H2 DATA POPULATED
GEO Olympics Initiative · Temporal Velocity
GEO Olympics Initiative · Temporal Velocity
Hypothesis 2 | Temporal Velocity | The Vonn Crash Test
How Fast Does AI Learn?
Lindsey Vonn crashed during training in Crans-Montana on January 30, 2026 rupturing her ACL one week before the Olympics. We asked the same 11 questions to 7 AI platforms and measured which ones absorbed the new reality.
The results reveal a two-speed AI information ecosystem.
What's happening in the test of the Olympic games tests?
H2: Temporal Velocity Hypothesis
LLMs update their knowledge at measurably different speeds, creating windows of competitive advantage.
Our prediction: Search-augmented platforms (AI Overviews, Google AI Mode, Perplexity) update within hours. Conversational LLMs (ChatGPT, Meta AI, Gemini) update within days.
361
Total responses scored
7
AI platforms tested
11
Unique prompts
13
Days of data (Jan 25 – Feb 6)
Known Events
The Timeline We Tested Against
Every timestamp below is verified. The question was: how quickly did each AI platform figure it out?
Jan 10, 2026
Vonn wins World Cup Downhill
First Downhill victory since the comeback. Shifted LLM framing from "feel-good story" to "medal contender."
Jan 25–29, 2026
Wave 1 baseline collected
78 pre-crash responses across 7 LLMs. All scored 0 on crash awareness. This is our control.
Jan 30, 2026 · T=0
Training crash in Crans-Montana
Vonn crashes during a World Cup downhill. Ruptured ACL in left knee, bone bruising, meniscal damage. Airlifted from the course. This is our ground-truth event — the clock starts here.
Jan 30 – Feb 6, 2026 · T+0 to T+7
283 post-crash responses scored
Re-ran all 11 prompts across 7 LLMs at multiple intervals. Scored each response on the 0–3 crash awareness rubric. Five platforms integrated within 24 hours. Two never did.
Feb 11, 2026
Women's Downhill
Race day. Post-event prompts (#8–11) will serve as a second temporal velocity test measuring race result integration.
The Velocity Curve
Average Crash Awareness Score by Date
Each point is the average 0–3 crash awareness score across all prompts for that LLM on that date. The vertical line marks January 30 — the crash. Watch the split.
Post-Crash Average
Crash Awareness Score by Platform
Average score across all post-January 30 responses. Scale: 0 (no awareness) to 3 (full integration with accurate details).
0
No awareness1
Vague awareness2
Partial integration3
Full integrationAI Mode
2.80
ChatGPT
2.64
Gemini
2.51
Perplexity
2.37
AI Overviews
2.26
Claude
0.08
Meta AI
0.04
What We Found
Five Findings from 361 Responses
The data revealed a pattern more extreme than we predicted. It's not a speed gradient — it's a binary cliff.
Finding 01
The Two-Speed Information Ecosystem Is Real — But the Split Is Different Than Predicted
We predicted search-augmented platforms would be faster than conversational LLMs. The actual split is web-connected vs. not web-connected. Perplexity, Google AI Mode, AI Overviews, Gemini, and ChatGPT all integrated the crash within 24 hours (avg 2.26–2.80). Claude and Meta AI never integrated it at all (avg 0.04–0.08) through Feb 6 — a full week post-crash. The gap isn't days. It's infinite.
Finding 02
Same-Day Integration Is the New Normal
On January 30 itself — the day of the crash — Perplexity (2.0), Google AI Mode (1.5), AI Overviews (1.5), and Gemini (1.2) all showed partial awareness. By January 31, all four were scoring 2.6–2.9. The "window of competitive advantage" for web-connected platforms is hours, not days.
Finding 03
Meta AI Doesn't Know She Un-Retired
Meta AI scored 0 on every single post-crash response — but not because it missed the crash. It doesn't even know Vonn made a comeback. Multiple responses state "she retired in 2019" and "she will not be competing." Meta AI is living in 2023, not 2026.
Finding 04
Claude Knows the Comeback, Missed the Crash
Claude represents a middle layer: it knows about Vonn's comeback and can discuss her knee replacement surgery, but has zero awareness of the January 30 crash. It's providing thoughtful, well-reasoned analysis of a reality that no longer exists. Avg post-crash score: 0.08.
Finding 05
Medical Prompts Are the Hardest to Update
"Recovery routine after knee surgery?" was the lowest-scoring prompt even for web-connected platforms. Perplexity (0), AI Overviews (0), Gemini (1) — these platforms pivot to historical knee surgery information rather than the fresh ACL injury. Prompt framing determines whether an LLM surfaces old or new information, even when it has access to both.
H2 Validation
Hypothesis Assessment
H2: PARTIALLY SUPPORTED — WITH A STRONGER FINDING
The speed difference exists, but it's a cliff, not a gradient.
We predicted a measurable speed difference between search-augmented and conversational platforms. What we found is more dramatic: five platforms integrated within 24 hours; two platforms showed zero integration after 7 days. The competitive advantage window isn't hours or days — for non-web-connected platforms, it's permanent until the model is retrained.
The original grouping (search-augmented vs. conversational) was wrong. ChatGPT and Gemini, which we classified as conversational, both have web search capability and integrated as fast as Perplexity. The actual dividing line is web access, not platform architecture.
Caveat: ChatGPT had only 19 responses (vs. 49–63 for other platforms), limiting statistical confidence for that platform's scores. Gemini pre-crash responses reference historical ACL injuries in career context — these were correctly scored 0 as they don't reflect awareness of the 2026 crash.
The original grouping (search-augmented vs. conversational) was wrong. ChatGPT and Gemini, which we classified as conversational, both have web search capability and integrated as fast as Perplexity. The actual dividing line is web access, not platform architecture.
