Prompt Quality Guide • June 2026
Best AI Prompt Scoring Tools (2026) — Rate & Improve Your Prompts
Prompt scoring is the practice of measuring AI prompt quality against objective, weighted dimensions — not just "does it look good?" AI Prompt Architect scores every prompt on 5 dimensions (Structure, Content Depth, Code Quality, Diagrams, Completeness) and returns a 0-100 score with a letter grade and per-dimension improvement suggestions. It's the only tool that generates prompts and scores them in a single workflow.
Score your first prompt free
Get a 5-dimension quality score in under 30 seconds.
Definition: Prompt scoring is the systematic evaluation of an AI prompt's quality using weighted, measurable dimensions — producing a numerical score and actionable feedback. Unlike subjective review, scoring applies a consistent rubric across every prompt, enabling teams to set quality baselines, track improvements, and catch incomplete prompts before they waste AI tokens.
Why Most Prompts Fail
The average ChatGPT or Claude prompt is a single paragraph with no structure, no measurable quality, and no feedback loop. Our analysis of 10,000+ prompts submitted to AI Prompt Architect reveals:
Without scoring, you have no way to know why your AI output is mediocre. Prompt scoring turns "I don't know what's wrong" into "Structure: 25/100 — add section headings and numbered steps."
AI Prompt Architect's 5-Dimension Scoring System
Every prompt scored by APA is evaluated across 5 weighted dimensions. The final score is a weighted sum — not an average — so critical dimensions like Content Depth (25%) have more impact than Diagrams (15%).
🏗️ Structure
20%Section hierarchy, headings, markdown formatting, logical flow between sections. Measures whether the prompt uses clear H1/H2/H3 structure, bullet lists, and numbered steps — not just a wall of text.
📊 Content Depth
25%Word count vs target for the selected depth level, subsection density, specificity of instructions. A "Quick" prompt should hit ~500 words; "Exhaustive" should exceed 3,000 with dense subsections covering every edge case.
💻 Code Quality
20%Presence and quality of code blocks, language tags on fenced blocks, error handling patterns, type annotations, and test coverage instructions. Penalises pseudo-code without language tags or code blocks missing error handling.
📐 Diagrams
15%Mermaid diagram count and variety — ERD, flowchart, sequence, class, state diagrams. Prompts that include visual architecture produce 40% more accurate AI implementations (based on our internal testing across 10,000+ generations).
✅ Completeness
20%Coverage of data models, functional requirements, API specifications, deployment instructions, and security considerations. Checks whether the prompt addresses the full software lifecycle — not just "build me X".
Forbidden Pattern Detection
AI Prompt Architect's scoring engine automatically scans for forbidden patterns — placeholder markers that signal an incomplete prompt. When detected, a scoring penalty is applied and the specific offending text is highlighted in the report.
Removing all forbidden patterns typically increases your overall score by 10-20 points. A prompt with a single TODO in a critical section (e.g., API specification) can drop from a B+ to a C.
Prompt Scoring Tools Compared
| Feature | AI Prompt Architect | Prompeteer (16-dim) | OpenAI / Anthropic |
|---|---|---|---|
| Scoring dimensions | 5 (weighted) | 16 (unweighted) | None |
| Score range | 0-100 + letter grade | 0-100 per dimension | N/A |
| Forbidden pattern detection | ✓ Auto-penalty | ✗ | ✗ |
| Prompt generation + scoring | ✓ Integrated | ✗ Score only | ✗ Generate only |
| CLI scoring | ✓ apa score | ✗ | ✗ |
| MCP / IDE integration | ✓ Native | ✗ | ✗ |
| Framework | STCO (transparent) | Proprietary | None |
| Improvement suggestions | ✓ Per-dimension | ✓ General | ✗ |
| Price | Free / £9.99 / £14.99 | Free / Paid | Free |
| Best for | Generate + score + iterate | Score existing prompts | Basic generation only |
How to Score Your Prompts in 3 Steps
Paste or generate your prompt
Visit aipromptarchitect.co.uk and either paste an existing prompt or describe what you need in plain English. AI Prompt Architect will generate a complete STCO-structured prompt from your description — or restructure the one you pasted.
Review your 5-dimension score
Instantly receive a 0-100 score with a letter grade and per-dimension breakdown. See exactly where your prompt is strong (e.g., "Structure: 88/100") and where it's weak (e.g., "Diagrams: 15/100 — add at least one Mermaid diagram"). Forbidden patterns are flagged with line numbers.
Iterate and improve
Apply the dimension-specific suggestions, re-score, and watch your score climb. Most users go from a C (45-55) to an A (90+) in 2-3 iterations. Pro users can automate this via CLI: apa score prompt.md --format json for CI/CD integration.
📌 Key Takeaways
- Most prompts score below 50/100 — no structure, no code blocks, no diagrams. That's why AI output is inconsistent.
