⚠️ This column shares the prompt creation process for free educational purposes and does not constitute investment, legal, or financial advice.** AI execution results may contain errors, so independent verification is required. All responsibility for their use lies solely with the user. They cannot be directly used for business decisions. [See all essential checklists below the 30-week timeline before use[]](https://www.notion.so/30-29f84ddcec3c80a1a87eda9883d81453?pvs=21)
but in Week 5, I learned how to filter future implementation policies by document status. I wrote a rule in the prompt to unconditionally track Investigations and Proposeds, and to check the implementation date for Finals. With about 100 lines of prompts complete, it was time to start the actual search.
I asked Claude. "Find documents containing the keyword 'Investigation'." The search returned two or three documents. "Keep searching." Another two or three more came up. "There must be more, keep searching." Several more came up.
How long should I keep doing this? How many times would it take to complete the search? This method was inefficient. I had to plan the number of searches.
At first, I tried requesting searches one by one. I'd say, "Search with the keyword 'Investigation'," and Claude would search and return the results. Usually, two or three documents would appear.
"There must be more, right?" So I'd ask, "Search again with the same keyword." Another two or three results would appear. But this time, some of them were duplicates. How many times would I have to repeat "Keep searching"? Five times? Ten times? I couldn't tell when it was over. If no new results appeared, was it over? But changing the keyword might produce new results.
This method was unsystematic. Simply repeating "Continue searching" was time-consuming, and I couldn't be sure I'd found everything. A systematic plan was needed.
I needed to understand how Claude's web search worked. Here are the characteristics I identified: