Introduction
At TheDayAfterAI News, we run a systematic experiment: asking six leading AI chatbots to forecast the weekly price movement of individual US-listed stocks, then comparing their predictions against actual market outcomes. Our goal is not to validate or debunk any single platform, but to explore the differences in reasoning, data awareness, and analytical depth across today’s most prominent generative AI tools.
This edition focuses on GRAIL, Inc. (NASDAQ: GRAL) — a biotech company pioneering multi-cancer early detection (MCED) through its Galleri blood test. The prediction window covers Friday, February 20, 2026, through Thursday, February 26, 2026, comprising five US trading days (Friday February 20 and Monday–Thursday February 23–26).
What made this week extraordinary was a seismic event that unfolded on the evening of February 19, 2026: GRAL collapsed approximately 47% in after-hours trading following the disclosure that the landmark NHS-Galleri clinical trial had failed to meet its primary endpoint. For our experiment, this created a rare real-world stress test — which chatbots recognised the crash, and which ones were still analysing a world that no longer existed?
Background: What Happened to GRAL?
The NHS-Galleri Trial Miss
On February 19, 2026, GRAL reported Q4 2025 earnings that actually beat analyst consensus: non-GAAP EPS of −$2.44 against expectations of −$2.66 to −$4.01, and revenue of $43.6 million (up 14% year-over-year). However, the company simultaneously released top-line results from the NHS-Galleri trial, involving 142,000 participants screened over three years across England’s National Health Service. The trial’s primary endpoint — a statistically significant reduction in combined Stage III and Stage IV cancer diagnoses — was not met.
The market reaction was immediate and brutal. GRAL fell from its $101.53 closing price to as low as $52.25 in after-hours trading, a decline of approximately 48%. By the February 20 pre-market session, the stock was quoted around $53.65.
The Silver Lining Within the Data
Critically, the trial did produce meaningful secondary findings that tempered the severity of the selloff:
- A greater than 20% reduction in Stage IV diagnoses across 12 pre-specified cancers in rounds 2 and 3 of screening
- A fourfold improvement in overall cancer detection versus standard of care
- Substantial increases in early-stage (Stage I–II) cancer identification
- No serious safety concerns reported
Additionally, the company’s commercial fundamentals remained intact: $904 million in cash with a runway to 2030, 17% annual revenue growth (26% for US Galleri sales), a prescriber base that expanded 30% to 17,000 providers, and the Nancy Gardner Sewell Medicare MCED Screening Coverage Act — signed into federal law on February 3, 2026 — creating a Medicare coverage pathway from 2028.
The Partial Market Recovery
During the regular February 20 session, the stock mounted a significant intraday rebound, reaching the $72–$75 range — approximately a 36% recovery from the overnight low, driven by short covering, dip buying, and the market beginning to digest the trial’s secondary findings. It was against this volatile backdrop that we queried our AI platforms.
Methodology
For each prediction round, TheDayAfterAI News queries six AI chatbot platforms using an identical structured prompt, requesting a day-by-day price forecast for the coming trading week. The platforms tested in this edition were:
- Claude (Anthropic) — Research + Extended Thinking mode
- ChatGPT (OpenAI) — Extended Thinking mode (deep research mode was not used due to extremely slow response times)
- Perplexity AI — Deep Research mode
- Grok (xAI) — DeepSearch + Expert mode
- Copilot (Microsoft/Bing) — Think Deeper mode
- Gemini (Google) — Deep Research + Pro mode — FAILED to respond (three consecutive failures within the session window)
All queries were submitted on February 20, 2026, ahead of the US market open. Where available, each platform’s “research” or “deep thinking” mode was activated to elicit the most thorough analysis possible. The exception was ChatGPT, whose research mode was not used because it proved extremely slow to generate a response. Each chatbot was given access to its own web browsing or knowledge retrieval capabilities. The resulting outputs were saved as PDF and are archived for this article.
Gemini’s Deep Research mode failed to generate a response on three consecutive attempts — the first consecutive failures we have recorded for any platform in this series.
The Five Predictions: At a Glance
The table below summarises the key numerical outputs from each responding platform. All figures are those stated by the respective AI platform in its own analysis.
| AI Platform | Open Est. (Feb 20) | Close Est. (Feb 26) | Period Low | Period High | P(Increase) | Bias |
|---|---|---|---|---|---|---|
| Claude (Sonnet) | ~$70.00 | ~$73.00 | $62 | $80 | 58% | Mildly Bullish |
| ChatGPT | ~$55.00 | ~$60.00 | $47 | $68 | 54% | Mildly Bullish |
| Perplexity* | ~$101.50 | ~$106.00 | $92 | $115 | 55% | MISSED CRASH |
| Grok | ~$55.00 | ~$65.00 | $50 | $70 | 70% | Bullish |
| Copilot (Bing) | ~$100.00 | ~$92.00 | $80 | $112 | 28% | Bearish |
| Gemini (Deep Research) | N/A | N/A | N/A | N/A | N/A | FAILED (3x) |
| Average | $76.3 | $79.2 | $66 | $89 | 53% | Flat |
* Perplexity’s estimates are based on pre-crash price levels (~$101.50), indicating it did not incorporate the NHS-Galleri trial disclosure. These figures are not comparable to the other platforms’ post-crash forecasts.
