Prompts help generate frameworks that can lead to deeper, more holistic insights that you would’ve never included in your original prompt.

For instance, compare the below two prompts.

Example Non-AI-Assisted Prompt:

Evaluate the different AI companies (ChatGPT, Claude, Perplexity, CoPilot), their products (ex: Perplexity Computer), and their advantages and disadvantages for finance professionals. Dive deep using clear examples.

Example AI-Assisted Prompt:

You are an expert in AI tools, enterprise software, and finance workflows. Conduct a rigorous, up-to-date evaluation of the leading AI platforms for finance professionals:

• OpenAI / ChatGPT

• Anthropic / Claude

• Perplexity

• Microsoft Copilot

Evaluate both the companies’ core AI assistants and their broader product ecosystems. This should include relevant products and capabilities such as ChatGPT Deep Research, Codex, Projects, custom GPTs, Claude Code, and any other significant products available at the time of the analysis.

Your primary audience is finance professionals working in roles such as:

• Strategic Finance and FP&A

• Investment Banking

• ….

Evaluation framework

Evaluate each company and product across the following dimensions:

1. Financial analysis and reasoning

Assess the ability to:

• Interpret financial statements

• Analyze budgets, forecasts, and actual results

• Diagnose variances

• Identify operational drivers

• Build or review financial models

• Maintain accuracy across long, multi-step analyses

• Explain financial concepts clearly to executives

Include examples such as:

• Analyzing a budget-versus-actual workbook

• .…

Required output structure

Part 1: Executive summary

Provide a direct summary of:

• The best overall platform for finance professionals

• The best platform for spreadsheet-heavy finance teams

• The best platform for deep research

• The best platform for document-heavy analysis

• The best platform for Microsoft-centric organizations

• ….

Analytical standards

Throughout the analysis:

• Use concrete, step-by-step finance examples.

• Avoid repeating company marketing language without testing or qualification.

• Separate native capabilities from capabilities that require APIs, custom code, connectors, or third-party products.

• Mention important limitations even when they weaken the recommendation….

The final analysis should be detailed enough that a finance leader could use it to decide which AI products to purchase, pilot, or approve for their team.

The second prompt introduces evaluation frameworks and dimensions that I probably wouldn't have thought to include myself. In practice, I've found this especially useful when I'm doing deeper market research, building frameworks to stress-test assumptions, or evaluating potential investments. Rather than simply answering the question I asked, the model broadens the scope of the analysis in ways that often surface angles worth exploring further.

That said, I've also found there's a point where adding more prompt engineering starts to work against you. Asking ChatGPT or Claude to generate a prompt that you then pass into another model doesn't automatically produce a better result. More often than not, the generated prompt becomes overly detailed, leading to responses that are comprehensive but difficult to work with. I still end up reading through the output, pulling out the handful of insights that actually matter, and using those as the starting point for the next iteration.