How Good Am I at Prompting? [Self-Test]

How good am I at prompting? If you've asked yourself that question, you've already outgrown the "just add 'act as an expert'" phase. But the answer isn't a single trick or a multiple-choice quiz — it's a structured look at three distinct skills that most people conflate into one. This post gives you a framework to self-rate across those three dimensions using the same prompt engineering scoring criteria that AISA uses in its AI fluency assessment, then shows you where self-assessment predictably breaks down.

Prompt engineering quizzes from Real Python, Testlify, Adaface, and QuizGecko all test factual recall: "What does temperature do?" or "Which token limit applies to GPT-X?" That's trivia. It tells you almost nothing about whether you can actually get a model to produce useful output under real constraints. A proper prompt engineering test needs to evaluate how you think, not what you memorise.

What "Good Prompting" Actually Means

Good prompting is a thinking process, not a bag of tricks. It's the ability to decompose a goal into instructions a model can act on, then refine those instructions across turns until the output meets your standard. The person who writes one perfect prompt is less skilled than the person who writes a decent prompt, evaluates the response critically, and steers the conversation to a better result in three turns.

This distinction matters because the field has matured. With models like Claude Opus 5 supporting 1M-token context windows and five-level effort controls, the skill ceiling has risen. You're no longer wrestling with a 4K-token window where every word counts for space reasons. Instead, the challenge has shifted to what you include, how you structure it, and whether you can manage a multi-turn session without the model losing the thread.

The Thinking Process, Not the Template

Templates like "You are a [role]. Your task is [task]. Output in [format]" are fine scaffolding. They're also table stakes. The Anthropic AI Fluency Index — which AISA's rubric was validated against at 93% overlap — treats prompting as a composite of design, dialogue, and context management. If you only do the first one, you're exercising a third of the skill.

Why Tricks Plateau Fast

Anyone can learn "chain-of-thought" prompting in an afternoon. The difference between a Competent and Proficient prompter isn't knowing the technique — it's knowing when chain-of-thought helps and when it just burns tokens and adds latency. A 2024 Stanford HAI report found that structured prompting strategies improved task accuracy by 15-40% depending on task type, but the variance was driven almost entirely by the user's ability to match strategy to problem, not by the strategy itself.

Rate Yourself: 3 Dimensions of Prompting Skill

Self-rating works best when you have concrete anchors, not vague labels. The AISA rubric breaks prompting into three scored criteria: Prompt Design (P1), Iterative Dialogue (P2), and Context & Memory Management (P3). Together they make up 23% of the overall AI fluency composite score. Here's how to use the score-band language to place yourself honestly.

P1: Prompt Design

This is the one most people think of when they hear "prompt engineering." It's about the quality of a single instruction — clarity, specificity, constraint-setting.

Be honest. If your prompts usually look like "Summarise this article" with no format spec, audience note, or length constraint, you're in the 3-4 range regardless of how good the output sometimes is.

P2: Iterative Dialogue

This is the dimension most people underrate in themselves — or don't think about at all. It measures your ability to steer a multi-turn conversation productively.

P3: Context & Memory Management

With 1M-token context windows now standard across frontier models, this skill has become more important, not less. More rope means more ways to hang yourself.

Write down your three scores. We'll come back to them.

3 Before/After Prompts That Show the Gap

The difference between a weak and strong prompt is usually specificity and structure, not length. Here are three real patterns, moving from common to less obvious.

Example 1: The Vague Request

Before (Novice/Developing):

Write a blog post about our new feature.

After (Proficient):

Write a 600-word blog post announcing [Feature X] for [Product]. Audience: existing customers who use [Related Feature]. Tone: direct, no marketing fluff. Structure: what it does (2 paragraphs), why we built it (1 paragraph), how to enable it (numbered steps). Don't mention competitors.

The second prompt isn't longer for the sake of length. Every added sentence removes a degree of freedom the model would otherwise fill with guesses.

Example 2: The Iteration Failure

Before (Developing iteration):

That's not what I wanted. Try again.

After (Proficient iteration):

The structure is right, but the opening paragraph assumes the reader knows what [Feature X] replaces. Rewrite the opening to explain the problem it solves before naming the feature. Keep everything else.

The second follow-up tells the model exactly what's wrong, where it is, and what to preserve. This is the P2 skill — iterative dialogue — and it's where most people leave points on the table.

Example 3: The Context Overload

Before (Developing context management):

[Pastes 40-page product spec] Summarise the key changes.

After (Proficient context management):

I'm attaching our product spec (40 pages). I need a summary of changes from v2.3 to v2.4 only. Focus on: API breaking changes, new endpoints, and deprecated features. Ignore UI changes. Output as a markdown table with columns: Change Type, Endpoint/Feature, Description, Breaking (Y/N).