Caveat: ChatGPT had only 19 responses (vs. 49–63 for other platforms), limiting statistical confidence for that platform's scores. Gemini pre-crash responses reference historical ACL injuries in career context — these were correctly scored 0 as they don't reflect awareness of the 2026 crash.
Strategic Implications
What This Tells You About Showing Up in AI Responses
The Human Layer Turns This Signal Into Strategy
If your brand experiences a product recall, leadership change, earnings surprise, or any rapid narrative shift, the AI platforms your customers use will tell different versions of reality for different lengths of time.
For crisis response: Monitor Perplexity and Google AI Mode first — they'll reflect new information within hours. Claude and Meta AI may continue amplifying the pre-crisis narrative for weeks or months.
For competitive intelligence: If you publish a correction, update, or new positioning, check whether the AI platforms that matter most to your audience have picked it up. A press release that reaches Perplexity in 4 hours may not reach Claude's training data for 6 months.
For content strategy: The "recovery routine" finding shows that even web-connected platforms can miss the new information if the prompt triggers historical context. How you frame your content determines whether AI surfaces the old story or the new one — even when it has access to both.
For crisis response: Monitor Perplexity and Google AI Mode first — they'll reflect new information within hours. Claude and Meta AI may continue amplifying the pre-crisis narrative for weeks or months.
For competitive intelligence: If you publish a correction, update, or new positioning, check whether the AI platforms that matter most to your audience have picked it up. A press release that reaches Perplexity in 4 hours may not reach Claude's training data for 6 months.
For content strategy: The "recovery routine" finding shows that even web-connected platforms can miss the new information if the prompt triggers historical context. How you frame your content determines whether AI surfaces the old story or the new one — even when it has access to both.
Methodology
Data Considerations
Data Completeness
361 total responses scored across 7 platforms, 11 prompts, and 13 collection dates (Jan 25 – Feb 6, 2026).
Uneven platform coverage: Google Gemini (63 responses), Perplexity (61), Claude (60), Meta AI (60), Google AI Mode (59), AI Overviews (39), ChatGPT (19). ChatGPT's limited sample size means its 2.64 average carries wider confidence intervals than other platforms.
AI Overviews gaps: No responses were collected for 2 of the 11 prompts ("Competing in all alpine events" and "Medal in downhill at Milan-Cortina 2026").
Date gaps: Not all platforms have responses for all dates. Feb 2–3 have no data for any platform. ChatGPT data exists only for Jan 25–26, Jan 29, and Jan 31.
Uneven platform coverage: Google Gemini (63 responses), Perplexity (61), Claude (60), Meta AI (60), Google AI Mode (59), AI Overviews (39), ChatGPT (19). ChatGPT's limited sample size means its 2.64 average carries wider confidence intervals than other platforms.
AI Overviews gaps: No responses were collected for 2 of the 11 prompts ("Competing in all alpine events" and "Medal in downhill at Milan-Cortina 2026").
Date gaps: Not all platforms have responses for all dates. Feb 2–3 have no data for any platform. ChatGPT data exists only for Jan 25–26, Jan 29, and Jan 31.
Scoring Methodology
0–3 Crash Awareness Rubric: Each response scored on a single dimension — does this AI know about the January 30 crash?
0 = No awareness. Response reflects pre-crash information only. Includes responses stuck at "she retired in 2019."
1 = Vague awareness. Generic "injury concerns" or "recent setback" without crash specifics.
2 = Partial integration. References crash, ACL, or new injury but incomplete details (e.g., no mention of Crans-Montana, wrong date).
3 = Full integration. Accurately describes the crash with specific details: ruptured ACL, left knee, Crans-Montana/Switzerland, January 30/late January timing.
Historical injury filtering: References to Vonn's career ACL tears (2013, etc.) were excluded from crash awareness scoring. Only references to the specific 2026 crash event count toward the score.
Pre-crash sanity check: All 78 pre-crash responses (before Jan 30) scored 0, confirming the rubric correctly distinguishes historical injury discussion from 2026 crash awareness.
0 = No awareness. Response reflects pre-crash information only. Includes responses stuck at "she retired in 2019."
1 = Vague awareness. Generic "injury concerns" or "recent setback" without crash specifics.
2 = Partial integration. References crash, ACL, or new injury but incomplete details (e.g., no mention of Crans-Montana, wrong date).
3 = Full integration. Accurately describes the crash with specific details: ruptured ACL, left knee, Crans-Montana/Switzerland, January 30/late January timing.
Historical injury filtering: References to Vonn's career ACL tears (2013, etc.) were excluded from crash awareness scoring. Only references to the specific 2026 crash event count toward the score.
Pre-crash sanity check: All 78 pre-crash responses (before Jan 30) scored 0, confirming the rubric correctly distinguishes historical injury discussion from 2026 crash awareness.
Post-Event Prompt Note
Prompts 8–11 ("What happened to Vonn's comeback?", "Did her comeback succeed?", etc.) were originally designed for post-Feb 11 collection. However, Scrunch collected responses to these prompts during the pre-event window as well.
Key insight: Web-connected platforms treated these as current-status questions and provided crash-aware responses. Claude and Meta AI treated them as future-event questions and either said "I don't have information" (Meta) or provided pre-crash analysis (Claude). This itself is a temporal velocity signal — the prompt assumes a reality that only some platforms can access.
Key insight: Web-connected platforms treated these as current-status questions and provided crash-aware responses. Claude and Meta AI treated them as future-event questions and either said "I don't have information" (Meta) or provided pre-crash analysis (Claude). This itself is a temporal velocity signal — the prompt assumes a reality that only some platforms can access.
Seer Interactive GEO Olympics Initiative · H2 Temporal Velocity
See what AI thinks before AI shapes what everyone else thinks.