- AI Prompt Architect's 5-dimension weighted scoring gives you a transparent, actionable breakdown — not a black-box "try again."
- Forbidden pattern detection catches TODOs, [TBD], and placeholder text that silently degrades AI output.
- Score via web, CLI (apa score), or MCP — integrate prompt quality checks into any workflow.
- Free tier available — score your first prompt now.
Ready to rate your prompts?
Stop guessing. Start scoring. Get a 5-dimension quality breakdown on every prompt you write — free.
Frequently Asked Questions
What is a prompt scoring tool?
A prompt scoring tool analyses an AI prompt against measurable quality dimensions — such as structure, content depth, code quality, diagrams, and completeness — and returns a numerical score (typically 0-100) with actionable improvement suggestions. Unlike subjective "does this look good?" reviews, scoring tools apply consistent, repeatable rubrics so you can track prompt quality over time.
How does AI Prompt Architect score prompts?
AI Prompt Architect scores prompts on 5 weighted dimensions: Structure (20%) evaluates section hierarchy and markdown formatting; Content Depth (25%) measures word count vs target and subsection density; Code Quality (20%) checks for code blocks, language tags, and error handling; Diagrams (15%) counts Mermaid diagram quantity and variety; and Completeness (20%) verifies data models, requirements, and API specs. Scores range from 0 to 100 with a letter grade (A+ through F).
Is the AI Prompt Architect scoring tool free?
Yes. AI Prompt Architect offers a free tier that includes prompt generation with STCO framework and quality scoring on all 5 dimensions. Pro plans (£9.99/month) and Team plans (£14.99/month) add more generations, CLI access, and advanced features like prompt versioning and team collaboration.
What is a good prompt score?
On AI Prompt Architect's 0-100 scale: 90-100 (A+/A) means production-ready with full structure, depth, and diagrams. 70-89 (B+/B) is solid but may lack depth in one dimension. 50-69 (C+/C) needs improvement in multiple areas. Below 50 (D/F) indicates major gaps — usually missing structure, no code blocks, or placeholder text. Most unscored prompts from ChatGPT or Claude score 30-45 on first pass.
How is prompt scoring different from prompt optimisation?
Prompt optimisation (like the now-defunct PromptPerfect) rewrites your prompt in a black box — you get a "better" version but no explanation of why. Prompt scoring gives you a transparent, dimension-by-dimension breakdown showing exactly where your prompt is weak (e.g., "Structure: 45/100 — missing section headings") so you can learn and improve systematically rather than depending on opaque rewrites.
What are forbidden patterns in prompt scoring?
Forbidden patterns are placeholder markers that indicate an incomplete prompt: TODO, [TBD], FIXME, "placeholder", "lorem ipsum", "add details here", and similar. AI Prompt Architect's scoring engine automatically detects these and applies a penalty, because prompts containing placeholders produce unreliable AI output. Removing all forbidden patterns typically increases scores by 10-20 points.
Can I score prompts via CLI or API?
Yes. AI Prompt Architect's CLI tool (apa score) lets you score prompts directly from your terminal or CI/CD pipeline. You can also score prompts via MCP integration in IDEs like Cursor and Claude Code, enabling real-time scoring as you write. The 33-command CLI supports batch scoring, JSON output for automation, and integration with prompt versioning workflows.
Prompt Scoring: The Evidence
Every claim below is sourced from peer-reviewed research and industry reports.Browse all 141 citations →
Prompt caching reduces static context costs.
Cached prompt tokens cost $0.30/MTok vs $3.00/MTok uncached on Claude 3.5 Sonnet — a 90% reduction on repeated system instructions.
Without prompt caching, enterprise pipelines re-tokenise and re-bill the same system prompt across thousands of requests, paying 10x more for identical static context.
Anthropic, 'Prompt Caching (Beta)' documentation, 2024Few-shot extraction minimizes context window usage vs zero-shot verbose.
3 well-crafted few-shot examples (150 tokens) outperform a 600-token verbose instruction block, saving 75% on input costs per request.
Without concise few-shot examples, developers write lengthy prose instructions that consume 4x more tokens for equivalent or inferior output quality.
Brown et al., 'Language Models are Few-Shot Learners', NeurIPS 2020JSON Schema enforcement eliminates parse errors.
OpenAI structured outputs with JSON Schema achieve 99.9% schema adherence vs <70% with unconstrained generation — a 30x reduction in parse failures.
Without schema enforcement, every 1M requests generate 300K+ malformed responses requiring retries, error handling, and downstream data corruption.
OpenAI, 'Structured Outputs: JSON Schema' documentation, 2024Template systems compress prompt authoring time.
Structured prompt templates cut development time from 4 hours to 20 minutes per prompt (8x reduction) by separating instructions from variables.
Without templates, every new prompt starts from scratch — copying, pasting, and re-debugging the same boilerplate across dozens of prompts.
LangChain, 'Prompt Templates' documentation, 2024