Early Lock
Based purely on AI visibility signals (SoP, SSI, PDI, sentiment, media velocity).
Forecast: $143M
AI visibility stabilized ~12 days earlier than tracking models.
10 Days Before Results
Wicked: For Good
Emberos conducted a first-of-its-kind test to evaluate whether AI visibility, what LLMs surface in response to user prompts, could forecast theatrical revenue.
We privately froze our forecast on November 14 using a timestamped, hashed protocol. After box-office results were published, the model landed within 0.8% of actual opening-weekend gross. This case study outlines the signals, methodology, findings, and opportunities for future optimization.
This study tested a simple question:
Can AI visibility, what LLMs surface in response to user prompts, forecast real-world behavior before search, social, or presales move?
Two Lock Forecasting
Lock #1 sets the early prediction; Lock #2 validates after reviews.
Nov 14
Nov 20
Nov 21
Reviews Lift
Track changes in SoP, sentiment, PDI, and media references post-embargo.
Capture pre- vs. post-review movement to measure convergence.
Opening Day Update
Final Lock
Uses only AI visibility + public review data (no box-office numbers).
Shows convergence or variance relative to Lock #1.
Confirms whether early AI signals held.
Verified Results
Domestic Opening Forecast
$114.2M → $114.0M
Model stabilized early; variance tightened post-reviews
SOP Lift
+0.2%
AI visibility held steady through review window
Early Signal Lead Time
~14 Days
Inflection detected before tracking shift
Visibility-Driven Uplift
$10M
Estimated incremental demand captured via AI visibility
Forecast Accuracy (Backtested)
4.6% MAE
across 300 titles
2-for-2
validation for 2 additional titles released
with mean 7% error
±$10M CI
on domestic weekend
AI Visibility Trends — Early Signal Advantage
Wicked: For Good’s Share-of-Prompt curve stabilized by Nov 14, demonstrating the model's ability to detect the inflection 10–12 days earlier than traditional tracking.
Performance Revenue & Impact
AI visibility movements correlated with opening-weekend demand ahead of reviews and pre-sales, supporting the viability of LLM-based forecasting.
Observed vs Predicted (Domestic Weekend)
0.8% Error
Signal Strength vs Forecast Adjustment
Negative Signal Movement
Positive Signal Movement
AI visibility surged well before traditional indicators updated, establishing SoP as a forward-looking demand signal.
The model locked onto its final range ~12 days earlier, and reviews confirmed, not reshaped, the forecast.
Traditional tracking sees what already happened.
AI visibility showed what was about to happen.
Simulation runs show that a +2–3 point lift in SoP during the final ten-day window could have translated to $2–3M in incremental opening-weekend revenue—opportunities future marketing teams could unlock using emberos' predictive optimization loop. This represents the next phase of GEO: moving outcomes, not just predicting them.
Lead Time Advantage
~14 Days
Forecast Error
0.2%
"When AI systems start surfacing a title, audiences start showing up."
Justin Inman Founder & CEO
Why This Matters
AI visibility has become the earliest measurable signal of audience intent. For Wicked: For Good, AI systems began surfacing Wicked content 10–12 days before traditional tracking curves moved. This marks one of the first cases where LLM reasoning stabilized the forecast ahead of social velocity, ticketing data, or critic reviews.
Top Query Themes Leading Up to Release
- Cast & character questions
- Music & song queries
- Comparisons to other holiday tentpoles
- Family/holiday positioning
- Franchise continuity ("Do you need to see Part 1 first?")
Methodology
AI Monitoring
Scout
Track SoP, SSI, PDI, CCI across AI platforms
Predictive lift engine
Pilot
Multi-model prediction with confidence intervals
Predictive Optimization
Flow
Real-time feedback loop for strategic adjustments
Resources
Supporting materials
Legal Notice: "Wicked," "Wicked: For Good," and all related titles, logos, characters, and artwork are trademarks of Universal Pictures. All references are used for commentary and analytical purposes. Emberos is an independent analytics platform and is not affiliated with or endorsed by Universal Pictures