[prompt-analysis] Copilot PR Prompt Analysis - 2026-05-27

[github-actions[bot]]

Summary

Analysis Period: Last 30 days (up to 2026-05-27)
Total PRs: 1,000 | Merged: 804 (80.7%) | Closed: 192 (19.3%) | Open: 4

Overall success rate is stable at ~80.7%, consistent with the past 10 days (range: 80.6%–81.7%).

Prompt Categories and Success Rates

Category	Total	Merged	Closed	Success Rate
Refactor	36	32	4	89% 🥇
Test	26	23	3	88% 🥈
Bug Fix	691	557	132	81%
Feature	186	147	37	80%
Docs	28	20	8	71%
Other	33	25	8	76%

Prompt Length Analysis

Body Length	PRs	Merged	Success Rate
Very short (<100 words)	103	82	80%
Short (100–300 words)	723	599	83% ✅
Medium (300–600 words)	45	33	73%
Long (600–1000 words)	7	3	43% ⚠️
Very long (>1000 words)	118	87	74%

Key finding: 100–300 word prompts have the highest success rate (83%). Long prompts (600–1000 words) perform poorly at 43%.

Title Verb Analysis

Opening Verb	Merged	Closed	Success Rate
optimize	17	0	100%
improve	13	0	100%
allow	8	0	100%
reduce	15	1	94%
normalize	13	1	93%
prevent	12	1	92%
enforce	10	1	91%
fix	140	26	84%
feat:	25	4	86%
add	92	20	82%
[WIP]	24	25	49% ❌

✅ Successful Prompt Patterns

Common characteristics in merged PRs:

• Average body length: 543 words
• Top action verbs: optimize, improve, reduce, normalize, prevent, enforce
• Titles using precision verbs (optimize, reduce, normalize) achieve 90%+ success rates

Example successful prompts:

View merged PR examples

1. PR Set 10-minute timeout on Playwright CLI installation steps #35270 — "Set 10-minute timeout on Playwright CLI installation steps" → Merged

   "Playwright CLI setup can hang...leading to long-running or stalled runs. This change adds explicit step-level time bounds so these installs fail fast instead of blocking the workflow indefinitely."

2. PR refactor(cli): consolidate local tokenClassWeights into types.TokenClassWeights #35274 — "refactor(cli): consolidate local tokenClassWeights into types.TokenClassWeights" → Merged

   "pkg/cli was redeclaring tokenClassWeights locally instead of using the canonical types.TokenClassWeights from pkg/types. This PR consolidates to a single source of truth with consistent underscore-based JSON tags."

3. PR Refactor docs landing-page guardrails section into visual security overview #35282 — "Refactor docs landing-page guardrails section into visual security overview" → Merged

   Clear, targeted scope with specific file references and concrete change description

❌ Unsuccessful Prompt Patterns

Common characteristics in closed PRs:

• Average body length: 706 words (30% longer than merged)
• Title prefix [WIP] has only 49% success rate vs 83% for clean titles
• Vague scope or "catch-all" PRs (e.g., "fix 26 consistency issues") tend to be closed
• Prompts mentioning codex, fallback, inference correlate with closure

Example unsuccessful prompts:

View closed PR examples

1. PR cli: fix 26 consistency issues across flags, naming, and docs #35234 — "cli: fix 26 consistency issues across flags, naming, and docs" → Closed

   Large-scope automated fix touching many unrelated concerns at once

2. PR Use local GitHub MCP mode in production report workflows #35273 — "Use local GitHub MCP mode in production report workflows" → Closed

   Correct intent but violated existing documentation constraints (remote vs local MCP mode)

3. PR Refactor BuildAWFCommand into composable helpers to reduce function size in pkg/workflow #35190 — "Refactor BuildAWFCommand into composable helpers..." → Closed

   Broad refactor without clear functional motivation

Key Insights

• Precision over breadth: Prompts with specific, targeted verbs (optimize, normalize, reduce) achieve 90–100% success. Broad catch-all verbs perform 10–20% worse.
• [WIP] is a red flag: PRs marked [WIP] have only 49% merge rate vs 83% for non-WIP. If work isn't ready, don't open the PR yet.
• Sweet spot is 100–300 words: This length range achieves 83% success. Very long prompts (600–1000 words) drop to 43%, suggesting bloated scope or overengineering.

Recommendations

1. DO: Use precision action verbs — optimize, normalize, reduce, prevent, enforce — to signal targeted, well-scoped changes
2. DO: Keep PR bodies in the 100–300 word range; describe exactly one well-scoped change with specific file/component references
3. DO: Use fix, add, feat: prefixes — they consistently achieve 80–86% success rates
4. AVOID: Opening [WIP] PRs — they are merged less than half the time; wait until the work is complete
5. AVOID: Prompts that bundle multiple concerns (e.g., "fix N issues across X areas") — these rarely merge

Historical Trends

Date	Total PRs	Merged	Success Rate	Top Category
2026-05-27	1,000	804	80.7%	Bug Fix
2026-05-26	1,000	802	80.6%	Bug Fix
2026-05-25	1,000	804	80.6%	Bug Fix
2026-05-24	1,000	810	81.3%	Bug Fix
2026-05-19	993	811	81.7%	Feature
2026-05-18	1,000	811	81.4%	Bug Fix
2026-05-17	1,000	809	80.9%	Bug Fix

Trend: Success rate has been remarkably stable at 80.6%–81.7% over the past 10 days. The slight dip from 81.7% (May 19) to 80.6% (May 25–27) may reflect a shift back toward bug-fix dominated PRs after a feature-heavy period.

---

References:

• §26539241491

Generated by 🔍 Copilot PR Prompt Pattern Analysis · sonnet46 1.1M · ◷

• expires on May 28, 2026, 9:22 PM UTC

[github-actions[bot]]

This discussion has been marked as outdated by Copilot PR Prompt Pattern Analysis.

A newer discussion is available at Discussion #35572.