How to Deep Research
A Protocol for Prompting a Deep Research Agent
DRIP - The prompt for creating the prompt
Deep Research has been around for a good while now, and it has become a common feature across major AI platforms. No wonder: it is still a highly efficient and versatile agentic research tool. It creates a research plan, conducts research across multiple sources, and returns a cited report, based on a single prompt from the user.
Prompting it well makes a difference. Deep Research prompt needs to state clearly what should be researched and what the finished report needs to do.
The prompt template shared here is called DRIP, short for Deep Research Interface Protocol. It is a protocol for creating the prompt that will be given to Deep Research.
On ChatGPT, Deep Research is available in Chat mode (currently found in the Plugins menu). But for DRIP, you DON’T use Deep Research mode, not yet. We’re interested in first creating the prompt that will then be used to prompt the model in the Deep Research mode.
DRIP uses information already present in the conversation, asks for missing information when needed, and turns the result into a complete Deep Research prompt.
Why not prompt Deep Research directly?
A fixed template works well when the research task is already fully specified. Often, the relevant information is spread across the current conversation or still needs some clarification before research begins.
DRIP uses ChatGPT to construct the research prompt from that material. It uses existing context, resolves consequential gaps, preserves important wording and background material, and then produces the prompt for Deep Research.
This keeps prompt construction and research execution as separate steps. Deep Research receives a finished brief instead of having to infer the brief while conducting the research.
The workflow
Step 1: Establish the research task
Start with what needs to be researched.
Use Chat mode first if the task still needs clarification.
State the research task in whatever form fits it best. It can be a direct question, a hypothesis to investigate, or a clear instruction such as mapping a field or tracing a concept through existing research.
Include the purpose of the research when it affects the evidence or the report.
Step 2: Run DRIP
Once the task is clear enough, copy and paste the DRIP prompt into the chat.
(Find the prompt in the end of this article. If you don’t want to subscribe, or need to pass the paywall, you can unlock one post for free).
DRIP uses the information already available to create the Deep Research prompt. It asks for clarification only when the answer would change the resulting prompt.
The protocol defines the scope of the research and adds any instructions the task requires. These can include source requirements or instructions for handling uncertain evidence.
DRIP also includes an optional Context Dump for background material that should remain available to Deep Research.
The Context Dump can stay rough. Notes do not need to be rewritten before inclusion. Unresolved claims can remain unresolved. Previous model output can be included as context without treating it as evidence.
Once enough information is available, DRIP produces the complete Deep Research prompt in a code block.
The generated Deep Research prompt is ready to go, but of course it is still editable before it is sent. Add, remove, or refine anything that needs adjustment.
The Context Dump section in the end of the prompt is exactly what it sounds like - the model dumps everything contextually relevant here, and so can you. If there are notes, source material, previous model output, or other project context that would help the research but was not already available in the conversation, paste or attach it there. It does not need to be cleaned up first.
Step 3: Configure the research environment
Start a fresh chat in Deep Research mode. Copy-paste the generated prompt into the chat.
If the prompt depends on particular sources, local or online, make those sources available there as well.
Deep Research can use uploaded files and selected websites in addition to the public web. Connected sources can also be used when they are available in the account.
The prompt tells Deep Research how those sources should be used. The interface gives Deep Research access to them.
Step 4: Review the plan and run the research
Deep Research shows its research plan before the run begins.
Read it before starting. Check that Deep Research has understood the task and has included the parts of the research that need to be there.
The plan can be changed before the run. The research can also be redirected while it is underway.
The finished report includes citations. Check important claims against the sources when the result will be used for consequential work.
The DRIP prompt
The actionable assets on this Substack are for paid subscribers. The paywall can be passed once.
If you find the Deep Research prompting protocol helpful, consider becoming a paid subscriber. This is a long-term process and more resources will be posted more or less regularly.
