If you find yourself repeatedly asking ChatGPT to research topics, summarize articles, or compile information the same way over and over, you’re wasting time re-explaining your preferences. Instead, you can build a personal AI research assistant that remembers exactly how you like your research done.
Custom GPTs in ChatGPT let you create a specialized version of the AI trained on your specific instructions. Once set up, your research assistant will consistently deliver results in your preferred format without needing to repeat yourself. Here’s how to build one in under 15 minutes.
What You’ll Need
To create a custom GPT, you need a ChatGPT Plus or Team subscription ($20/month). Free accounts can’t access the GPT builder yet. You’ll also want to think through how you typically conduct research—what sources you prefer, how detailed you want summaries, and what format works best for your workflow.
Creating Your Research Assistant
Start by navigating to ChatGPT and clicking your profile icon, then select “My GPTs” and click “Create a GPT.” The GPT Builder will walk you through setup with a conversational interface.
Tell the builder what you want. For example: “I want to create a research assistant that helps me investigate topics thoroughly, summarizes findings in bullet points, and always cites sources with links.”
The builder will suggest a name and generate a profile picture. You can accept these or request changes. More importantly, it will draft initial instructions. Review these carefully—this is where the magic happens.
Writing Effective Instructions
Click “Configure” to see and edit the full instructions. This is where you define exactly how your assistant behaves. Here’s a template that works well:
- Role: Define what the GPT is. Example: “You are a thorough research assistant who investigates topics deeply and presents findings clearly.”
- Process: Explain the steps it should follow. Example: “When given a topic, search for current information, prioritize authoritative sources, and cross-reference multiple perspectives.”
- Output format: Specify exactly how you want results. Example: “Present findings as: 1) Executive summary (3-5 sentences), 2) Key findings (bullet points), 3) Important caveats or limitations, 4) Sources with clickable URLs.”
- Tone and style: Set the voice. Example: “Write in clear, professional language. Avoid jargon unless necessary, and explain technical terms when used.”
You can also add constraints. For instance: “If information is outdated or unverified, flag it clearly” or “When sources conflict, present multiple viewpoints rather than choosing one.”
Adding Conversation Starters
Under “Conversation starters,” add 3-4 prompts that demonstrate what your GPT does well. These appear as buttons when someone (including you) opens the GPT. Examples:
- “Research the latest developments in [topic] from the past 3 months”
- “Compare different approaches to [problem] with pros and cons”
- “Summarize the key papers on [academic topic]”
- “Find expert opinions on [controversial issue]”
These aren’t just helpful hints—they train users (and remind you) what the GPT is designed to do.
Testing and Refining
Before saving, test your GPT thoroughly using the preview pane. Try several research requests that represent your typical use cases. Does it cite sources properly? Is the format consistent? Does it ask clarifying questions when needed?
If something’s off, return to Configure and adjust the instructions. You might need to be more specific. For example, instead of “cite sources,” try “include the full URL and publication date for every source in a numbered list at the end.”
Common refinements include adding: “Always check your knowledge cutoff date and mention if the topic may have recent developments you can’t see” or “If you’re uncertain about any information, say so explicitly rather than speculating.”
Making It Even More Powerful
Once your basic research assistant works well, consider these upgrades:
Enable web browsing in the GPT settings so it can access current information beyond ChatGPT’s training data cutoff. This is essential for timely research.
Upload reference documents under Knowledge. If you frequently research within a specific domain, upload key papers, glossaries, or style guides. Your GPT will reference these automatically.
Add specific output templates. If you always need research formatted for reports, blog posts, or presentations, include those templates in your instructions with placeholders.
Privacy and Sharing
Under “Settings,” choose who can access your GPT. “Only me” keeps it private. This is smart if your instructions contain proprietary information or workflow details. You can always share it later via link or publish it to the GPT Store if you want others to benefit from your creation.
One useful practice: create different research assistants for different domains. You might have one for technical research, another for market analysis, and a third for academic literature reviews. Each can have domain-specific instructions and knowledge bases.
Your personal research assistant gets smarter as you refine it. After each use, note what worked and what didn’t, then update the instructions accordingly. Over time, you’ll have a genuinely customized tool that knows exactly how you work.
Want to discover more practical AI techniques like this? Subscribe to the One Two Three AI newsletter and get one useful AI idea delivered daily—no hype, just actionable tips you can use right away.