The second version doesn't reduce what's in context — it tells the model how to navigate the context. That's the difference between dumping information and managing it.

Why Self-Rating Undersells Your Iteration Skills

People consistently overestimate their design skills and underestimate their dialogue skills. Across 1,064 completed AISA assessments, the average prompting dimension score is 45.3 out of 100 — firmly in the Developing-to-Competent range. But the more revealing number is the prediction gap: candidates who predicted their scores beforehand (n=166) overestimated by an average of 19.2 points overall.

Why does this happen specifically with prompting? Three reasons.

You Don't See Your Own Multi-Turn Patterns

Most people evaluate their prompting by the final output they got, not by the process they used to get there. If you spent 12 turns getting a mediocre result that you could have reached in 3 turns with better feedback, you're less skilled than you think — even though the output was "fine."

One-Shot Success Creates False Confidence

Sometimes a model nails it on the first try. You attribute this to your prompt. In reality, the model may have been trained on nearly identical examples, or your task was simple enough that almost any prompt would have worked. This inflates your self-assessed P1 (Prompt Design) score while telling you nothing about P2 or P3.

Quizzes Can't Measure Dialogue

This is the fundamental problem with every prompt engineering quiz on the market. A multiple-choice test can ask "What is chain-of-thought prompting?" It cannot observe whether you actually steer a conversation well. That requires a conversational format — which is why AISA uses an AI facilitator that conducts a real dialogue, scored by a separate AI evaluator. It's a different approach from quizzes, and it's specifically designed to capture P2 skills that static tests miss.

If you scored yourself above a 6 on P2 (Iterative Dialogue) in the self-rating above, ask yourself: could you describe your last three multi-turn AI conversations and explain why you sent each follow-up? If not, you might want to revise that number.

From Self-Assessment to Actual Measurement

Self-rating is a starting point, not an endpoint. It's useful for identifying which of the three dimensions you should focus on, but it has a ceiling. The Dunning-Kruger pattern is well-documented in skill self-assessment — a 2023 Gartner survey found that 64% of employees rated themselves as "advanced" or "expert" AI users, while objective assessments placed the majority at intermediate levels or below.

If you want a number you can trust, you need an assessment that actually watches you prompt. AISA's AI fluency assessment does this through conversation, not checkboxes. But even without a formal assessment, the framework above gives you something most prompt engineering tests don't: a way to think about prompting as three separate skills rather than one.

The practical next step is simple. Pick the dimension where you scored lowest. Spend a week deliberately practising it. If it's P1, focus on adding constraints and format specs to every prompt. If it's P2, force yourself to give specific, structural feedback instead of "try again." If it's P3, start summarising context at turn 5 of every conversation and see how output quality changes.

That's more useful than any quiz score.

---

Related reading: How Good Am I at AI? A Self-Assessment Framework — the broader version of this self-test, covering all 5 dimensions.

Related reading: AI Know-How Test: Do You Know AI, or Do You Know How to Use It? — why knowledge and skill aren't the same thing.

Related reading: AI Fluency Assessment: 9 Methods Compared [2026] — how quizzes, self-assessments, and conversational tests stack up.

Frequently Asked Questions

How do I know if I'm good at prompting?

Rate yourself separately on three dimensions: prompt design (clarity and structure of a single instruction), iterative dialogue (ability to steer multi-turn conversations with specific feedback), and context management (how you organise and reference information across long sessions). If you score Competent (5-6) or above on all three using the anchors in this post, you're ahead of most users. For a validated score, a conversational AI fluency assessment like AISA measures all three through actual dialogue rather than self-report.

What makes a prompt good vs bad?

A good prompt removes ambiguity. It specifies the task, audience, format, constraints, and what to exclude. A bad prompt forces the model to guess on multiple dimensions simultaneously. But prompt quality is only one-third of the skill — a good prompter also iterates effectively and manages context across turns. The strongest prompters treat each prompt as one move in a longer strategy, not a standalone command.

Is prompt engineering still relevant in 2026?

Yes, but it has evolved. With frontier models supporting 1M-token context windows and multi-step tool use via protocols like MCP, the skill has shifted from "fit your request into a small window" to "structure complex workflows and steer multi-turn sessions." The core ability — translating intent into instructions a model can act on — is more relevant than ever. What's less relevant is memorising specific syntax tricks, since models have become better at interpreting loosely structured inputs. The thinking process behind the prompt matters more than the format.

Can a quiz measure prompt engineering skill?

Multiple-choice quizzes can test knowledge about prompting — terminology, techniques, model capabilities. They cannot measure whether you actually apply that knowledge effectively in practice. Iterative dialogue skill (P2) and context management (P3) require observation of real behaviour across multiple turns, which static quizzes fundamentally cannot provide. That's the gap conversational assessments are designed to fill.