What Meta's brain prediction model TRIBE v2 tells advertisers

FAIR, Meta's fundamental AI research team, announced TRIBE v2 on 26 March 2026: a model that predicts what happens in the brain of someone watching a video, listening to a sound or reading a text. It comes with a demo site anyone can try in the browser; upload a video and a colour map shows which brain regions are expected to respond more strongly.

The visuals are striking and the model is often shared under headlines such as "AI that measures an ad's effect on the brain". This piece separates what the model actually does, what it does not do, and what it means for marketing teams.

What exactly does TRIBE v2 do?

TRIBE v2 is a prediction model: it takes a video, audio or text as input and predicts the activity an fMRI scanner would measure in the brain of an average person exposed to that content. According to Meta, it builds on the architecture that won an award in the Algonauts 2025 challenge; that model was trained on low-resolution recordings of four people, while the new version scales up to more than 700 volunteers.

According to the research paper, the training data consists of more than 1,000 hours of fMRI recordings from 720 participants. To understand content, the model relies on existing AI models: V-JEPA 2 for images, Wav2Vec-BERT for audio and Llama 3.2 for language. The paper was published on arXiv on 5 May 2026, and Meta says the model can also make predictions for people, languages and tasks it has not seen before.

Does it read the brain or decode thoughts?

No. TRIBE v2 does not read signals from the brain to work out what you are thinking; it works in the opposite direction and predicts the response a piece of content is expected to trigger. It does not measure a real person's brain; it produces a prediction based on the average it learned from the volunteers in its training data.

There is another limit: fMRI measures changes in blood flow in the brain over a matter of seconds. It does not directly capture sub-second responses, wandering attention or the influence of personal history. The paper itself notes that the model remains limited by the spatial and temporal resolution of fMRI.

Can brands use TRIBE v2 to test ads?

Not directly. The code and model weights are open on GitHub and Hugging Face, but the licence is CC BY-NC 4.0: research and non-commercial use are allowed, commercial use is not. An agency or brand using the model to score creative for a client campaign would breach the licence.

Beyond the licence there is a gap. A stronger predicted brain response does not mean an ad will sell, be remembered or make people like the brand. No study has yet established that link; the model was designed as a neuroscience tool, not to forecast campaign success.

So why does it matter for marketing?

Because cost has been the biggest barrier to neuromarketing: measuring brain responses required a lab, equipment and participants. Prediction models open a path to moving part of that measurement into software. If an approach that is limited to a research licence today turns into commercial tools, it could change how creative testing is done.

In the short term the main effect is on expectations: more tools and services will arrive promising "ads tested with brain prediction". Judging those promises requires knowing what the model measures and what it does not.

What should brands and agencies do?

• Check the licence: before using a TRIBE v2-based tool in a campaign, ask in writing whether commercial use is permitted.
• Do not confuse brain prediction with sales data: base creative decisions on real campaign data (clicks, conversions, sales), not on a prediction map.
• Keep A/B testing: comparing creatives with a real audience, equal budgets and a success metric set in advance is still the most reliable method.
• Use the demo for internal training: seeing how images, sound and text activate different brain regions can help a creative team build a shared language.
• Protect the personal data boundary: if you plan research that collects brain or biometric data from customers, review the rules on sensitive personal data with your lawyer first.

For a measurement set-up that tests your creatives on real campaign data and shifts budget to the winning variation, see our approach to CRO and analytics and performance marketing.