Google Search Console rolled out something that looks like a small UI improvement but is, for many websites, a significant shift in how you measure SEO performance.
The new branded queries filter. Let me show you why it matters more than most announcements from Google.
Our Data: 10 Projects, Old Method vs. New Filter
We pulled two weeks of data (23 Feb – 8 Mar 2026) across 10 projects and compared the brand share from our old regex approach against the new GSC branded filter. Here’s what we found:
Brand Share: Old Regex Method vs. New GSC Filter
| Project | Total Clicks | Brand % (regex) | Brand % (GSC filter) | Difference |
|---|---|---|---|---|
| Project A | 38 200 | 2.6% | 0.8% | +238% overstated |
| Project B | 45 000 | 52.5% | 18.9% | +178% overstated |
| Project C | 76 000 | 6.9% | 3.6% | +91% overstated |
| Project D | 445 000 | 39.7% | 21.2% | +87% overstated |
| Project E | 8 670 | 10.8% | 6.5% | +67% overstated |
| Project F | 527 | 41.7% | 27.1% | +54% overstated |
| Project G | 34 000 | 4.2% | 5.2% | -20% understated |
| Project H | 16 100 | 19.1% | 20.3% | -6% (similar) |
| Project I | 122 000 | 10.6% | 11.0% | -4% (similar) |
| Project J | 1 700 | 4.7% | 4.5% | +4% (similar) |
Key Summary Stats
| Metric | Old Method (regex) | New GSC Filter |
|---|---|---|
| Average brand share | 19.3% | 11.9% |
| Average query coverage | ~41% of total clicks | 100% of total clicks |
| Projects with >50% difference | 6 / 10 | — |
| Projects within ±10% | 3 / 10 | — |
How Much of Your Data Did the Old Method Actually Use?
The regex approach only works with queries GSC shows you. Here’s how much of total traffic was actually covered per project in our sample:
| Project | Query coverage (old method) |
|---|---|
| Project A | 22.6% |
| Project B | 29.4% |
| Project C | 33.8% |
| Project D | 40.2% |
| Project E | 34.9% |
| Project F | 58.3% |
| Project G | 29.8% |
| Project H | 64.4% |
| Project I | 50.2% |
| Project J | 50.9% |
In the worst case, the old method was making a brand/non-brand judgment based on less than a quarter of actual clicks. And as mentioned, that visible subset is biased. Brand queries surface in GSC at a higher rate, inflating the apparent brand share.
The Problem: You’ve Been Measuring Brand Share With Incomplete Data
Google Search Console has a well-known limitation. A large chunk of clicks, often 60–80% or more, have no search query attached. These are the “not provided” queries. GSC shows you impressions and clicks for them, but you don’t know what keywords drove them.
This created a real problem for anyone trying to answer: what share of our SEO traffic is branded vs. non-branded?
Because you can only filter queries you can see.
The Two Workarounds We’ve Been Using
Method 1: Regex filter on known queries
You define all brand terms (e.g. brand|name) and apply a regex filter in GSC. You get a count of brand clicks and non-brand clicks. But only within the queries that GSC actually shows you.
Then you extrapolate. If GSC shows you queries for 35% of your total clicks, you apply a rule of three: assume the unattributed 65% splits in the same ratio as what you can see.
It works. But it’s an estimate. And the estimate has a systematic bias: brand queries tend to be “known” in GSC at a higher rate than generic queries, which inflates the apparent brand share.
Method 2: Homepage traffic as a brand proxy
For sites with a strong brand (think direct navigation, brand searches), homepage clicks are a reasonable proxy for branded traffic. You take total clicks minus homepage clicks = estimated non-brand traffic.
This works for some sites, not others. If brand queries land on product pages (not just the homepage), you’re undercounting brand. If the homepage gets non-brand clicks, you’re overcounting.
Both methods are workarounds. Both introduce error.
What the New Filter Does
Right now, Google Search Console classifies in many projects all queries as branded or non-branded. Including the “not provided” ones.
The classification is done by Google’s AI, based on patterns it learns from your site: your brand name, variations, misspellings, product-brand combinations, and so on.
This means for the first time, you get brand vs. non-brand split across 100% of your clicks, not just the subset GSC happens to show you.
The filter is available in the GSC interface. As of this writing, it does not work via the API, which means no Looker Studio export, no BigQuery, no automated reporting. UI only for now.
Key Findings
1. The old method systematically overstates brand share
In 6 out of 10 projects, the regex approach overstated brand traffic by more than 50%. In two cases (Project A and Project B), the old estimate was nearly 3x higher than the new filter shows.
The direction is almost always the same: old method makes brand look bigger, non-brand look smaller.
2. For high-brand projects, the difference is strategically significant
Project D (a well-known e-commerce brand) showed 39.7% brand share with the old method. A number that might make you question whether SEO is driving real discovery traffic.
The new filter says 21.2%. Still significant brand traffic, but the non-brand contribution is nearly double what the old numbers suggested.
3. For low-brand projects, the methods largely agree
Projects G, H, I, and J (all with brand share under ~20%) show very small differences between the two methods (within ±20%). When brand is a small share of total traffic, the bias from the old method has less room to distort the picture.
4. The new filter isn’t perfect
In one project in our sample (not in this table), the AI classification was clearly off. It was assigning non-brand queries as branded and vice versa. Before relying on these numbers for KPI reporting, it’s worth doing a manual spot-check of the branded query list GSC is using.
What This Means in Practice
If you set SEO targets based on brand vs. non-brand splits (which you should, especially for established brands), your baseline numbers may need to be recalculated.
The old approach wasn’t wrong for the data it had. But if you’re now using the new filter and comparing against historically regex-based numbers… It’s much better now.
