How Soma compares

Predicted from the file. Not a panel, a webcam, or an LLM guessing.

Everyone else measures reactions from recruited humans (panels, webcams, surveys) or simulates them by prompting an LLM to role-play a person. Soma predicts the actual cortical response from the video file, and is the only one that draws a visible line between what is validated and what is still a labeled hypothesis.

the real difference · where the signal comes from

Same output shape. Different source.

Per-second attention curves and weak-spot callouts already exist. The question that matters is where the signal underneath them comes from. That is the whole comparison.

panel · neuro
Webcam & eye-tracking
Recruited viewers, watched by a camera. Realeyes, Neurons.
panel · survey
Surveys & ratings
Recruited panels rate the ad, tied to norms. System1, Nielsen.
synthetic
LLM personas
An LLM role-plays a consumer and self-reports. Aaru, Simile.
predicted biology
Brain response, from the file
A model trained on real fMRI predicts the cortical response. Soma.
at a glance

The comparison in one table.

  Human panelsRealeyes · Neurons · System1 SyntheticAaru · Simile VidCognitionsame model Soma
Signal source faces, gaze, self-report LLM guess predicted brain predicted brain
Needs recruited people yes, per study none none none
Result in minutes days
Cheap enough for every cut
Grounded in biology ~a proxy
Shows validated vs hypothesis claims it as fact on every claim

Honest framing: the panel column is a real human reaction, which we don't have, and we don't claim to be more accurate than a panel. We compete on source, speed, cost, and honesty. Figures for named competitors are from their own material; verify before quoting.

head to head

Where each one is strong, and how Soma differs.

vs Realeyes

webcam facial-coding + eye-tracking

Where they're strong

Real human reactions, not predicted. An established brand with agency trust and years of normative data.

How Soma differs

We predict from the file. No webcam, no recruited panel, no scheduling, in minutes at a fraction of the cost, so you test every variant. Their read is a facial proxy for an internal state; ours is the predicted cortical state, with the evidence tier on every claim.

Why won't they build it? A public, from-the-file model undercuts their own pitch of recruiting a panel and charging per study. It is a business-model conflict, not an engineering gap.

vs Neurons Inc

AI attention / gaze prediction

Where they're strong

The closest shipping product to ours, polished, with a real per-frame attention model that is mature in its domain. Established with enterprise marketing teams.

How Soma differs

Neurons predicts where the eye goes. Soma predicts the full cortical response, attention and a coarse affect read, from a brain-encoding model. We are priced for performance teams, not five-figure enterprise contracts, and we carry the honesty boundary they don't.

Honest caveat: their attention model is mature; ours is validating, not validated. We differ on breadth of signal, price, and honesty, not on an accuracy claim we haven't earned. Pricing figure (about €15,000/year for five seats) is from a third-party page; verify before quoting.

vs System1 & Nielsen

panel emotion + normative databases

Where they're strong

Outcome-linked norms and credibility with brand marketers who care about long-term brand building.

How Soma differs

Panel-based, per-study, slow, and built for a handful of hero spots at brand budgets. Soma is the high-volume, low-cost, from-the-file read for teams shipping dozens of paid-social videos a week, the segment their model is too slow and pricey to serve.

vs VidCognition

same public TRIBE v2 model

Where they're strong

Built on the same public encoder we use, with a clean creator funnel and a good plain-English way of describing patterns. On the model itself, we are even; neither of us owns it.

How Soma differs

Three real forks. Honesty: they present the activation-to-engagement step as settled fact while admitting they run no validation of their own; we show the tier and run the held-out test. Buyer: they sell creators a hook score; we sell performance teams a decision instrument tied to real media spend and a data flywheel. Disclosure: we volunteer the negative prior; they quote only the flattering number.

Since the model is shared, the race is honest validation and real outcome data. Whoever earns those first wins. Verify their current claims and pricing before citing them by name in public.

vs Aaru & Simile

synthetic research, LLM personas

Where they're strong

Genuinely fast and cheap, and flexible enough to answer any question you can phrase, not only second-by-second attention.

How Soma differs

No biology. They predict what a person might say, and LLM personas lean toward agreeable, plausible answers. We predict the actual neural response from a model trained on real fMRI, reproducible run to run.

Honest caveat: it is partly a different job. They run broad simulated surveys; we read a second-by-second neural arc on video. The contrast is grounded versus ungrounded for the video-reaction question, not strictly either-or.
where we hold the line

What Soma does not claim.

A comparison page that only lists strengths is a sales sheet. Here is what we refuse to say, on purpose, because our whole edge is being trusted by a reader who checks.

See the demo Read the science Request early access