For
- SEOs
- Content leads
- Analysts
Is this you?
- Traffic changes nobody can explain
- Reports that disagree about the same site
- Search Console data that is only ever looked at, never exported
Why Search Console misleads the people who trust it most
Search Console is Google’s own data about your visibility in Search. So teams trust it. And they read it like a finished report, when it is a data set with rules.
Some queries are never listed. The table shows a cut of the rows. A counting change can move a number overnight. Each rule is documented, and each one bends a verdict if you do not know it.
Google says it plainly: “Anonymized queries are those that aren’t issued by more than a few dozen users over a two-to-three month period.” (Google Search Central, A deep dive into Search Console performance data filtering and limits)
In a 2026 session I put my goal for the day this way: “it’s not about the totals anymore.” Your team reads the relationships, and answers the next “why did it move?” without me.
The day, in the order of the questions
Each block ends with a number from your own property.
- What the report counts. Clicks, impressions, average position: what each one is, and which one to trust first.
- Why traffic moved. Compare periods, split by every tab, and date the change from the data, not from the ticket.
- Brand and non-brand. One rule that separates them, and the denominator to read them against.
- Page clusters. Every URL in one page type, so you read types, not a top-10 list.
- What Search Console hides. Anonymized queries, the row limit and the 16-month window, measured on your property.
- Export. Your data through the API, and when BigQuery is the better road.
- Analysis with AI. An assistant that works through your files, and the checks that keep it honest.
- Your routine. What you look at, how often, and what counts as a finding.
What I see teams get wrong in 2026
These are the reading mistakes I measured in my own 2026 sessions and analyses. The day is built around them.
| What goes wrong | What I measured | What I specify |
|---|---|---|
| A query table read as the whole traffic | About 60 percent of a site’s web clicks were in queries the table does not list | Chart total, row sum and hidden share side by side |
| A screen mistaken for the data | The screen export stops at 1,000 rows; one 90-day pull had about 50k query rows | The API with paging, or the bulk export |
| A brand share with the wrong denominator | Just over half of the clicks you can see by query, about a fifth of all clicks | Brand pattern, denominator and window written down |
| A counting change read as a ranking loss | Impressions nearly halved, clicks down about a tenth, position about 5 to about 9 | Clicks first, Google’s data anomalies page second |
| Top-10 lists instead of page clusters | About 47k URLs of one type earned over 40 percent of clicks, under 1 click per URL | One rule that assigns every URL to a type |
| AI impressions added to web impressions | About a fifth of one site’s impressions came from AI features, and they are a subset | The AI view as a slice, never as an addition |
| Last year’s baseline gone | Nothing before mid-2025 came back from the API | Store the data now, before you need it |
A query table read as the whole traffic
Ahead of one 2026 session I pulled 90 days of Search Console data for a site with a large long tail. Queries that Search Console lists carried about 35k clicks. The property total was about 90k.
- Hidden in anonymized queries: about 55k clicks, about 60 percent of all web clicks.
- A table that is filtered by query leaves these out. A chart total includes them.
Google documents both sides: anonymized queries “are included in chart totals, unless you filter by query” (source).
What I specify: three numbers next to every query view - the chart total, the sum of the listed rows and the hidden share. A query table is evidence about the visible long tail, not about the whole site.
A screen mistaken for the data
In the session itself I showed the live report and pointed at the limit of its filtered views: “1,000 rows”. The 90-day pull behind it had about 50k query rows. A screen export reaches 2 percent of them.
Google: “The maximum you can export through the Search Console user interface is 1,000 rows of data.” (source) The API limit is “50,000 rows per day per site per search type” (source).
What I specify: pull through the API with paging, not from the screen. On very large sites, switch on the bulk export: “the bulk data export is not affected by the daily data row limit” (source). It leaves out anonymized queries, too.
