Google Hides Half Your Clicks. I Counted.

The numbers are wrong and the room knows it

There is a moment in every reporting meeting where someone reads a number off a dashboard and everyone nods, and nobody in the room believes it, including the person reading it. That moment has gotten longer this year.

On September 1, Google Analytics 4 standard reports showed zero traffic for the day across accounts, a processing bug, not a real drop, though the chart did not include a note saying so. On August 26, Google confirmed it now routes clicks on search results through a google.com/goto redirect, and its statement about why was that it has "a long history of deploying technical measures against evolving forms of abuse." It has not published what that does to your referrer data. And Search Console, the one report everyone treats as the clean one, withholds queries on purpose and has for years.

Three separate pieces of Google measuring Google. The referee, the scorekeeper, and the team, and they all share a Slack. So I stopped arguing about whether the numbers are off and counted how far.

775 clicks. 400 of them have a query. The rest are a shrug.

My agency, Local Howl, stores a daily Search Console snapshot for the clients who connected it. I took the latest 28-day window, August 22 through September 19, for five of them: a window company, a car wrap shop, an HVAC contractor, a dentist, and a pool builder. Small local sites, the kind that run the country.

Total clicks across the five, as reported at the property level: 775. Clicks the page report could account for: 781, which is every click plus a little double counting when one click matches two page rows. Clicks the query report could account for: 400. Fifty-two percent. The other 375 clicks happened, Google counted them, and Google will not tell you what the person typed.

Per site, the query report explained 71 percent of the window company's clicks, 57 percent of the dentist's, 50 percent of the HVAC contractor's, 36 percent of the wrap shop's, and 22 percent of the pool builder's. The pool builder got 45 clicks in a month and can see the words behind 10 of them. (The other 35 people were presumably looking for pools. Bold assumption. I stand by it.)

Before anyone blames the export: the pull keeps 500 query rows per site, and on all five sites the 500th row had zero clicks. The cap did not eat anything. Google did.

This is not a small-site problem. It is the average.

Ahrefs measured this across 22 billion clicks and 887,534 Search Console properties and found 46.77 percent of traffic sitting in anonymized queries. My five sites came in at 48. I am not special. I am the median with a spreadsheet.

Google's own help text used to call these queries "very rare." It quietly changed the word to "some." Half is some. Technically. In the way that a shark is some fish.

The practical damage is not the missing words. It is what people do with the half they can see. A client looks at the query table, sees brand terms at the top, and concludes search only brings people who already knew them. Then they cut the content budget that was producing the other half, the half Google would not label. Last click gets the credit, the blog gets the blame, and the decision was made on a table that was missing 375 rows.

Direct traffic is where attribution goes to be forgotten

Search Console is the tidy version of this. GA4's direct bucket is the messy one. AI assistants that send a visitor without a referrer, links shared in texts and Slack, consent banners that wipe the cookie before the first pageview, bots with unusual screen sizes and two-second sessions. All of it lands in Direct, and Direct has a conversion rate someone is using to justify ad spend.

I do not have a clean number for how much of Direct is real, and neither do you. That is the whole problem. Server-side tagging does not fix it either. It relocates the broken data into a system that is harder to debug, which is progress the way moving the mess to the garage is cleaning.

Here is the test I use in the meeting, borrowed from a practitioner who was tired of defending decoration. If this number changed by 30 percent tomorrow, would anyone in this room do anything differently? If the answer is no, it is not a measurement. It is a wall hanging. Impressions are a wall hanging. Reach is a wall hanging. A bounce rate on a page whose only job is to show a phone number is a wall hanging with a frame.

Triangulate. Nobody gets a single source of truth anymore.

One. Report search at the page level, not the query level. The page report explained every click in my count. The query report explained half. Use the report that adds up, and treat the query table as a sample of what people typed, never as the total.

Two. Put your own capture where the money is. A tracking phone number per channel. A form that records the referrer on the server, before the consent banner and the redirect get a vote. A field that asks how they heard about you, answered by an actual human, which is the least sophisticated attribution model ever built and still the hardest to argue with.

Three. Get an estimate that does not run through your own tags. This is the job the Semrush Traffic & Market Toolkit does for me. Traffic Analytics estimates any site's visits and channel mix without touching its analytics, yours or a competitor's, so when GA4 says your direct traffic doubled you have a second signal that either agrees or laughs. Market Overview gives you the channel mix for your whole category, which is how you answer "is this number even plausible" before the meeting instead of during it.

Four. Report decisions, not dashboards. Every number on the page should sit next to the thing it changed. If nothing changed, delete the number. The report gets shorter. The meeting gets shorter. Everyone pretends to miss it.

The numbers were never precise. They were just confident.

The measurement side of AI search, what to track when clicks stop being the point, is in the AI search KPI check. The reason your query report is full of brand terms and not much else is half explained above and half explained by the zero-click piece.

Half your search clicks have no label. Your direct bucket is a landfill with a conversion rate. Google's referee just started wearing the team jersey. None of that is a reason to stop measuring. It is a reason to stop pretending one report was ever going to do it. Count what you can count yourself, get a second opinion that does not run on Google's tags, and put a decision next to every number that survives.