Should You Put Google's Branded Search Volume in Your MMM?

Branded search volume can make a marketing mix model (MMM) more accurate. It can also decide how much credit your TV and YouTube get, using a number Google owns. In September 2026, Google made it easy to feed branded search volume, which it calls Branded Google Query Volume or bGQV, into its Meridian model as a brand signal, so TV and YouTube can get credit for the sales they drive indirectly, through brand demand, instead of the model stripping that demand out or handing it to lower-funnel channels.

That fixes a real problem. It also raises the stakes, because your TV and YouTube ROAS now rides on a chain of assumptions, not just a model that looks accurate. Here is what to check before you move a budget on it.

What is branded search volume, and what did Meridian change?

Branded search volume is how many people search your brand name over time. Google packages it as Branded Google Query Volume (bGQV) and hands it over as a data feed. It has been around for a while to clean up your search numbers. What is new is that Meridian now models an indirect brand pathway, so TV and YouTube can pick up the sales effect that travels through brand demand.

Here is the problem it solves. When brand demand is high, more people search your name and you spend more on brand-search ads at the same time. So those ads look great for reasons that have nothing to do with the ads. Tell the model to account for branded search and it stops handing that free credit to search.

But there is a catch. If a TV or YouTube campaign is what drove those searches, accounting for branded search also erases the effect you wanted to measure. Leave it out, search takes credit it did not earn. Put it in, brand media loses credit it did earn. The problem is real, and Meridian's setup is a reasonable attempt at it.

How does the new two-step model change things?

Meridian splits the work into two steps so branded search can do two jobs at once. The old model made you choose: strip branded search out, or leave it in. The two-step model does not make you choose. It treats branded search as the middle of the story.

First it estimates how much your brand media moved the brand-demand signal. Then it estimates how that signal, alongside the channels themselves, moved sales. That lets the model recover the indirect path, brand media to searches to sales, without forcing all of TV or YouTube's effect through search. The direct effects still count.

On paper, clean. The trouble is, a model can match your history perfectly and still not know what caused what. Matching the past is drawing a line through numbers you already have. Knowing what caused a sale means telling "the TV did it" apart from "it would have happened anyway." A model can nail the first and miss the second.

Google says as much in its own guidance: the two-step model has more effects to estimate and more places where a missing control can bias the answer (Google's docs). That is a reason to look hard at the result before you trust it.

What has to be true for the number to hold up?

The credit rests on a chain: brand media lifts searches, searches turn into sales, plus the direct effects that skip search. Full-funnel MMM solves one problem by creating a harder one. The old model only had to separate your marketing from everything else moving sales. Now it has to do that in two places at once.

Link one: did your brand media actually cause the extra searches? TV goes up, searches go up. But a launch, a promo, or a press hit could lift both at the same time. Brands run their biggest TV during their biggest launch, exactly when searches climb on their own. The model has to separate "the TV drove searches" from "the launch drove both," and account for every big cause like that (Meridian control variables).

Link two: do those searches really stand for the demand that drove sales? Same trap. Searches and sales can rise together because demand lifted both, or because a launch, promo, or price change did. The model has to pull the search effect apart from everything else moving sales.

Full-funnel MMM does not remove the causal assumptions. It multiplies the places they matter. Miss an important hidden cause at either stage and the ROAS can look precise and still be wrong about what drove what.

Can you trust Google's branded search number?

Partly. The common worry, that Google delivers it as an index instead of raw counts, does not hold up. For cleaning up your search numbers, rescaling it changes nothing (MMM Data Platform). You can also compare its movement, directionally, against independent demand signals like Search Console, Google Trends, direct traffic, or brand tracking. None is a perfect substitute, so treat disagreement as a reason to investigate, not a red flag on its own.

The real question, once the number is carrying credit, is whether it measures the right thing: does Google's count of "your brand" searches actually track the brand demand you want the model to represent? Google builds that list from the brand names you submit. So ask whether the definition captures your demand, whether it holds steady over time, and whether the series behaves like the real thing. Then test it: does the TV result survive if you swap in a different brand measure, like ad awareness or direct traffic? If it only shows up with Google's number in the middle, it is riding on Google's number, not your demand.

Does Google owning the number make it wrong?

No. Google sells YouTube and Search, and it supplies the number the model runs on. That does not make it wrong. It means you want evidence you did not get from Google before you move real budget. A conflict of interest earns a double-check on its own.

Google even says you can swap in a different brand measure. So if your ROAS only appears when Google's number carries it, you want to know that first. Same care you would bring to picking the measurement partner itself. PPC.land raised this when the tool launched, while most coverage just cheered (PPC.land).

