Filing out extensive paperwork for a new apartment lease or a home mortgage can be a tedious and exhausting undertaking. While applicants are typically prepared to provide standard financial records, employment histories, and credit checks, there is one particular administrative hurdle that frequently catches people off guard: detailing every single past address where they have lived over the past five, ten, or even twenty years, complete with precise dates for when they moved in and out.
This meticulous requirement recently tripped up tech journalist Ben Patterson while he was completing a complex co-op application. Faced with the daunting prospect of digging through years of digital and physical files for leases dating back a decade or longer, he was left trying to mentally reconstruct his distant past. Trying to remember exact move-in dates and apartment numbers, such as when he officially moved into a specific Brooklyn apartment on Henry Street, highlights a common friction point in modern bureaucratic processes.
While Patterson eventually managed to get his address history ducks in a row through painstaking manual searches, he soon realized that there is a significantly easier and faster way to accomplish the task with the help of modern agentic artificial intelligence.
For this specific administrative challenge, users can leverage advanced AI tools like Gemini’s Spark agent, which features a dedicated Gmail connector. Utilizing such tools requires granting the AI agent permission to comb through a user’s Gmail inbox to extract relevant personal data. For individuals who harbor privacy concerns regarding letting artificial intelligence loose in their personal email archives, opting out is entirely understandable. However, for those willing to explore the capability, experts recommend reviewing configuration settings and privacy considerations beforehand to ensure comfort with data access permissions.
Once an AI agent with a Gmail connector is properly configured, users can start a new thread and supply a carefully crafted prompt to automate the search. The prompt used in Patterson’s test, originally crafted by Claude Opus 5.5, is designed to systematically comb through years of archived and active messages.
The prompt instructs the AI to search all of Gmail, including archived mail, from a specified start year to the present day. It directs the system to look for anything that shows where the user lived and when, encompassing a wide variety of documentation such as leases, rental applications, move-in and move-out notices, security deposit emails, utility and internet service start or stop notifications, USPS change-of-address confirmations, Department of Motor Vehicles or voter registration confirmations, mortgage or closing emails, insurance policies, and shifts in the "ship to" address on personal e-commerce orders.
Furthermore, the prompt instructs the AI to compile the findings into a new spreadsheet called "My address history" featuring one row per address ordered from newest to oldest. It requests specific data points including the full street address, city, state, ZIP code, the earliest and latest months the user lived there, a confidence rating based on whether the data is confirmed by a lease, closing, or USPS email or merely inferred from orders and bills, and direct links to two or three evidentiary emails proving each date.
The prompt also commands the AI to flag any gaps or overlaps exceeding one month between successive addresses and to list what kind of email would settle the discrepancy. Crucially, the instructions restrict the AI to using only addresses and dates that can be directly linked to an email, instructing it to write "unknown" rather than guess if a date cannot be pinned down. It also bars the AI from including other people’s addresses found in gift shipments, filling out or submitting any forms, signing into external accounts, or contacting third parties.
When tested on a personal Gmail account containing messages stretching back more than 20 years, the Spark agent produced a remarkably accurate spreadsheet within a matter of minutes. It successfully pieced together a comprehensive rental history going back to 2006 by cross-referencing everything from formal lease agreements and move-in notices to security deposit returns and utility confirmations.
The resulting spreadsheet helpfully specified which dates the AI was confident about and which ones required inference. For example, the agent inferred the last month lived at an oldest address based on a collection of past orders and bills discovered in the archive. While this falls short of iron-clad legal proof of a move-out date, it functions as a highly educated guess that provides the user with clear transparency regarding what is factual versus what is inferred.
This agentic AI approach is not restricted to a single platform and can be executed using several other sophisticated AI agents. Similar tests conducted with ChatGPT’s "dot" agent, Claude Cowork, and Meta Muse yielded comparable results, demonstrating that the underlying prompt strategy is broadly effective across different artificial intelligence ecosystems, though Spark’s spreadsheet output proved particularly thorough.
Leave a Reply