How to Use Prompt Architect
A Beginner's Guide
What Is Prompt Architect?
Prompt Architect helps you write better instructions for AI chatbots like ChatGPT, Claude, and Gemini.
What you type into an AI chatbot is called a "prompt." A better prompt gets you a better answer. This tool builds that better prompt for you.
You describe what you need, answer a few quick questions, and the tool gives you a ready-to-use prompt you can copy and paste into any AI chatbot.
Try it now — the tool is right above this guide. It's free, no account needed.
What it can do
-
Turn a vague idea into a clear, structured prompt
-
Recommend the best prompting technique for your task
-
Explain why each technique was chosen so you learn as you go
-
Show a side-by-side comparison of your original input vs the optimised prompt
-
Give you a prompt you can copy in one click
What it can't do
-
It doesn't run AI — it builds the prompt, you paste it into your chatbot
-
It doesn't guarantee perfect answers — always check AI output before using it
-
It doesn't replace your judgement — it's a starting point, not a finished product
See the Difference: Before and After
Here's what the tool actually does to your input.
Before (what you'd normally type into a chatbot):
Write me a blog post about SEO
After (what Prompt Architect builds for you):
You are a senior SEO content strategist who writes for small business owners. Write a 600-word blog post covering 3 key SEO trends for 2026. Use a conversational but authoritative tone. Include a practical takeaway for each trend. Be specific and actionable — avoid generic advice.
The "after" version tells the AI who to write as, what to cover, how long to make it, what tone to use, and what to avoid. That's why it produces a better answer.
How to Use It: Step by Step
Step 1: Describe Your Task
Type what you need into the text box.
What this does: Everything the tool builds is based on what you type here.
Why it matters: More detail gives better results. Compare these two inputs:
-
❌ Weak: "write something about SEO"
-
✓ Strong: "write a blog post about technical SEO tips for small business owners who manage their own websites"
The second one gives the tool your topic, format, audience, and context. The first gives it almost nothing to work with.
The Prompt Strength bar underneath the text box tells you if you've written enough:
-
✗ Red — Too short. Add specifics like your topic, audience, or what you want the output to look like.
-
◐ Amber — Fair. Try adding who it's for or what format you need.
-
✓ Green — Good. Enough detail to generate solid prompts.
-
✦ Bright green — Strong. Plenty of context to work with.
There's a 500-character limit, so keep it focused.
[Screenshot placeholder: the text input screen with strength bar]
Where this happens: In Prompt Architect on searchwithsean.com. No AI chatbot involved yet.
Step 2: Pick Your Goal Type
Choose the one that best describes your task.
-
Creative — Writing, brainstorming, drafting content. Pick this when voice and style matter.
-
Analytical — Researching, comparing, evaluating. Pick this when you need reasoning and logic.
-
Technical — Code, debugging, building systems. Pick this when accuracy and precision matter.
-
Communication — Emails, messages, pitches, briefs. Pick this when tone and audience awareness matter.
What this does: Different tasks need different prompting approaches. A blog post needs a different prompt structure than a debugging task. This step makes sure the tool picks the right technique.
Not sure which to pick? Go with whichever describes your final output. Writing a strategy report based on data analysis? That's Analytical. Writing a blog post that references data? That's Creative. Think about what you want to end up with, not what goes into it.
[Screenshot placeholder: the goal type selection screen]
Where this happens: Still in Prompt Architect.
Step 3: Choose Your Depth
This controls how much work you need the AI to do.
-
Quick — A short, fast answer. One prompt, one output.
-
Detailed — A thorough, well-structured response. The prompt asks the AI to go deeper.
-
Multi-step — A complex task broken into 3 stages. The tool gives you 3 separate prompts to run one after another.
What this does: A quick question needs a very different prompt than a detailed report. This makes sure the prompt matches your expectations.
How Multi-step works: You get 3 prompts. Run the first in your chatbot, copy the output, paste it into the second prompt, then repeat for the third. Each step builds on the last. The tool labels every step and tells you exactly where to paste.
[Screenshot placeholder: the depth selection screen]
Where this happens: Still in Prompt Architect.
Step 4: Pick Your Format and Preferences
Two quick decisions on one screen.
Format — What shape should the AI's answer take?
-
Paragraph, List, Email/Message, Report/Doc, Code, or Other
Preferences — The tool sets smart defaults based on your earlier choices, but you can adjust them.
-
Tone: Casual (friendly), Professional (direct), or Formal (polished). If you picked Technical + Code, it defaults to Professional. Creative + Paragraph defaults to Casual.
-
Experience level: Beginner (the prompt asks the AI to explain things simply), Intermediate (standard), or Advanced (tighter, no hand-holding).
If you see an extra text box: For some tasks, the tool recommends a technique called "Few-shot" that works best with an example to match. Paste in a paragraph, email, or code snippet that shows the style you want. This is optional — the prompt still works without one.
[Screenshot placeholder: the format and preferences screen]
Where this happens: Still in Prompt Architect. Your choices here shape the prompt you'll paste into your chatbot next.
Step 5: Get Your Results
This is where the tool delivers your optimised prompts.
The Delta View
A side-by-side comparison showing your original input on the left and the optimised prompt on the right. Below it, small tags show what was added — things like +Role, +Format, +Audience, +Reasoning.
This is the most useful part for learning. Over time, you'll start recognising what's missing from your first attempts and write better prompts on your own.
