Correspondence in the age of AI
A few weeks ago, I came across some correspondence addressed to my wife Kim from a friend of hers in North Carolina. The cards were sitting in a pile bound for the shredder, but before I disposed of them, I took pictures.
I uploaded the photos to ChatGPT for text extraction. Because the cards were dated, it arranged them in chronological order — and then it summarized them.
💬 A Note on This Process
What AI can do is always interesting, and amazing at times. I know Kim has more correspondence from Donna, but it's not a priority to find it. We are too busy living our best life.
I did ask ChatGPT to write a blog post about the process. It was much too wordy, but I've left it below for posterity.
📋 Narrative Summary
These cards form a short, chronological snapshot of Donna's life from mid-2023 through spring 2025 as she writes to Kim.
Work & Daily Life
Donna works in healthcare at West Point, often on overnight shifts. She prefers nights because......
Family & Personal Life
Family structure becomes increasingly central to Donna's life, especially with the t............................................daily logistics................
Overall Tone
Across all the letters, the tone remains steady and practical. The correspondence conveys resilience, humor, and a stable routine shaped by work, health management, and the arrival of a new generation in the family. What's notable is not any single detail, but the stability of the overall pattern. The writer does not drift dramatically in outlook or circumstance. Instead, the narrative shows continuity — adaptation without collapse, routine without stagnation.
📝 From Handwritten Cards to Structured Insight: Turning Personal Correspondence into Usable Narrative
(Written by ChatGPT — kept for posterity, though admittedly a bit long-winded.)
There's something deceptively simple about a stack of handwritten cards. At first glance, they look like fragments — seasonal updates, casual notes, small talk. But when examined collectively, they form something far more valuable: a longitudinal record of a life in motion.
Step 1: Extraction
The process begins with extraction. Handwriting, especially informal cursive, is inconsistent. Dates may be abbreviated, sentences run together, and punctuation is optional at best. The goal at this stage is fidelity, not interpretation. Every word is captured as written, preserving tone, quirks, and ambiguity. This matters because small details — hesitations, repetition, phrasing — often carry meaning that gets lost in early summarization.
Step 2: Transcription & Organization
Once transcribed, the material shifts from image to text. That's the first major transformation. Text can be searched, reorganized, and compared across time. Patterns begin to emerge almost immediately. Recurring topics — health, work, family — become visible threads rather than isolated mentions.
Step 3: Normalization
The next phase is normalization. Dates are aligned. Events are sequenced. Inconsistent references ("this fall," "last May") are anchored to a timeline. This step converts anecdotal fragments into a coherent chronology. Without it, narrative distortion is almost guaranteed.
Step 4: Synthesis
Then comes synthesis. This is where the raw text is compressed into meaning — not just "what was said," but "what persists across time." Several stable themes emerge:
- A chronic health condition being actively managed but not defining daily function
- A consistent work environment, with deliberate choices about schedule and workload
- A family structure that becomes increasingly central, especially with the transition to grandparenthood
- A tone that remains pragmatic, even when describing uncertainty or physical limitation
What's notable is not any single detail, but the stability of the overall pattern. The writer does not drift dramatically in outlook or circumstance. Instead, the narrative shows continuity — adaptation without collapse, routine without stagnation.
🔍 What This Makes Possible
At this point, the material has moved from "letters" to "data." That opens up several additional possibilities.
1. Timeline Construction
Events can be plotted precisely: diagnosis points, job changes, family milestones. This is useful for both personal history and medical context. It also allows cross-referencing with external records if needed.
2. Health Trajectory Mapping
Mentions of symptoms, treatments, and perceived effectiveness can be tracked over time. This creates a subjective health record that complements clinical data — particularly valuable for conditions that are chronic or slowly evolving.
3. Emotional Baseline Analysis
Sentiment across letters reveals an emotional baseline. There's no dramatic oscillation. That consistency itself is a signal — suggesting resilience, or at least a stable cognitive framing of events.
4. Social Proximity & Support Structure
Physical proximity of family (living across the street) emerges as a key structural factor. It influences childcare, emotional tone, and daily logistics. This kind of detail is easy to overlook in isolated readings but becomes obvious in aggregate.
5. Narrative Reconstruction
Perhaps the most powerful outcome: the ability to reconstruct a life narrative that was never explicitly written as one. The cards were not intended to be a story — but together, they become one.
💡 The Bigger Picture
This process highlights a broader point. Most people already generate rich longitudinal data about their lives — through emails, texts, cards, and notes. The limitation isn't data scarcity. It's lack of structure.
Once structured, even small, informal artifacts can support:
- Personal history reconstruction
- Medical context enrichment
- Family archive building
- AI-assisted memory systems
- Biographical writing
The key constraint is accuracy during the early stages. If transcription or sequencing is flawed, every downstream interpretation inherits that error. But when done carefully, the transformation is substantial.
A handwritten card is ephemeral. A structured narrative is durable. The gap between the two is not technological — it's methodological.
