How to Write a Good Prompt

Summary

Writing a prompt is like briefing a colleague: the clearer you define the task, the better the result. Clear prompts help Bynder AI Enrichment Agents generate accurate, brand-safe metadata that saves your team hours of manual work. 

This guide outlines best practices for drafting, previewing on sample assets, and refining prompts to optimize your enrichment workflows.

How? 

1. Define Prompt Requirements

  • Always include a destination for the agent's output. Specify whether the result should be saved to the asset's Title, Description, or a specific metaproperty. Example: "Describe the asset in 1 sentence. Save it in the /Description field."

2. Getting Started

  • Start small: Run your prompt on a small batch of assets to evaluate results before scaling up.
  • Iterate: Prompting is an iterative process. Small wording changes can lead to big improvements, so testing and refining really helps.
  • Don't delete output if something looks off: Keep the results and contact us. This helps us investigate and improve.

3. Follow Best Practices for Effective Prompt Writing

Think of the agent as a specialized technician: the clearer your work order, the better the output.

  1. Define the source and the destination. Always specify exactly where the output belongs and what field, if any, the agent should use as input. Use the / character to reference fields.
  2. Treat the agent like a new colleague. Think of it as a brilliant but literal-minded intern: it has all the world's knowledge but zero context on your company culture or "how we do things here."
  3. Provide role and context. Tell the agent who they are ("You are an SEO specialist"), what is requested ("Write SEO-optimized alt text"), and how to do it ("Use short, natural sentences"). Assigning a persona acts as a mental filter for the AI.
  4. Use step-by-step logic. For complex tasks, use chain-of-thought prompting: break the task into a sequence.
  5. Show examples (few-shot learning). Examples are the highest-leverage part of a prompt. Showing two or three inputs with their corresponding outputs eliminates ambiguity about the style and depth you expect.
  6. Use formatting and Markdown. Agents "see" structure through punctuation and spacing. Use Markdown (#, *, ---) to create visual boundaries. Bullet points and numbered lists are also very effective.
  7. Pivot to positive instructions. LLMs struggle with negative constraints because they focus on the keywords themselves. Instead of "Don't use slang," try "Use formal yet approachable department store terminology."
  8. Group conceptually linked tasks. Avoid context switching. Asking an agent to check legal compliance and brainstorm catchy TikTok captions in the same prompt will likely lower the quality of both.
  9. Repeat critical rules. If a rule is break-or-fix (like a character limit for a database), state it at the beginning and the end of your prompt.
  10. Use meta-prompting. Don't be afraid to ask another AI for help. Copy your draft prompt and ask it to improve it. Context is key here too: the more detail you provide, the better the refined prompt.

Tip: The agent editor also includes an Improve prompt button that adds structure, specificity, and best practices in one click. Review the improved prompt before saving; you stay in control of the final instructions.

4. Test and Refine Prompts with Agent Preview

You can test your prompt directly in the agent editor using real assets from your Asset Bank without launching a live run or consuming credits.

  1. Navigate to Agent Library.
  2. Previews are available for both new and existing agents
    1. Click the Create agent button at the top-right of the agent library to preview a new agent
    2. Click the ... button next to the agent you want to inspect, then Edit details to preview an existing agent.
  3. Create/edit a prompt, then click Preview to switch to the preview tab
  4. Click Select asset, scroll to find or search an asset to test the agent against. Once you click Add asset the output preview will be generated.
    1. Output preview can be generated for only one asset at the time. You can change asset to preview by clicking on the Replace asset button at the bottom of the page.
  5. Once satisfied with the output, select Create agent or save your changes to the agent.

Example Prompt: E-Commerce Alt Text Generator

Role & Context: You are an expert E-commerce SEO Specialist. Your task is to write high-quality, descriptive alt text for product images to improve web accessibility and search engine rankings.

Instructions:

  1. Analyze the uploaded asset image.
  2. Identify the primary product, its color, and any key defining features visible in the image.
  3. Write a concise, 1-sentence description based on your analysis.
  4. Keep the description under 125 characters.

Style Constraints:

  • Use literal, objective language describing what is visually present.
  • Use formal yet approachable terminology (e.g., "navy blue running shoe" rather than "sleek midnight sneaker").

Examples:

  • Input#1 (attached example asset #1)
    • Output #1: Red stainless steel reusable water bottle standing upright against a plain white background.
  • Input#2 (attached example asset #2)
    • Output#2 : A woman smiling while wearing a bright yellow, short-sleeved linen dress in a sunny outdoor park.

Critical Rule: Do not include fluff words like "image of" or "photo of." Keep the final text under 125 characters.

Destination: Save the final 1-sentence description into the /Alt Text metaproperty field.

Related Articles

Enrichment Agent: Automate Asset Metadata

Updated