Over the past few months, a tool has been making the rounds in creator communities, marketing agencies, and social media consultancies with a hard-to-ignore promise: drop a video, and an AI model will simulate the cortical response of hundreds of human brains to predict, with scientific precision, whether that content will go viral.
The tool is called TRIBE v2, it's from Meta, and the pitch around it is seductive: "drop a clip, we simulate the cortical response of 720 modeled brains, second by second, and rate how hard it hits." An overall score from 0 to 100. Hook Score. Hold Rate. Visual cortex activation, auditory cortex, language network. A dashboard styled like a neuroscience lab, a brain image with areas highlighted in red, and a number that supposedly tells you whether your video has potential or not.
The problem is that I tested it. And the numbers don't hold up.
What Meta's Documentation Actually Says
Before trusting any score, I went straight to the source: the official Meta AI blog, where TRIBE v2 was announced in March 2026.
The research team's own description is unambiguous about what the model was built to do:
"This offers unprecedented speed, accuracy, and a 70x resolution increase as compared to similar models to predict how the brain responds to almost any sight or sound — enabling neuroscientists and clinical researchers to test theories without requiring human subjects."
Neuroscientists and clinical researchers. Not content creators. Not marketing agencies.
The model was trained on fMRI data from over 700 volunteers exposed to images, podcasts, videos, and text — and its stated objective is to predict brain activity in functional magnetic resonance imaging, to accelerate discoveries in neuroscience and the treatment of neurological conditions.
At no point does Meta claim that TRIBE v2 predicts virality, content performance, or social media engagement.
There is another detail that gets overlooked: the license is CC BY-NC — non-commercial use only. Any paid product built on TRIBE v2 without explicit Meta authorization is, at minimum, operating in a legal gray zone.
What has reached the market as a "virality prediction tool" is a third-party product that took a legitimate research model, repackaged it with a marketing narrative Meta never made, and started selling access on that premise.
The Tests: Viral Videos, Low Scores
The theory was easy to validate or disprove. I took videos that had already gone viral — known outcomes, real metrics — and ran them through the tool.
One test was with a 15-second Reel showing a cabin with the caption "See the cabin after 3 nights." The video has 87,600 likes and over 4,500 comments. Metrics any creator would recognize as genuine virality.
The tool's score: 44 out of 100. Hook Score of 27.
The dashboard flagged 69% "Focus Drift" — interpreted by the model as a negative signal. Hold Rate of 100%, the only number that matched reality: nobody skipped the video.
The model penalized the exact mechanism that made the video work. "See after 3 nights" is a classic open loop — the deliberately restrained opening creates narrative tension, and the viewer stays to the end wanting to see the result. Low sensory stimulation at the start is not a content failure; it is the mechanics of the reveal format.
A model measuring cortical activation does not capture this. It does not understand narrative, does not understand FOMO, does not understand the behavior of commenting "how much does it cost?" without even watching to the end. Those are the variables that drive sharing — and none of them appear in the dashboard.
The Structural Flaw in the Premise
Even setting aside the misappropriation of purpose, the central premise has a logical flaw that no technical refinement resolves: cortical activation is not sharing behavior.
Virality is a network phenomenon. It depends on the algorithm, publishing timing, community behavior, cultural context, and identity triggers. A video can generate low visual cortex activation and be shared millions of times because it touched on something people needed to pass on. Another can have a visually striking opening, a high score, and die at 200 views.
The training sample compounds the problem. The 700-plus fMRI study volunteers are demographically specific — people willing to lie inside an MRI machine for scientific research. That does not represent the audience of a Brazilian Reel about hospitality, food, humor, or any other niche with its own dynamics.
The real test of a prediction tool would be straightforward: take a thousand videos before publishing, run them through the tool, publish all of them, and verify whether the high-scoring ones went viral more often than the low-scoring ones. That study has not been conducted — at least not in any published, auditable form.
The Verdict
TRIBE v2 is a genuine scientific contribution. A model that simulates brain activity at high resolution, made openly available to researchers, with real potential to accelerate discoveries in neuroscience and neurological treatment. That has value.
What does not have value — at least not the value being sold — is using that model as a proxy for predicting virality. The leap from "we simulated cortical response" to "we know if your video will blow up" has no support in the documentation, no published validation, and empirical tests point in the opposite direction.
When a tool gives a score of 44 to a video with 87,000 likes, it is not being conservative. It is wrong on the most basic case.
Before incorporating any of these metrics into your creative process, ask the simple question: is there any published evidence that this score correlates with virality? If the answer is no — and here the answer is no — the dashboard is lab aesthetics, not applied science.
Written by
Paulo Victor Fraga