Overview
The--analyze flag enables Grok-X to use its code_interpreter tool to perform quantitative analysis on X data. This allows for statistical analysis, trend identification, and data aggregation beyond what text summarization alone can provide.
Use Case
Data analysis is ideal for:- Quantifying trends (post volume, engagement patterns, sentiment shifts)
- Identifying peak activity periods
- Aggregating mentions by company, topic, or geography
- Performing statistical analysis on text data from X
Example: Tech Layoff Announcements
Analyze tech layoff announcements with quantitative data processing:--analyze— Enable code execution for quantitative analysis--from 2026-02-01— Start date for the analysis window
Output
Grok-X searches X, extracts relevant posts, and performs quantitative analysis:Tech Layoff Activity — February 2026 Analysis Grok searched X for layoff announcements and ran quantitative analysis via code interpreter across 847 relevant posts from Feb 1–Mar 4, 2026. By company (posts mentioning layoffs):Posting velocity: Layoff-related posts peaked on Feb 14 (Valentine’s Day — 203 posts in 24h) and again on Feb 28 (end-of-month HR cycles — 178 posts). Sentiment in layoff posts: 61% negative/critical, 24% neutral/informational, 15% supportive of company rationale. The supportive cluster skews toward accounts in finance and investor relations. Geographic signal: 71% of geo-tagged posts originated from US accounts, 14% from India (Salesforce and Workday have large India engineering hubs), 8% from UK. Most amplified single post: An Intel engineer’s firsthand account of receiving a severance email with 48 hours notice — 94K likes, 41K reposts, 6.2K replies. Conclusion: February 2026 is tracking ~18% above the rolling 6-month average for layoff post volume. The Intel and Salesforce announcements are dominating signal. No signs of a sectoral recovery narrative emerging yet.
Company Posts Estimated headcount reduction Salesforce 143 ~1,200 Intel 98 ~2,800 Snap 67 ~400 Twilio 54 ~350 Workday 41 ~600
What Code Interpreter Enables
When--analyze is used, Grok-X can:
Aggregate Data
- Count posts by company, topic, or keyword
- Group posts by time period (hourly, daily, weekly)
- Extract entities and count their mentions
Statistical Analysis
- Calculate averages, medians, and percentiles
- Identify outliers and anomalies
- Compute sentiment distributions
- Analyze engagement metrics (likes, reposts, replies)
Pattern Recognition
- Detect posting velocity spikes
- Identify temporal patterns (day-of-week, time-of-day)
- Cluster similar content or accounts
- Track geographic distributions
Comparative Analysis
- Compare current period to historical baselines
- Benchmark against previous events
- Identify relative changes in volume or sentiment
Output Format
Analysis outputs typically include:Tables
Structured data with columns for entities, counts, and metrics:Quantitative Findings
Key statistics extracted from the data:- Peak activity periods
- Percentage breakdowns
- Comparative metrics
Notable Examples
High-impact individual posts identified through engagement analysisConclusions
Contextualized interpretation of the quantitative findingsMore Analysis Examples
Sentiment Trend Analysis
Track how sentiment changes over time:Engagement Pattern Analysis
Analyze posting and engagement patterns:Comparative Company Analysis
Compare discussion volume across companies:Geographic Distribution
Analyze where conversations are happening:The code interpreter has access to Python libraries for data analysis including pandas, numpy, and matplotlib. Complex aggregations and transformations are automatically handled.
Combining with Other Flags
When to Use —analyze
Use the--analyze flag when you need:
- Quantitative answers — “How many posts?”, “What percentage?”, “Which company was mentioned most?”
- Trend identification — “When did activity peak?”, “Is sentiment increasing?”
- Comparative data — “How does this compare to last month?”, “Which account gets more engagement?”
- Aggregated metrics — “Total engagement”, “Average sentiment score”, “Median post length”
Analysis works best with queries that span enough data for statistical significance. For single-day snapshots or very narrow queries, regular search without
--analyze may be more appropriate.Related Flags
--from/--to— Define precise time windows for analysis--handles— Limit analysis to specific accounts--web— Include web sources in the data set--extract sentiment— Combine sentiment extraction with quantitative analysis