A brand share with the wrong denominator
On the same site, a live read of the screen said 60 to 70 percent branded, for 28 days. In my 90-day pull the answer depended on what I divided by:
- Brand queries: about 20k clicks.
- Just over half of the clicks listed by query (about 35k).
- About a fifth of all web clicks (about 90k). That is my arithmetic and my inference: it assumes the hidden queries are not brand, and the brand queries are far too frequent to be hidden.
Three honest-looking answers to one question, from different windows and different denominators. A filtered unbranded view gave me the same warning in my own capture log: query-filtered totals leave the hidden clicks out, so the result “is not a clean unbranded share”.
What I specify: one brand pattern that catches misspellings (a short stem of the name, not the full word), one window, one denominator, written next to the number. Brand and non-brand are two curves with their own totals.
A counting change read as a ranking loss
In one 2026 analysis I recomputed a report’s table from the raw API data. Its numbers were right to the digit:
- Impressions: from about 210k to about 115k a day. That is nearly half.
- Clicks: from about 830 to about 730 a day, about a tenth less.
- Average position: from about 5 to about 9.
That is the pattern Google documents for a logging fix on 27 April 2026: “A logging error prevented Search Console from accurately reporting impressions from May 13, 2025 until April 27, 2026” (Search Console Help, Data anomalies). Impressions, CTR and position were affected, but “clicks were not affected by the error” (source).
The full forensic case is on my search data page.
What I specify: read clicks first. Impressions, CTR and position are derived from the same log, so check the anomalies page before you explain a jump in any of them.
Top-10 lists instead of page clusters
A list of the top URLs hides the shape of a site. In one 2026 pull I assigned every URL to a page type with one rule and counted:
- One type: about 47k URLs with impressions, over 40 percent of all clicks, under 1 click per URL.
- The home page, a handful of URLs: about a fifth of clicks, about 2k clicks per URL.
- The pages that convert earned a few percent of clicks.
None of that shows in a top-10 list. It shows in clusters, and in clicks per URL.
What I specify: one rule that assigns every URL to a type, the same rule in every report. Read clicks, impressions and clicks per URL per type, with the URL count beside them. Do the same for countries and devices.
AI impressions added to web impressions
In 2026 Google added reports for AI features. On one site they showed about 2M impressions in 11 weeks, about 22k a day, against roughly 105k impressions a day in total. About a fifth.
That is a slice, not an addition. Google: “This data is included in the overall performance report” (Google Search Central, 3 June 2026). The report counts impressions only. It has no clicks, no CTR and no queries.
What I specify: use the AI view to see which pages appear, not what they earn. Never add it to the web numbers, and never read it as traffic.
Last year’s baseline gone
Search Console keeps 16 months. On 12 November 2026 I asked the API for my own property’s data from 1 January 2024. The first day it returned was 30 June 2025.
Google’s advice: “If you’d like to extend the 16 months, you could use the Search Analytics API or bulk data exports to pull data and store it in your systems.” (Google Search Central, Debugging drops in Google Search traffic, updated 10 December 2025)
What I specify: a monthly pull of the full data into your own files, starting the week of the workshop. In a year, last year’s April is the baseline you will wish you had.
How it runs
- Access first. I need read access to your Search Console property before the day, and the two or three questions your team could not answer last quarter.
- The morning: the report as data. Every block ends with a number from your property: the hidden share, the brand split, the page types.
- Midday: the export. You pull your data through the API into files. In one half day, about 40 scripts ran against the API in four hours. That is what a team can do in a few hours.
- The afternoon: analysis with AI. An assistant works through your files. Every statement it makes carries a tag, as in my 2026 reports: measured, Google’s own words, or my assessment. Then you recompute one number by hand.
- The routine. In a 2026 handover I set it at Search Console twice a week, and a check of every change after 2 weeks and after 2 months.
- What you keep. Your exports, the scripts, the prompts and the checks.
I teach this in two settings: on your property in-house, or as one module of a longer SEO training. My AI First SEO workshops in 2025 carried the same analysis with AI as a module.
