Anatomy of a NewBizBot Prompt

Why Prompt Structure Matters

The quality of AI research output directly depends on the quality of instructions it receives. Think of a prompt as a detailed brief you'd give to a new research analyst—the clearer and more specific your instructions, the more relevant and useful the results.

NewBizBot structures prompts into four distinct components because each serves a different purpose in guiding the AI's behavior. This modular approach lets you customize specific aspects of research without rewriting everything, and it ensures consistency across different research tasks.

Core Prompt Structure

Every NewBizBot prompt consists of four essential components that work together to define the scope, context, and expected output of your research request:

Task

The core instruction that defines what specific research action to perform. This should be a clear, concise statement of the primary objective (e.g., "Produce detailed, structured profiles" or "Identify top 20 companies in X sector").

Backstory

Provides essential context about who is making the request and establishes the researcher's expertise and perspective. This helps the AI understand the professional context and adopt the appropriate tone and depth of analysis.

Goal

Contains the detailed objectives and specific requirements for the research. This section outlines what information to gather, how to structure findings, search strategies, and quality standards for the final output.

Expected Output

Shows the exact format, structure, and level of detail required in the final response. This includes examples of how the information should be presented, what sections to include, and the overall organization of results.

Output of The Example Prompt

See how your research results will actually appear in the NewBizBot interface:

Understanding Each Component's Role

When these four components work together, they create a comprehensive framework that guides the AI to produce consistent, high-quality research results that meet your specific needs and professional standards.

---

The Principles Behind Effective Prompts

Before diving into how to write each component, understanding a few core principles will help you craft prompts that consistently deliver excellent results.

Be Specific, Not Vague

The AI performs best when it knows exactly what you want. Compare these two approaches:

Vague: "Find information about this company" Specific: "Identify the ownership structure, key executives with their backgrounds, annual revenue, recent funding rounds, and any M&A activity in the last 24 months"

Specific instructions eliminate guesswork. The AI doesn't have to interpret what you might want—it knows precisely what to find and include.

Provide Context for Better Judgment

The AI makes thousands of micro-decisions during research: which sources to prioritize, how much detail to include, what level of technical language to use. Your Backstory component shapes these decisions.

When you establish that the researcher is "an investment banker conducting preliminary due diligence," the AI understands to focus on financial metrics and deal-relevant information, use professional terminology appropriate for sophisticated investors, prioritize accuracy over comprehensiveness when time-sensitive, and flag potential red flags that would matter in a transaction context.

Without this context, the AI might produce a general overview that misses the specific angles a banker would need.

Show, Don't Just Tell

The Expected Output component isn't just about describing what you want—it's about demonstrating it. When you include an example output, the AI can pattern-match against a concrete reference rather than interpreting abstract descriptions.

This is why the default configurations include detailed sample profiles. The AI sees exactly how to format employment history, how to structure financial metrics, and what level of source citation you expect.

Guide the Process, Not Just the Result

Your Goal component can include instructions about how the AI should conduct research, not just what it should find. This matters because research quality depends on methodology.

For example, guidance like "Always assess the searched information with a critical lens" encourages source verification, while "Be very careful for different entities with the same names" prevents embarrassing mix-ups. Instructions to "Adapt to your search based on the search results" enable intelligent pivoting when initial approaches don't work.

These process instructions create a more thoughtful, reliable research approach.

---

Deep Dive: Writing Each Component

Crafting an Effective Task

Your Task statement sets the fundamental direction for research. It should answer one question clearly: What is the AI supposed to produce?

Characteristics of Strong Task Statements:

Task Statement Templates:

The default prospect profiling Task reads:

You can adapt this pattern for other research types. For competitive analysis, you might frame it as "Analyze and compare competitive positioning across [specified dimensions] for the requested companies." For lead identification, consider "Identify and qualify [number] [entity type] that match the specified criteria for [use case]."

Building the Right Backstory

The Backstory establishes the AI's persona and expertise level. Think of it as briefing a consultant before they start work—you want them to understand who they're working for and what lens to apply.

An effective Backstory establishes four key elements: the professional identity (what kind of expert is conducting this research), the domain expertise (what specific knowledge areas should the AI draw upon), the output purpose (how will this research be used), and the quality expectations (what standards should govern the work).

Example Backstory Construction:

Industry-Specific Backstory Adaptations:

Designing Comprehensive Goals

The Goal component does the heavy lifting. It specifies exactly what information to gather and how to approach the research process. Well-designed Goals balance thoroughness with focus.

Structuring Your Goals:

Organize goals into clear categories that match your research subjects. The default configurations structure objectives by entity type—individuals versus companies—with specific data dimensions for each.

For individuals, the default wealth management prompt focuses on gathering employment history and career trajectory, net worth composition broken down by asset class (not just a number), family circumstances and dynamics, board affiliations and philanthropic involvement, and public presence and recent news.

For companies, the focus shifts to ownership structure and cap table, executive team profiles, financial metrics (revenue, funding, profitability), strategic milestones (M&A, partnerships, pivots), and competitive positioning.

Adding Search Guidance:

Beyond specifying what to find, guide how the AI should search. The default prompts include guidance to adapt research based on search results rather than robotically following a format, assess searched information with a critical lens, and be careful about entities with the same names. When the AI encounters conflicting information, it should take time to verify from multiple sources.

Including Quality Standards:

Quality standards ensure consistent, verifiable output. The default prompts instruct the AI to use the report formatting as a guideline while including additional relevant information discovered during research. They emphasize that the final report should have sufficient length to tell the cohesive story of findings, and require citing sources with links in markdown format.

Common Goal Additions by Use Case:

Defining Expected Output

The Expected Output component serves as a template for results. The AI uses this as a pattern to follow, so the more concrete and detailed your example, the more consistent your results.

