A production AI workflow is a repeatable three-stage process (Research and Gather, Reason and Analyze, Structure and Write) that combines ChatGPT and Claude so tasks like competitor analysis or customer feedback synthesis produce consistent output every time, without picking a new prompting pattern for each run.
What you will be able to do
- Build a three-stage workflow (Research and Gather, Reason and Analyze, Structure and Write) that runs the same way every time instead of starting from scratch
- Assign each stage to the model it suits best using the Handoff Pattern: ChatGPT for browsing and current research, Claude for analysis, writing, and structured output
- Set temperature per stage: 0.1 to 0.3 for fact extraction, 0.5 to 0.7 for analysis, 0.6 to 0.8 for final writing
- Apply quality gates between stages so missing citations, unresolved 'Not found' entries, or weak reasoning get caught before they reach the final draft
- Adapt the worked competitor-analysis example (with its JSON schema and prompt structure) as a template for your own research and writing workflows
Before you start
- Access to both ChatGPT and Claude, a browser tab for each is enough, no API required
- Working knowledge of Lessons 1 through 7 in this series: roles, temperature and top-p settings, few-shot examples, structured JSON output, chain-of-thought reasoning, plan-draft-critique, and RAG/source grounding
- Source material to run through the workflow, such as scraped competitor pages, support tickets, or article text
Reference
| Stage | Purpose | Patterns Used | Temperature | Output |
|---|---|---|---|---|
| Stage 1: Research and Gather | Gathers facts from sources and extracts structured data | RAG/source grounding, zero-shot extraction rules, structured output | 0.1 to 0.3 | Structured research notes with citations, as JSON or organized sections |
| Stage 2: Reason and Analyze | Reasons through the research and connects insights | Chain-of-thought reasoning, few-shot examples, role setting | 0.5 to 0.7 | Synthesized findings with a visible reasoning chain |
| Stage 3: Structure and Write | Transforms analysis into the polished final deliverable | Structured output, multi-stage self-critique, citation enforcement | 0.6 to 0.8 | Final document in the required format |
| Model pick: writing and editing | Newsletters, articles, tone-sensitive content | n/a | n/a | Claude, for more natural prose |
| Model pick: current-data research | Competitor pricing, recent news | n/a | n/a | ChatGPT, for web browsing |
| Model pick: code generation | Automation scripts, technical output | n/a | n/a | ChatGPT, for Code Interpreter |
Common errors and fixes
| What goes wrong | The fix |
|---|---|
| Stage 1 output makes a claim with no citation to a source document | Reject it at the quality gate and rerun Stage 1 rather than passing it to Stage 2 |
| Stage 1 has 'Not found in documents' entries that matter to the task | Resolve the gap before moving on, since Stage 2 and 3 will build on whatever Stage 1 hands them |
| Stage 2 analysis has no visible reasoning steps | Retry Stage 2 with adjusted instructions that explicitly ask it to show its reasoning |
| Stage 2 insights aren't tied back to Stage 1 data, or the logic has gaps | Send it back to Stage 2 instead of letting Stage 3 write from shaky analysis |
| Stage 3 output drops citations or doesn't match the required structure | Rerun Stage 3, the built-in self-critique step should catch this before you accept it as final |
Read the full walkthrough
The complete lesson, with screenshots and any downloads, is published on Substack as part of Prompt Engineering for AI Automation.
More in this section
- Lesson 7: Lesson 7: RAG Ground the Model with Your Own Sources
- Lesson 9: Stop Wasting 3 Hours Per Competitive Analysis
Continue the course
Browse all lessons in the Prompt Engineering for AI Automation course, or subscribe to the GenAI Unplugged newsletter to get new lessons in your inbox.
