Prompt crafting — also known as prompt engineering — is the art of interacting with AutogenAI in a meaningful way, using clear instructions to guide the tool's response. It's an essential skill for proposal writers in the era of augmented intelligence.
A prompt is the text you give a large language model to respond to. It can range from a single word to multiple paragraphs, and it serves as the input the model uses to generate its output. Successful prompts should be unambiguous, direct, and relevant — and worth curating as reusable, scalable solutions.
The prompting formula
Every successful prompt follows one of these formulas:
- Basic: Context + Specific Request
- Advanced: Purpose + Topic + Context + Structure + Specific Requests
Purpose — The "why." What outcome you want (inform, solve a problem, make a decision, etc.). It sets the direction for the response.
Topic — The "what." The main subject, so the AI focuses on the right area instead of answering too broadly.
Context — Background, nuance, and specifics that frame the question precisely — history, geography, constraints, or relevant events.
Structure — How you want the response organized: a list, a detailed explanation, a comparison. This makes the output easier to digest.
Specific Request — The precise instructions or questions within the larger ask — examples, a step-by-step process, a particular type of analysis.
Choosing your sources
Source Providers let you specify which sources AutogenAI uses when generating content. You'll find this option on the Home page or the Editor page. There are three source types:
- The Knowledge Hub / Library — your organization's library documents
- Datasets — structured data you've selected, including specific columns
- The Internet (Web Search) — general web results
Library AI
Toggle off "Generate using all Library Documents" to select specific folders or files instead. Every library and folder you have access to will appear in this menu.
Datasets
Toggle off "Generate using all datasets" to choose exactly which datasets to draw from.
Web Search
Use Search Only to restrict generation to specific websites you add, or Avoid Searching to allow the full internet except sites you exclude.
Clear vs. vague prompts
A poorly written prompt is ambiguous, indirect, and lacking specific detail. A well-written prompt is unambiguous, direct, and includes all relevant detail.
Vague: How to run an assessment.
Clear: How to run an occupational health assessment for shift workers in a biscuit factory.
Vague: What are the key risks in a structural engineering project?
Clear: What are the key risks that a structural engineer should consider when undertaking a large-scale construction project involving high-rise buildings within a dense UK/US city?
Vague: Detail what strategies our company will implement to ensure continuous improvement of the Soft FM services over the duration of the contract. Begin with stating our company's high commitment to continuous improvement. Describe how we have a system in place to ensure quality standards are consistently met and outline how compliance with industry standards and regulations will be monitored and maintained. Include relevant examples from past projects.
Clear: Same as above, but with the contract duration specified (e.g., "over the duration of [insert summary of the contract]") instead of left ambiguous.
How to prompt
Ask AI is the most effective way to prompt within AutogenAI. It generates content from multiple sources — the Knowledge Hub, the Internet, or Datasets — depending on the task.
- Highlight text, or press the Ask AI button, to activate the prompt box.
- Type in your prompt.
- Include all relevant context and avoid ambiguous language — if you don't ask, you don't get.
- If you don't like the result, use Regenerate for a new set of options.
Research lets you ask direct questions to the AI from your document Library or the Internet, with filters on which specific documents or websites to use.
- Select your Source Providers (Library, Datasets, Web Search).
- Type in the question you'd like to ask.
Saving prompts
Saving a prompt turns a one-off effort into reusable infrastructure. Instead of rewriting an instruction or hunting through old chats, you store a proven prompt once and call it up instantly across the platform. This keeps outputs consistent — everyone draws from the same vetted wording — and lowers the barrier for less experienced users, who can reuse what already works instead of figuring it out from scratch.
- Knowledge Hub: Find the Saved Prompts section under your library sources.
- Saved Prompt Menu: View your collection of saved prompts, or create a new one.
- Create New Prompt: Give your prompt a name, write it out, and save.
- Edit Prompts: Revisit any saved prompt to edit it.
- Saved Prompt icon: Click it in the editor to see all stored prompts.
- Select a saved prompt from the list.
- The prompt auto-populates into the prompting module, ready to send.
- Saved prompts are also available from the Research function on Home.
Prompting use cases
Producing content from scratch
Instruct the model to carry out the set of instructions in your prompt. For the best results, include as much context as possible and avoid ambiguous language or internal jargon.
Repurposing existing text
Save time by turning one piece of text into a different content form:
- Repurposing old proposal responses
- Converting content into bullet points
- Converting content to LinkedIn posts
- Converting content to blog posts
- Converting content to cover letters
Expanding existing text
Expand your content by leaving a trailing sentence for the AI to complete, or by prompting it to add more relevant detail — such as case studies or statistics. For example: "A case study to support this is..."
Key takeaways
- Prompts with AutogenAI should be clear, unambiguous, and relevant. Avoid ambiguous language and always provide context.
- Prompts can range from a single word to multiple paragraphs, guiding the model to generate responses based on its understanding of the input.
- AutogenAI has different tools and features to prompt with, depending on your use case.