Does a matching experiment prove it?

Not the part you most want proven. Run a geo experiment, hold TV back in some regions, and say it finds TV drove about $500K. The model lands near $500K too. That agreement is worth having. But it pins down TV's total effect, because you actually moved TV and watched sales. It does not trace the path, TV to branded searches to sales.

So the model can match the experiment's total and still be wrong about how the money traveled. TV could have worked through retargeting or direct response instead of brand demand. Use experiments to pin down the total (a real GeoX test). Just don't let "the experiment matched" become proof that every step of the story is right.

Six questions before you move a budget

Ask these of whoever runs your model, agency or in-house. Fuzzy answers are the answer.

1. Is our model the simple one-step kind or the new two-step kind? If two-step, walk me through what it assumes.
2. What rules out a launch or promo, not the brand media, causing the extra searches?
3. What rules out that same launch or promo causing the sales?
4. How much of each brand channel's ROAS is direct, and how much runs through branded search?
5. Does the result still hold with a different brand measure instead of Google's?
6. Does a real experiment agree with the total, and what is the range, not just the headline number?

These sit under the model, not inside it. A model can pass the usual fit and prediction checks and still fail to prove the causal story you are using to move budget.

Is a free model enough on its own?

The bGQV question is one version of a bigger one. Meridian is a capable model, and it is free, which is why so many advertisers reach for it. But it is only the model. Measurement is the system around it, and that is the part a free tool leaves to you.

The tool is capable. Meridian's 2026 release can ingest geo experiments and calibrate the model to them (Meridian docs). What the software does not do is the measurement work around it. Someone still has to design the experiments, decide which evidence counts, choose the controls and priors, chase down disagreements between methods, refresh the analysis as conditions change, and turn the output into budget decisions.

That is the work Stella automates and orchestrates. It runs always-on, recalibrating as new data and new experiments land instead of going stale between rebuilds. It keeps incrementality tests, surveys, and attribution feeding the model on a schedule, not as a one-off. It can push incremental signals back to the ad platforms so measurement sharpens bidding. And the results are queryable over MCP, so you can put them in front of an AI assistant like Claude and plan a budget against real numbers.

None of that replaces a good model. It is what turns a model into measurement you can act on, and it is the difference between a number that reads confident and one you can defend.

So should you use it, or not?

Use it. The problem it tackles is real, and branded search may be a good stand-in for demand. But once it becomes the bridge from your TV spend to your revenue, a marketing mix model owes you more than a clean fit: credible evidence at both stages, and a total that holds up to a real experiment.

Branded search being the middle of the story is a theory about how your money worked. It is not proof that it did.

So be clear about what you are asking bGQV to prove. An experiment can tell you whether TV drove incremental revenue. Full-funnel MMM can help explain where that effect traveled. Those are not the same claim. If the budget decision rides on the first, validate the first independently. Then use the model to understand the rest.

Book a demo

FAQ

What is Branded Google Query Volume (bGQV)?

It is the number of searches for your own brand name over time, delivered by Google as a data feed. You can use it two ways in a marketing mix model: to clean up your paid-search numbers, or, since Meridian's 2026 update, as a brand signal that lets TV and YouTube pick up the sales effect that runs through brand demand.

Is branded search volume a good input for an MMM?

As a way to clean up your search numbers, yes, when it matches the search you are measuring. As the signal that credits your TV and YouTube, the bar is higher, because the credit then depends on two separate links both being real, not just the model fitting well.

What is full-funnel MMM?

A model that measures brand building and performance together. Meridian does it in two steps: how much brand media moved branded searches, then how those searches, alongside the channels themselves, moved sales. Part of a channel's credit runs through that search path; the model still estimates the effects that do not.

Does an experiment that matches the model prove it?

No. If a geo experiment and the model agree on how much revenue TV drove, that gives you independent evidence for the total. It says nothing about how it happened. The model can be right about the size and wrong about the path.

What does a free MMM like Meridian leave you to handle?

The measurement work around it. Meridian's 2026 release can even ingest geo experiments and calibrate the model to them. But someone still has to design those experiments, choose the controls and priors, chase down disagreements, refresh the analysis as things change, and turn the output into budget calls. Stella automates and orchestrates that, running always-on, feeds signals back to the ad platforms, and lets an AI assistant query the results to plan against them. Free covers the model, not the system around it.

What should I ask my vendor about branded search in my MMM?

Whether the model is one-step or two-step, what rules out a promo or launch causing the searches, how much of the ROAS runs through branded search, whether the result survives a different brand measure, and whether a real experiment agrees with the total.