My own property, read this way
I ran the same pull on this site’s Search Console property, sc-domain:fullstackoptimization.com, for the 90 days from 12 August to 9 November 2026 (final data only):
- Chart total: 223 clicks, 56,961 impressions.
- Sum of 1,183 query rows: 61 clicks (27 percent), 35,116 impressions (62 percent). 162 clicks, 73 percent, sit in hidden queries.
- Sum of 114 page rows: 229 clicks and 60,582 impressions. More than the chart total. Google puts such gaps down to aggregation: “The chart totals can sometimes differ from the table totals. This is usually due to differences in aggregation (property vs. page).” (Search Console Help, Performance report)
- Rows: the API returned exactly 1,000 rows by default. With a row limit of 25,000 it returned all 1,183.
- History: nothing before 30 June 2025.
My verdict: for this site the query table is a sample. The page and country views are the reliable ones. I only know this because I counted.
Questions I get before a Search Console day
Do we need a big site?
No. A small property shows the same rules: mine hides 73 percent of its clicks in queries it does not list. A large property needs the API and the bulk export earlier.
Is Search Console data sampled?
For privacy and size, yes, in two ways: anonymized queries are left out, and very large sites hit a daily row limit. Google calls the remaining data “a representative sample of data” (source). Requests without a query or URL dimension, such as countries or devices, return all the data.
Why do Search Console and analytics disagree?
They count different things. Google: “Clicks and sessions are calculated differently, which means that when you compare the data, you’ll likely see different numbers.” (Google Search Central, Using Search Console and Google Analytics data for SEO, updated 7 January 2026) Compare the pattern, not the digits.
What does average position mean?
Google: “The average position of the topmost result from your site.” (Search Console Help, Performance report) It moves with the mix of queries and with counting changes. Filter by one country and read it next to clicks.
Can an AI analyse our data on its own?
It can work through your files faster than any person. It cannot tell you which number is wrong. That is why the day ends with a hand recomputation, and why I tag every statement as measured, quoted or assessed.
Does the day change anything on our site?
No. It changes how your team reads the data. If a finding needs a fix on the site, the search data and traffic forensics service is where I take a cause to a verdict.
Sources for the Google quotes, read on 2026-11-12: A deep dive into Search Console performance data filtering and limits, Bulk data export, Debugging drops in Google Search traffic, Using Search Console and Google Analytics data for SEO and Search Generative AI performance reports, Google Search Central; Performance report and Data anomalies in Search Console, Search Console Help. The client numbers come from my own sessions and analyses in 2026, my own Search Console pull of 2026-11-12 and my workshops since 2022.
Some of our customers
Examples, not the full list
From real workshops
- 2026: a session prepared on 90 days of Search Console data; about 60 percent of the site's clicks sat in queries it does not list.
- 2026: a report's table checked against raw Search Console data; its numbers were right to the digit, its cause was a counting change.
- 2025: Search Console analysis with AI as a module of my own AI First SEO workshops.
- 2024: one half day from Search Console screens to about 40 scripts against its API.
- 2022: a Search Console workshop with a handover pack.
- A standalone day, and one module of a longer SEO training.
Consulting for the next step
How consulting worksOther workshops
All workshop formats- SEO & GEO Workshop
- AI First Workshop
- Technical SEO Workshop
- AI Computer Use Workshop
- AI Content System Workshop
- Content SEO Workshop
- AI SEO Workshop
- Web Performance Workshop
- Relaunch Workshop
- Post Relaunch Workshop
- SEO Concept Workshop
- Newsroom SEO Workshop
- Google Discover & News Feed Workshop
- AI Newsroom Workshop
- International SEO Workshop
- Growth Workshop
- AI Agents Workshop
- AI Thinking Workshop
- Dev Thinking Workshop
- SEO Training for In-house SEOs
- SEO Setup Workshop