window. All of it. Your baseline was wrong from day one.

I ran a brand new website through this exact window. Watched the impressions climb from 150 to over 10,000 and felt good about the trajectory. The trajectory was probably real in shape if not in scale. The direction was right. The speed was a lie. And I had no way to know because the only speedometer in the car was the one Google installed.

Here’s what I’d actually trust right now. Clicks. Clicks were never affected. If someone clicked through to your site from a search result, that click is real. Your click data is clean. Everything else has an asterisk.

Ranking positions are probably reliable, though I haven’t seen Google explicitly confirm that position data was unaffected. That silence is worth noting. If position data was clean, saying so costs nothing and reassures millions of users. The absence of that statement makes me uneasy in a way I can’t fully justify with evidence. (My read could be wrong here. Google might consider it obvious. But “obvious” and “confirmed” are different things and I’ve been doing this long enough to know the difference matters.)

The Industry Response

Muted. That’s the honest word for it.

Some SEO publications covered the disclosure. Passionfruit Labs ran a proper study. The information exists if you go looking. But I haven’t seen the kind of industry-wide reckoning you’d expect from learning that the primary free measurement tool for organic search had been feeding bad data to every website owner on earth for eleven months.

Agencies didn’t send urgent emails to clients saying every report since May 2025 was based on inflated numbers. SEO tool companies didn’t flag their dashboards with a warning that the upstream data source was compromised. Conference speakers didn’t cancel their 2025 year-in-review talks because the data underpinning their conclusions was wrong. Everybody adjusted the narrative slightly, noted the bug in a footnote, and kept going.

I’ve worked around enough measurement systems to know that an eleven-day error with this kind of scope would cost someone their job in most industries. Eleven months didn’t even trend. That gap between the severity of the failure and the volume of the response tells you something about how deeply the industry depends on not questioning the platform it’s built on.

Platform Dependency

The deeper issue isn’t the bug. Bugs happen. Systems break. The deeper issue is that there is no alternative.

Google Search Console is the only tool that shows you how Google sees your site. Third-party tools like Ahrefs, Semrush, and Moz estimate what Google sees based on their own crawling. They’re guessing. Educated guesses, sometimes very good guesses, but guesses. GSC is the source. And the source was broken for eleven months while currently running over a day behind.

The entire SEO industry, every agency, every freelancer, every best AI tools roundup that references search visibility, every content strategy built on impression data, is downstream of one company’s free tool. No SLA. No compensation. No contractual obligation to accuracy. No alternative provider. The tool breaks, Google posts 47 words, and we adjust our spreadsheets and move on because what else would we do.

This is the same dependency pattern I wrote about when covering the guest post spam ecosystem. The entire link building industry prices its product using Ahrefs and Moz metrics that Google doesn’t use. And now the measurement tool Google does provide turns out to have been broken for a year. We built an industry on a single company’s infrastructure and we have no recourse when that infrastructure fails. At every layer. The metrics are borrowed. The tools are borrowed. The shelf space is rented. Nothing is owned.

What I Can’t Resolve

The question I keep circling is whether the asymmetry between free and paid data quality is a structural inevitability or something closer to a strategy. I’ve gone back and forth on this more than I’d like to admit and I still don’t have a clean answer.

The structural argument says ads data carries legal consequences for inaccuracy. Organic data doesn’t. Different standards emerge from different incentive structures. I find this persuasive most days.

Then I remember that the same company selling the ads also operates the free measurement tool that defines whether organic is working. In most industries, that arrangement would be called a conflict of interest. In search, it’s called the ecosystem. A system can be structurally biased toward a certain outcome without anyone consciously designing it that way. The result is the same either way.

I don’t know which explanation is correct. I used to think the structural argument was sufficient. After watching the delay climb from 13.5 hours to 26 while the ads dashboard updated in real time, I’m less sure. That’s an honest shift in my thinking, not a conclusion. The data moved me but it didn’t land me anywhere I’m confident enough to plant a flag.

The Meter

There’s an old story about a gas station that gets caught with a rigged pump. Reads a gallon but delivers nine tenths. Nobody notices for months because the difference per fill-up is small enough to escape suspicion. When it gets caught, the station owner says calibration error. Maybe it was.

Google’s impression meter was reading high for eleven months. Showing more than it delivered. Nobody noticed because the difference was spread across billions of data points and nobody had a second pump to compare against.

The difference between the gas station and Google is that the gas station had a regulator who checked the pumps. Google is the pump, the regulator, and the gas.

There’s nobody to check. And the meter is still broken. Over 26 hours stale and counting.

AI Autopsy is a series documenting platform failures from inside the machine. When the tools we depend on break, someone should be writing it down.