Google’s Meridian GeoX, its open-source tool for geographic incrementality experiments (testing whether ads caused outcomes in specific regions, not just coincided with them), moved from beta to general availability globally. The launch gives advertisers a structured, platform-agnostic way to measure causal ad impact and feed those findings back into a marketing mix model (MMM, a statistical model that estimates how different marketing channels contribute to business outcomes).
What Is Meridian GeoX and What Does It Actually Do?
Meridian GeoX lets marketers run experiments by dividing geographic areas into treatment and control groups. One region sees a campaign; another does not. The difference in outcomes estimates how much the ad investment actually drove results, rather than simply appearing alongside them. According to Search Engine Journal’s coverage of the launch, GeoX works across advertising platforms, so an advertiser can test a campaign running on a non-Google platform, or across several channels simultaneously.
Google first previewed GeoX in May. The global general availability release makes it part of the core Meridian open-source package, available to any team with the technical capacity to run it.
How Does Meridian GeoX Connect to Marketing Mix Modeling?
Meridian is Google’s open-source MMM tool. It uses historical data to estimate how different media investments contributed to outcomes. The limitation of pure historical modeling: when multiple channels run simultaneously, or when external factors shift market conditions, the model’s estimates become less precise.
GeoX addresses this by adding causal evidence. Per Search Engine Journal, marketers can incorporate incrementality results from GeoX experiments directly into their Meridian MMM, giving the model an additional, experiment-backed source of data. Google describes this as creating a closer connection between modeling and experimentation: Meridian provides the broad performance view while GeoX adds specific causal proof from individual tests.
Geographic testing also offers a practical alternative when user-level experiments are not feasible. Instead of tracking individuals across devices and platforms, the experiment operates at the regional level, which sidesteps attribution (the process of crediting a sale to a specific ad or channel) challenges tied to cookie deprecation and privacy changes.
What Else Is New in Meridian?
Google is rolling out three additional updates alongside the GeoX global launch, according to Search Engine Journal:
- Agentic model-building tools: New AI-assisted capabilities can audit data quality, help resolve errors, and guide marketers through the model-building process in real time. This reduces manual troubleshooting, particularly when data quality issues need to be diagnosed before an analysis can begin.
- Faster backend performance: Google is updating Meridian’s backend infrastructure to run analyses faster and more efficiently.
- Brand signal integration: Meridian models can now incorporate brand signals, including Branded Google Query Volume, to help account for the longer-term effects of brand investment. A TV campaign or out-of-home (OOH) placement, for example, may lift brand awareness well before that interest converts to a purchase.
On the data infrastructure side, Google’s blog.google announcement highlighted broader measurement stack updates running in parallel. Data Manager is now integrated directly into Google Analytics and Display and Video 360 (DV360). Advertisers who connect offline and app data to Data Manager see an average 26% increase in incremental ROAS (return on ad spend), according to Google. Enhanced conversions, now available in GA and DV360, deliver an average 11% increase in Search conversions compared to standard conversion imports. Advertisers using Google tag gateway observe an average 14% conversion uplift, with Demand Gen campaigns seeing over 20% uplift, per the same Google source.
What Does a Team Need to Run Meridian GeoX?
Meridian is open-source and installed via PyPI (Python’s public package repository). Per Google’s developer documentation, it requires Python 3.11 or 3.12 and at least one GPU (a graphics processing unit, used here for heavy statistical computation rather than rendering graphics). The model uses a sampling algorithm called No-U-Turn Sampler, or NUTS, which is compute-intensive by design. Google recommends running it on GPUs to reduce training time and get real-time optimization results.
This means Meridian GeoX is not a plug-and-play dashboard. Teams either need in-house data science capacity or a technical partner to build, validate, and interpret the models.
What This Means for AI-Search Visibility
The simplest way to understand the brand signal update: Google is now treating branded search volume as a modeling input, not just a reporting output. That shift matters beyond paid media.
Here is what that looks like in practice. A food brand runs a broad awareness campaign in Q3. Branded searches for that brand name rise first, before any uptick in direct purchases. Previously, most MMM setups had no input layer for that awareness-to-search step. Meridian can now model that lag using Branded Google Query Volume. The implication: brand investment gets credited more accurately in the model, rather than being absorbed as unexplained variance.
The same causal chain appears in AI-search visibility. When a brand is cited frequently in AI-generated answers, branded searches for that brand often lift in the weeks that follow, as consumers follow up on what an AI tool mentioned. That lift then flows downstream to conversions. The problem is that almost no MMM setup today includes an input signal for AI citation frequency, so the incremental contribution of appearing in ChatGPT, Gemini, or Perplexity answers is invisible to the model.
Hingewise’s read: the inclusion of Branded Google Query Volume as a Meridian signal is a quiet but meaningful acknowledgment that brand-level awareness, wherever it originates, is now measurable as a business input. Advertisers treating GEO (generative engine optimization, the practice of improving how often AI tools cite or recommend a brand) as a separate, unmeasurable experiment are missing the architecture that already exists to capture it. GeoX-style causal thinking, isolating a variable, measuring the incremental branded search lift, is a methodology that could plausibly be applied to test AI-driven brand awareness just as it tests paid media. The key insight is not that Meridian solves AI attribution directly: it is that the brand signal layer is now open, and teams that add AI visibility metrics to that layer will have measurement infrastructure that most competitors do not. That is an experiment worth designing.
Five Things to Check Before Building or Updating a Meridian Model
- Confirm your team has Python 3.11 or 3.12 and GPU access for compute-intensive NUTS sampling.
- Audit whether your media data covers all active channels, not just Google, since GeoX is explicitly platform-agnostic.
- Assess geographic feasibility: GeoX requires sufficient regional variation in media exposure to create meaningful treatment versus control comparisons.
- Decide which brand signals to include: Branded Google Query Volume is now a supported Meridian input and may improve model accuracy for awareness-led or brand-heavy campaigns.
- Plan for a data quality review before modeling begins: the new agentic tools assist with error identification, but human judgment on which business data belongs in the model remains essential.
What to Watch Next
Google has not specified a timeline for expanding the agentic model-building tools or clarified their performance boundaries for smaller teams without dedicated data science staff. How well those tools handle real advertiser data at scale will determine whether Meridian becomes a broader cross-industry measurement standard or remains a tool for enterprise and agency teams with significant technical capacity. The more consequential long-term question is how quickly the MMM field adopts non-paid brand signals, including AI citation activity, as standard modeling inputs alongside traditional media spend. That adoption curve is now a meaningful variable in how well measurement reflects actual business performance.
Sources: Search Engine Journal; blog.google; Google Developers, Meridian documentation. Published September 2026.
This article was drafted with AI assistance from the sources cited above and checked against them before publishing. The "What this means for AI-search visibility" section is Hingewise's editorial analysis, built on the frameworks and positions we use in our GEO work. How we work: Editorial policy.