Platform-by-Platform Analysis
1. Claude (Anthropic) — Post-Crash Inflection Thesis
Claude produced the most comprehensive post-crash analysis among all responding platforms. Fully aware of the NHS-Galleri trial results, Claude framed its forecast around what it termed a “post-crash asymmetric setup,” arguing that the magnitude of the 47% decline appeared disproportionate to the actual clinical damage given the encouraging secondary findings. Key elements of Claude’s reasoning:
- Predicted opening price of approximately $70.00 on Monday February 23, reflecting the intraday bounce from the $52.25 overnight low
- Period high of $80 (aligned with post-crash analyst targets from Baird and Canaccord) and period low of $62 (representing a retest risk)
- 58% probability of a net price increase over the period
- Elevated short interest of 18.9% of float (5.91 million shares, 4.54 days to cover) identified as a key technical support for continued bouncing
- NVIDIA earnings on Wednesday February 25 flagged as the single most important external variable for the week
Claude’s day-by-day forecast:
| Day | Open | Close | Key Driver | Tone |
|---|---|---|---|---|
| Mon Feb 23 | ~$70 | ~$72 | Weekend digestion of analyst reports; short covering continues; light economic calendar supports stability. | Recovering |
| Tue Feb 24 | ~$72 | ~$74 | Consumer Confidence data; additional analyst notes expected; short-covering momentum pushes toward $75 resistance zone. | Slight Gain |
| Wed Feb 25 | ~$74 | ~$71 | Pre-NVIDIA caution introduces hesitation; profit-taking after the $53-to-$74 bounce; healthy consolidation pullback. | Pullback |
| Thu Feb 26 | ~$71 | ~$73 | Post-NVIDIA reaction; Natera earnings provide peer context; MSCI rebalance flows; stock stabilises in low-$70s. | Stabilise |
Notable omission: Claude’s day-by-day forecast begins on Monday February 23 and does not include a prediction for Friday February 20 — the first trading day of the five-day window. Given that February 20 was the immediate post-crash session (with the stock opening near $53 and rebounding to the $72–$75 range), this was arguably the single most volatile and consequential day of the period. Its absence from the day-by-day table represents a gap in Claude’s otherwise comprehensive analysis.
Claude’s conclusion was balanced: the stock’s crash to $53 priced in something closer to a terminal scenario, but the rapid bounce to $70+ on February 20 confirmed the market was recalibrating. The $62–$80 range explicitly acknowledged wide uncertainty following a crash of this scale.
2. ChatGPT (OpenAI) — Cautious Recovery, Conservative Range
Opening Prediction: ~$55 | Closing Prediction: ~$60 | Bullish Probability: 54%
ChatGPT also correctly recognised the post-crash environment, anchoring its forecast to a February 20 opening of approximately $55 — consistent with the pre-market level following the 47% after-hours decline.
- Predicted open (February 20): ~$55
- Predicted close (February 26): ~$60
- Period range: $47–$68
- Probability of net increase: 54%
ChatGPT correctly noted that February 20 was the monthly options expiration date, flagging it as a potential amplifier of intraday volatility and “pinning” behaviour around strikes. Its day-by-day path was more cautious than Claude’s, with the stock expected to remain range-bound between $52 and $68 throughout the week before modestly settling higher.
ChatGPT also transparently acknowledged the limitations of its real-time data access and invited the user to share the actual pre-market price to allow for a tighter estimate.
3. Perplexity AI — Critical Data Gap: The Missed Crash
Opening Prediction: ~$101.50 | Closing Prediction: ~$106.00 | Bullish Probability: 55%
Perplexity’s analysis stands apart from all other platforms for a fundamental reason: it appears to have been conducted without knowledge of the NHS-Galleri trial results or the 47% after-hours crash. Perplexity anchored its forecast to a pre-market price of approximately $101.50 — the pre-crash level. Its entire analysis proceeded as though no material event had occurred:
- Predicted open (February 20): ~$101.50
- Predicted close (February 26): ~$106.00
- Period range: $92–$115
- Probability of net increase: 55%
The platform identified a “slightly bullish week with a wide trading range” — a perfectly reasonable assessment of normal conditions for GRAL, but one that bore no relevance to the extraordinary week that actually unfolded. Perplexity did correctly identify the stock’s high beta and elevated short-term volatility, but it had already missed the biggest catalyst of the year.
This outcome raises an important question about real-time web access and information recency across AI platforms. Even with web search enabled, Perplexity did not surface the February 19 after-hours trial disclosure before being queried on the morning of February 20.
