Should brand search be its own channel in a mix model?

Yes, usually. Google's Meridian docs say brand and generic search campaigns are very different and fit best as separate media channels. Each gets its own query volume as a control. Brand campaigns are often modeled on clicks, since they aim to drive direct web traffic.

Why does my mix model give brand search the best ROI?

Usually because people search your name whether or not you bid on it. Without branded query volume as a control, the model credits the ads for demand they only collect. Meridian's docs say leaving query volume out can overestimate paid search.

Does a big baseline mean my brand is strong?

Not by itself. The baseline holds whatever the model cannot tie to your marketing: trend, seasonality, holidays, outside factors and any channel you left out. Meridian does not split the baseline by control variable, so the model cannot say how much of it is brand.

How does MMM handle brand effects?

Usually in three places. Carryover catches brand effects that fade within weeks, and the baseline absorbs slow brand building. Branded search volume goes in as a control or a separate stage. Leave that volume out and search ads look too good; put it in as a control and brand ads look too weak.

By Joris van Huët, Founder & CEOUpdated 2026-10-017 min read

Usually in three places. Carryover catches brand effects that fade within weeks. Slow brand building lands in the baseline, which no channel gets credit for. Branded search volume goes in as a control or a separate stage. Leave it out and search ads look too good; put it in as a control and brand ads look too weak.

The usual answer is that a mix model sees brand because it reads total sales, not clicks. That is half right. A model credits a channel for the sales that rise and fall with its spend, inside a window you set. Brand effects slower than that window, or routed through search, end up somewhere else.

Carryover catches the short part. Meta's Robyn docs describe carryover as the effect of brand equity measures such as ad recall or campaign awareness. Google's Meridian cuts lagged effects off at a setting called max_lag. Its docs suggest 2 to 10 periods with geometric decay and 4 to 20 with binomial decay. In a weekly model, even the top end is under five months.

The baseline holds the slow part. Meridian's baseline is what the model expects without paid media, organic media or other treatments. Robyn's docs say the long-term effect should describe media's impact on baseline sales, and that adstock says nothing about it. So a brand that grew for years shows up as a rising baseline, credited to nobody.

Branded search is where credit moves. Meridian's docs call query volume an important confounder for search ads. When demand rises, people search more and search ads spend more, so the ads collect credit for demand they did not create. Leave branded query volume out and search looks too good. Put it in as a control and brand media loses credit for the searches it caused.

Meridian's way out, when your data can carry it, is a two-stage model. Brand media builds branded search first; then branded search and all other media drive sales.

What one store's data shows

One store's anonymised GA4 export, 1 January 2024 to 21 August 2026. It holds shares of revenue only: no ad spend, no order counts.

What the export shows	Share of revenue	Source cell
Direct, in last click, first click and touched views	57.7%	Channels sheet, Direct row
Journeys with 1 touch (0.5 days to buy)	79.5%	Journeys sheet, 1 touch row
Paid Social, in all three views	0.0%	Channels sheet, Paid Social row

This export is not a mix model. It has no weekly spend and no weekly sales, so it cannot measure a brand effect. It shows where a mix model would have to look for one.

On the Channels sheet, Direct holds 57.7% of this one store's revenue in last click, first click and touched alike. Google defines Direct as arriving through a saved link or by typing the address. Both need a buyer who already knows the shop.

A mix model has no spend line for Direct. Those sales sit in its baseline unless some channel's weekly ups and downs explain them. Whatever taught those buyers the name gets no credit there.

On the Journeys sheet, journeys with 1 touch hold 79.5% of revenue and took 0.5 days to buy. That fits buyers who had decided before the one visit GA4 recorded. Where they decided, the export cannot say: an old video, a friend and a past email look the same from here.

On the Channels sheet, Paid Social holds 0.0% of revenue in all three views. The export does not say whether this store ran social ads. If it ran awareness video, a zero here would not mean the video did nothing. A viewer who later types the address lands in the Direct row instead.