The key principles for crafting Expected Output are to show actual formatting rather than describing it—don't just say "use bullet points," include a bulleted list. Examples should include realistic detail that reflects the depth you actually want. Demonstrate source citation by showing how links should be incorporated inline. If your Goal handles both individuals and companies, your Expected Output should show examples of both scenarios, as the default configurations do.

Adjusting Output Length and Detail:

Your Expected Output implicitly signals how much detail you want. A 200-word example suggests concise summaries; a 2,000-word example signals comprehensive profiles.

For quick briefs, you might include just the essential fields: role, background in 2-3 sentences, one key fact relevant to your meeting, and recent news.

For comprehensive profiles, include the full structure shown in the default configuration—detailed sections for each category of information, with multiple bullet points and source links throughout. The default Expected Output examples demonstrate exactly this level of detail with complete individual and company profile templates.

---

Prompt Writing Best Practices

Start with Defaults, Then Iterate

NewBizBot's default configurations represent tested, proven prompt structures. Rather than starting from scratch:

This iterative approach prevents the common mistake of over-engineering a prompt before understanding what actually needs to change.

Use Concrete Language Over Abstract Instructions

The AI interprets language literally. Abstract instructions often produce inconsistent results because they require interpretation.

Anticipate Edge Cases

Good prompts handle unusual situations gracefully. Consider adding guidance for private versus public companies (different information availability requires different approaches), individuals with common names (disambiguation strategies), limited information scenarios (what to do when standard searches don't yield results), and conflicting information (how to handle and flag discrepancies).

The default prompts address some of these by instructing the AI to "always assess the searched information with a critical lens" and to "be very careful for different entities with the same names." You can extend this by adding explicit instructions like noting "Ownership structure: Not publicly disclosed" rather than omitting the section when information isn't available, or presenting both perspectives with their respective sources when sources conflict.

Balance Comprehensiveness with Focus

Prompts that try to capture everything often capture nothing well. The AI has limited context and attention—overloading instructions can dilute focus on what matters most.

Test with Diverse Subjects

A prompt that works perfectly for one type of subject might fail for another. Test your configurations with well-known public figures (high information availability), obscure individuals (low information availability), large public companies (abundant data), small private companies (limited data), and subjects in your specific target industry. Testing reveals assumptions in your prompt that only surface with certain subject types.

---

Troubleshooting Common Issues

Research Results Are Too Generic

Symptom: The AI returns surface-level information that anyone could find with a quick Google search.

Solutions: Add more specific requirements to your Goal. For instance, the default wealth management prompt specifies to "go beyond just a net worth figure" and break down into stock holdings, real estate, and business equity. Include search guidance pointing to specific source types like SEC filings, property records, and business registration databases. You can also adjust your Backstory to establish higher expertise expectations.

Important Information Is Missing

Symptom: Results consistently omit categories of information you need.

Solutions: Explicitly list the missing category in your Goal with specific sub-requirements. Include that category in your Expected Output example to demonstrate its importance—the AI learns from concrete examples. You can also add a "Required elements" note at the end of your Goal to emphasize must-have information.

Output Format Doesn't Match Needs

Symptom: Results are too long, too short, or structured differently than you want.

Solutions: Provide a complete Expected Output example in exactly the format you want—the AI pattern-matches against your example. Add explicit length guidance such as "Maximum 500 words" or "Include at least 10 bullet points per section." Specify structure requirements like "Use headers for each major section" or "Format as a single-paragraph executive summary."

Information Quality Concerns

Symptom: Results include unverified claims, outdated information, or obvious errors.

Solutions: Add verification requirements to your Goal, such as "Cross-reference key facts across multiple sources." The default prompts already require source attribution—"Make sure to cite your sources with links in markdown format"—which you can emphasize or extend. Include recency requirements like "For financial metrics, use the most recent available data and note the date." Add disambiguation guidance to "Verify entity identity using multiple data points before attributing information."

---

Advanced Techniques

Creating Industry-Specific Variations

Once you've mastered the basic structure, create specialized configurations for each industry you regularly research:

Healthcare/Biotech: Consider adding FDA approval status and regulatory pathway, clinical trial information from ClinicalTrials.gov, patent portfolio and expiration dates, key opinion leader relationships, and reimbursement considerations.

Technology/SaaS: Focus on product-market fit indicators (customer growth, retention), technology stack and technical moat, competitive landscape positioning, key integrations and partnerships, and developer community engagement.

Real Estate: Include property portfolio by type and geography, development pipeline and permits, tenant quality and lease terms, zoning and regulatory considerations, and environmental or historical designation issues.

Chaining Multiple Research Objectives

For complex research needs, consider creating a sequence of configurations rather than one monolithic prompt. Start with initial discovery for a broad sweep to identify key areas, then move to a deep dive for focused investigation of specific aspects, and finally verification for cross-checking critical claims. This approach prevents prompt overload and allows targeted refinement at each stage.

Adapting for Different Output Audiences

The same research might serve different audiences with different needs. For internal analyst use, emphasize high detail with technical language, include all supporting data, and provide extensive source documentation. For client deliverables, focus on polished presentation with executive summaries leading and key findings highlighted. For quick internal briefings, use a bullet-point format covering just the essential facts with focus on recent developments.

---

Next Steps

Now that you understand how NewBizBot prompts work, you're ready to create configurations tailored to your specific workflows. Start by reviewing the default configurations on the Prompt Configuration Page to see proven examples in action. Create your first custom configuration by copying a default and modifying the Goal for your industry, then test with real research subjects to refine your prompts based on actual results. As you identify recurring research patterns, build a configuration library for your team.

---

Quick Reference