4. Grok (xAI) — The Most Bullish Post-Crash View
Opening Prediction: ~$55 | Closing Prediction: ~$65 | Bullish Probability: 70%
Grok correctly processed the post-crash environment and delivered the most optimistic forecast among the responding platforms:
- Predicted open (February 20): ~$55
- Predicted close (February 26): ~$65
- Period range: $50–$70
- Probability of net increase: 70%
Grok’s 70% bullish probability was the highest across all platforms. Its reasoning centred on the argument that the ~50% drop was an overreaction given the positive earnings beat, strong US Galleri commercial growth (31% year-over-year), and encouraging secondary trial findings. It cited the active FDA PMA submission, 2026 revenue guidance of 22–32% growth, and upcoming detailed trial data presentation at ASCO mid-2026 as near-term catalysts.
Grok also flagged the February 20 monthly options expiration as a potential volatility amplifier.
5. Copilot (Microsoft/Bing) — The Contrarian Bearish Case
Opening Prediction: ~$100 | Closing Prediction: ~$92 | Bearish Probability: 62%
Copilot was the sole platform to deliver a meaningfully bearish forecast for the period. However, its analysis contains a key caveat: Copilot appeared to anchor to a pre-market price near $100 rather than the ~$55 pre-open level, suggesting partial or incomplete incorporation of the crash’s full magnitude.
- Predicted open (February 20): ~$100
- Predicted close (February 26): ~$92
- Period range: $80–$112
- Probability of net increase: 28% (i.e., 62% probability of net decrease)
Copilot’s bearish thesis was driven by the immediate post-earnings volatility and what it interpreted as elevated options activity skewed toward puts, suggesting downside pressure. The platform noted that technical momentum indicators were likely to flip bearish if the stock failed to reclaim post-open highs, and that sector rotation out of healthcare/biotech represented a headwind.
One important distinction from Perplexity’s miss: Copilot did reference the Q4 earnings release and elevated after-hours activity, indicating awareness of the catalyst event. However, its opening price assumption of ~$100 suggests it did not fully resolve the magnitude of the decline. Its bearish directional call remains the most distinctive outlier position across all responding platforms.
6. Gemini (Google Deep Research) — Three Consecutive Failures
On this occasion, Gemini’s Deep Research mode failed to produce a response on three consecutive attempts within the pre-market query window. This marks an unprecedented event in our series — the first time any platform has failed three consecutive times within a single session.
February 20, 2026, represented an exceptionally high-demand period for AI research tools, given the significance of the GRAL event and the concurrent activity surrounding NVIDIA’s upcoming earnings. Whether the failures were attributable to server load, a timeout in the deep research pipeline, or content-related restrictions is unknown. Gemini has previously provided detailed analyses in this series, and we will retest in the following week’s edition.
Key Observations Across Platforms
Real-Time Awareness Varied Dramatically
The most significant differentiator this week was not analytical sophistication but data recency. Claude, ChatGPT, and Grok all correctly identified the post-crash starting point near $53–$55. Perplexity did not — it produced an entirely different analysis for a different stock. Copilot appeared partially aware but mis-anchored its opening price. This disparity underscores that web search access alone does not guarantee information recency in fast-moving news cycles.
Directional Consensus Was Bullish Among Informed Platforms
Among the three platforms that correctly identified the crash (Claude, ChatGPT, Grok), all three held a mildly-to-strongly bullish view, with probabilities of a net price increase ranging from 54% to 70%. The shared rationale included elevated short interest creating covering pressure, analysts maintaining Buy ratings with targets at $80–$82, and the intact commercial fundamentals despite the clinical setback.
Copilot Offered the Only Bearish Conviction
With a 62% probability of a net price decrease, Copilot’s forecast diverged sharply from the other post-crash analysts. Whether this reflects a more sophisticated interpretation of options market dynamics or a data anchoring artefact remains an open question — one that the actual week’s price action will resolve.
Analytical Depth Differed Significantly
Claude and Grok produced the most comprehensive analyses, incorporating day-by-day price paths, specific macro catalysts (NVIDIA earnings, Consumer Confidence data, MSCI rebalance), short interest mechanics, and sector rotation context. ChatGPT was well-structured but shorter, explicitly acknowledging its data limitations. Copilot provided detailed technical metrics. Perplexity’s analysis, while internally thorough, was rendered irrelevant by its data gap.
The NVIDIA Wildcard
Multiple platforms independently identified NVIDIA’s earnings release on Wednesday February 25 as a significant external risk. A positive NVIDIA print was expected to lift broader risk appetite and provide tailwinds for growth and innovation names including genomics stocks; a miss could create the reverse effect. This macro dependency illustrates how individual stock forecasting cannot be fully decoupled from mega-cap technology earnings in the current market environment.
Methodology
Each AI chatbot was given an identical prompt requesting a five-day stock price forecast. The models used their own web-search and data-retrieval capabilities; no proprietary data was provided. Responses were collected without modification. Variations in depth, format, and analytical approach reflect each platform’s native capabilities.






