What the export cannot show is timing. It holds shares for the whole period, not the weekly swings a model needs to tie sales to a brand burst.

Why does "MMM sees brand" mislead?

Because a model can only credit what moves with spend, inside its window. Three things get in the way.

Always-on brand spend gives the model little to learn from. Robyn's guide gives the example of TV spend that stays constant while sales move: the model struggles to tell what TV did. Flat brand spend has the same problem.

Short windows cut off the tail. Meridian's decay docs flag geometric decay as vulnerable to long-term effect underestimation. Binomial decay and a longer max_lag help, but Meridian warns that long lags slow the model and raise its variance.

Search collects the credit. Meridian's paid search page says failing to control for query volume can overestimate paid search. Brand search is the clearest case: a search for your name means the buyer already had you in mind.

What can a mix model not tell you about brand?

It cannot label the baseline. Meridian does not split the baseline by control variable, so a rising baseline carries no name. It could be brand ads, a better product, reviews or a growing category.

Robyn files trend under baseline too, next to season, holidays and outside factors such as competitors and weather. The model treats all of them as things you cannot steer.

And a lift test only covers its own weeks. Meridian's calibration guide asks whether an experiment ran long enough to capture the long-term effects of marketing. A four-week test of a brand video measures four weeks.

What to do this week

Chart Direct and Organic Search by week in GA4. In Explore, start a Free form exploration and switch the visualization to a line chart. Set Granularity to week and break down by Session default channel group. Pass: you can line up each awareness burst with the weeks after it. Fail: the bursts are not on a calendar, so log their dates first.
Check that your brand search campaign only buys your name. In Google Ads, select Campaigns, then Search terms in the Insights and reports sub-menu, and read the brand campaign's rows. Pass: nearly every term contains your brand name. Fail: generic terms crept in, so move them to their own campaign before any model sees the data.
Check whether your awareness spend ever moved. In Meta Ads Manager, select your Awareness campaigns, click the Breakdown icon, then By time, and pick week. Pass: weekly spend has clear highs and lows a model can learn from. Fail: the same spend every week, which leaves a model guessing.

Check the homework. Your GA4 Attribution paths export already holds the evidence. Causality Engine reads that one file and shows what each channel caused next to what last-click gave it, in 1 to 2 minutes, for €99 once (excluding VAT), refundable within 30 days. Check the homework

Sources, 1 October 2026: Key Features (Meta); Set the max_lag parameter (Google); Assess the baseline (Google); Paid search modeling (Google); Full-funnel MMM: unifying upper-funnel and lower-funnel measurement (Google); Default channel group (Google); An Analyst's Guide to MMM (Meta); Set the adstock_decay_spec parameter (Google); Control variables (Google); Calibrate treatment priors (Google); Free-form exploration (Google); About the search terms report (Google); Navigate to breakdowns in Meta Ads Manager to understand ad performance (Meta); About breakdowns, metrics and filtering in Meta Ads Reporting (Meta)

Frequently asked questions

Should brand search be its own channel in a mix model?
Yes, usually. Google's Meridian docs say brand and generic search campaigns are very different and fit best as separate media channels. Each gets its own query volume as a control. Brand campaigns are often modeled on clicks, since they aim to drive direct web traffic.
Why does my mix model give brand search the best ROI?
Usually because people search your name whether or not you bid on it. Without branded query volume as a control, the model credits the ads for demand they only collect. Meridian's docs say leaving query volume out can overestimate paid search.
Does a big baseline mean my brand is strong?
Not by itself. The baseline holds whatever the model cannot tie to your marketing: trend, seasonality, holidays, outside factors and any channel you left out. Meridian does not split the baseline by control variable, so the model cannot say how much of it is brand.

Go deeper: Causal attribution, explained.

Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.

Keep reading

Terms in this article

Browse the full glossary