Master Long Documents with Gemini’s Massive Memory
The topic
Google Gemini boosts context window to 1 million tokens
Google recently announced a significant increase in the context window for its Gemini 1.5 Pro model, expanding it to 1 million tokens. This allows the AI to process and understand much longer documents, entire codebases, or hours of video/audio in a single prompt.
Today's framework: BAB — Before · After · Bridge
Master Long Documents with Gemini’s Massive Memory
This prompt helps you use AI to 'read' and organize huge amounts of information at once, saving you hours of manual review.
The prompt
Before: I have several long documents, meeting transcripts, and project notes for my current work. It is overwhelming to read through hundreds of pages to find connections, summarize key points, or spot missing details manually.
After: I want to have a clear, concise summary of all these files combined. I need to understand the main themes, get a list of action items, and be able to ask specific questions about any detail hidden deep within these documents as if I had an expert assistant who memorized everything.
Bridge: Using your massive capacity to process long documents (context window), please analyze the text I am providing. Create a high-level summary of the main goals, a bulleted list of all deadlines mentioned, and a list of who is responsible for each task. Let me know once you are ready for me to ask follow-up questions about specific parts of the files.
How to use it
1.Copy several long reports or meeting notes into a single document or upload the files directly.
2.Paste the prompt into Gemini (or ChatGPT) along with your text.
3.Review the summary and action items generated by the AI.
4.Ask follow-up questions like 'What did Sarah say about the budget in the third meeting?'
Things to watch for
·Double-check important dates and names to ensure the AI didn't miss a detail.
·Do not upload highly sensitive or private company secrets unless using a secure work version.
·If the documents are extremely long, be patient as the AI takes a moment to 'read' everything.
Why this prompt works
The Before-After-Bridge (BAB) framework works by establishing a clear baseline of current struggles and defining an ideal future state, which provides the AI with critical context. The Before section clarifies the current pain points of information overload, while the After section defines the desired outcome of clarity and speed. Finally, the Bridge section serves as the specific instruction set that guides the AI on exactly how to transform that messy input into structured, usable data using its expanded context window.
Example AI response
I have completed a comprehensive analysis of the provided 450 pages of documentation and transcripts. The primary goals identified are the synchronization of the Q4 software rollout with the regional marketing campaigns and the mitigation of the current supply chain delays. Key deadlines include the final UI polish by October 12th and the stakeholder sign-off on November 1st. Responsible parties are Sarah Jenkins for engineering oversight and Marcus Thorne for logistics coordination. The documents also highlight a potential bottleneck in the testing phase scheduled for late September. I have indexed all sections regarding budget allocations and technical specifications; I am now ready for your specific follow-up questions regarding any particular document or detail.
Sample only — your AI's answer will be different. That's the fun part.
Try a variation
Codebase Audit
Before: Our development team is struggling to onboard new members because our entire codebase and documentation are spread across dozens of files, making it nearly impossible to see the architecture at a glance. After: I want a comprehensive architectural overview that identifies core dependencies, hidden technical debt, and a logical roadmap for new developers to follow. Bridge: Utilizing your massive token capacity, please ingest these source files and documentation. Create a visual-style text map of the system architecture, list the most critical functions, and identify any inconsistent naming conventions or missing documentation. I will then ask for specific refactoring advice.
Legal/Academic Review
Before: I am conducting research across a dozen lengthy legal transcripts and academic papers totaling hundreds of thousands of words, and I am worried about missing subtle contradictions or cross-references. After: I need a centralized synthesis that flags every time a specific topic is mentioned across different sources and highlights any conflicting statements between authors. Bridge: Act as a senior research analyst. Using your large context window, analyze these uploaded PDFs. Provide a cross-referenced table of key themes, note any direct contradictions in the data, and summarize the consensus. Confirm when you are ready for me to query specific page citations.
Video/Media Analysis
Before: I have hours of raw footage transcripts and interviews from a documentary project, making it tedious to find the most impactful quotes or consistent narrative threads. After: I want a structured storyboard draft that organizes the best quotes by emotional tone and identifies the strongest narrative arc from the collective footage. Bridge: Please process these multi-hour transcripts. Identify the five most compelling stories within the text, categorize the tone of each interviewee, and suggest a sequence for a 10-minute highlight reel. Stand by for my requests to pull specific timestamps or verbatim quotes for each segment.
Power tips
Upload multiple files at once instead of copy-pasting to maintain document structure and metadata.
Ask for specific page or document citations to verify the AI isn't hallucinating facts from its general training data.
Use a 'Chain of Thought' request by asking the AI to show its reasoning before providing the final summary.
Be specific about the output format, such as asking for a CSV-ready table or a Markdown-formatted report.
Test the context limit by asking a question about a very obscure detail located in the middle of your longest file.