[Screenshot placeholder: the delta view comparison]
The Recommended Prompt
The tool's top pick for your task. Each prompt card shows:
-
The technique name — like Chain-of-Thought or Role/Persona
-
Why it was chosen — a plain-English explanation
-
Heads up — an honest warning about when it might not work perfectly
-
Model tip — which AI chatbots this works best with
-
The prompt itself — in a grey box, ready to copy
The Alternative Prompt
A second option using a different technique. If the first doesn't feel right, try the second.
The Copy Button
Click the red "Copy Prompt" button. Then open your AI chatbot and paste it in.
[Screenshot placeholder: the full results screen]
Where this happens: Results are built in Prompt Architect. You then paste the copied prompt into your chatbot:
-
ChatGPT — chat.openai.com → paste → Enter
-
Claude — claude.ai → paste → Enter
-
Gemini — gemini.google.com → paste → Enter
-
Any other AI chatbot — the prompts work with any major AI tool
The 8 Techniques (In Simple Terms)
You don't need to memorise these. The tool picks the right one for you. But here's what each does.
⚡ Zero-Shot — Asks the AI directly with clear instructions. No examples needed. Best for simple, quick tasks.
📝 Few-Shot — Includes an example so the AI can match a specific style. Like showing a chef a photo of the dish you want.
🧠 Chain-of-Thought — Asks the AI to think step by step before answering. Helps prevent it jumping to wrong conclusions. Best for logic and analysis.
🎭 Role/Persona — Tells the AI to act as a specific expert. Changes the tone, priorities, and type of advice it gives.
📐 Constrained — Sets strict rules: word count, format, what to include, what to avoid. Best when you need a precise shape of answer.
🔗 Prompt Chaining — Breaks a big task into 3 smaller prompts run in sequence. Best for complex work that's too big for one prompt.
🌳 Tree of Thoughts — Asks the AI for 3 different approaches, compares them, and picks the best. Best for decisions with no obvious right answer.
🪞 Meta-Prompting — Asks the AI to improve your prompt before answering it. Best when you're stuck and don't know how to phrase what you need.
Quick Tips
-
Edit your task any time. There's an Edit button at the top of each screen. No need to start over.
-
Skip the wizard. Already know which technique you want? Click "I know what I need" on the first screen.
-
Try both options. The recommended and alternative prompts use different approaches. Test both and see which gets better results.
-
Always check AI output. The tool builds great prompts, but AI chatbots can still get things wrong. Verify facts, data, and code before using them.
-
Use the feedback buttons. Thumbs up or down at the bottom helps improve the tool.
Where Each Step Happens
Step
Where
What You Do
Describe your task
Prompt Architect
Type your task
Pick goal type
Prompt Architect
Tap one of four options
Choose depth
Prompt Architect
Tap Quick, Detailed, or Multi-step
Format and preferences
Prompt Architect
Tap your choices
Get results and copy
Prompt Architect
Click "Copy Prompt"
Paste and run
Your AI chatbot
Paste and press Enter
Prompt Architect is free to use. No account needed. No data stored.
Final Stage - Testing Checklist By Claude
Scenario 1: Quick creative task Type: Write me 5 LinkedIn post ideas about AI in SEO Goal: Creative → Depth: Quick → Format: List → Tone: Casual → Level: Intermediate
What to check: Should recommend Zero-shot. The prompt should add specificity and anti-generic constraints. The delta view should show a visible difference between your rough input and the optimised version. If the before and after look too similar, the template isn't working hard enough.
Scenario 2: Detailed analytical task Type: Compare the pros and cons of migrating from Universal Analytics to GA4 for a small ecommerce business Goal: Analytical → Depth: Detailed → Format: Report → Tone: Professional → Level: Intermediate
What to check: Should recommend Chain-of-Thought. The prompt should include step-by-step reasoning instructions. The alternative should be Constrained. Both outputs should feel genuinely different from each other, not just slightly reworded versions of the same prompt.
Scenario 3: Technical debugging task (keyword override test) Type: Debug why my structured data is throwing errors in Google Search Console Goal: Technical → Depth: Detailed → Format: Code → Tone: Professional → Level: Advanced
What to check: The matrix would normally recommend Few-shot for Technical + Detailed + Code, but the keyword override should detect "debug" and "errors" and bump Chain-of-Thought to the primary recommendation. If it still shows Few-shot first, the override isn't firing. Also check that the Advanced level suffix strips out introductory hand-holding.
Scenario 4: Multi-step communication task (chaining test) Type: Write a pitch email to a potential podcast guest explaining why they should come on Search With Sean Goal: Communication → Depth: Multi-step → Format: Email → Tone: Professional → Level: Intermediate
What to check: Should recommend Prompt Chaining and display 3 separate steps (Draft → Refine → Polish) each with their own copy button. Check that the paste-here placeholders are clear between steps. The persona should be something like "a business communications expert" not a generic "communications director." Try copying all 3 steps and check the copy button grabs everything.
Scenario 5: Vague question-format input (edge case test) Type: What's the best way to improve my website's Core Web Vitals? Goal: Technical → Depth: Quick → Format: Paragraph → Tone: Casual → Level: Beginner
What to check: The tool should detect the question mark and reframe it as a task instruction internally — the generated prompt should start with something like "Provide a detailed, actionable answer to the following question" rather than just repeating the question as-is. The Beginner level should produce a softer persona ("a patient developer who explains concepts as they go") and add the "explain in simple terms" suffix. The prompt strength bar should show weak or fair since the input is short and vague.