“Keep track of it and let me know at the next visit” is one of the most common instructions in medicine, and one of the hardest to follow well. Two weeks later, what you actually have is a handful of scattered notes-app entries, a memory of “it was worse on the weekend, I think,” and no way to tell your clinician anything more precise than that. The record fails not because you did not notice things - it fails because nothing enforced a consistent format in the moment.
This is a method for building that format with AI’s help, while keeping a hard line around what the log contains: observations, not conclusions. A model can help you design a consistent structure and organize entries you already wrote. It should never generate a severity score, a likely cause, or an opinion about whether something needs urgent attention.
Do not ask AI to score how serious an entry sounds, guess at a cause, or tell you whether a pattern needs urgent attention. If your care team gave you specific red-flag instructions - symptoms or combinations that mean “call us” or “seek care now” - follow those instructions exactly as given, copied verbatim into your log. Do not let a model generate its own version of that guidance.
Step 1: Design the format once, before you start logging
The most common failure in symptom tracking is changing the format halfway through, which makes two weeks of entries impossible to compare against each other. Design the columns once, with AI helping you turn a vague tracking instruction into a specific structure.
My clinician asked me to track [what you were asked to observe] between
now and my next appointment. Help me design a simple log format with
columns for: date and time, what I observed, the context (what I was
doing, any relevant timing), the impact on my day, any action I had
already been instructed to take, and the source (my own observation vs.
a device reading). Do not suggest what I should be looking for beyond
what my clinician already told me to track.
Keeping “source” as its own column matters more than it looks: a symptom you noticed yourself and a reading from a device (a thermometer, a blood pressure cuff, a glucose meter) carry different kinds of certainty, and your clinician will want to know which is which.
Step 2: Log in the moment, in your own words
Write each entry as close to the moment as possible, in plain factual language - what happened, when, and what you were doing. Resist writing a conclusion into the entry itself (“probably from the medication” or “definitely worse than last time”). Those are exactly the interpretations your clinician is positioned to make once they see the pattern; baking them into the raw entry can quietly bias how you read your own data later.
Date/time: [when]
Observation: [what happened, factually]
Context: [what you were doing, any relevant timing - e.g. before/after
a meal, medication dose, activity]
Impact: [how it affected your day - e.g. missed part of an activity,
no impact, had to stop what I was doing]
Action already instructed: [only if your clinician gave you a specific
instruction for this situation]
Source: [my own observation / device reading, and which device]
Step 3: Use AI to organize, not to conclude, between entries
Periodically - weekly, or before an appointment - ask a model to organize your raw entries into a clean table, explicitly instructing it not to draw conclusions from the pattern.
Here are my raw log entries from the last [period]: [paste entries].
Organize these into a clean table with the columns: date/time,
observation, context, impact, action instructed, and source. Do not
identify patterns, suggest a cause, estimate frequency trends, or
comment on severity - only reformat what I already wrote into a
consistent table.
If you notice a pattern yourself while reviewing the organized table - it happens more on weekdays, or after a specific activity - write that observation down as your own note, clearly separated from the factual log rows, and bring it up directly with your clinician rather than asking a model to confirm or interpret it.
Step 4: Keep clinician-given red flags exact and separate
If your care team gave you specific instructions about when to seek care sooner - a fever above a certain point, a symptom lasting more than a certain number of days, a specific combination to watch for - copy those instructions verbatim into a clearly labeled section of your log, exactly as given. Do not paraphrase them through AI, and do not ask a model to generate its own version of “when to worry.” Clinician-issued red flags are instructions to follow, not content to summarize.
Step 5: Bring the organized log, not just a verbal summary
A clinician reviewing a clean, consistently formatted table can spot patterns far faster than listening to a verbal recap of two weeks. Combine this log with the appointment-preparation method in preparing for a medical appointment in 20 minutes to turn the table into a one-page summary alongside your questions. If the log is spreadsheet-based rather than a simple table, AI for spreadsheets covers the formatting and formula side of that setup.
The full symptom impact log gives you this structure ready to use, with a dedicated section for clinician-given red flags kept separate from your own entries.
A symptom log accumulates detailed personal health information over weeks. Store it somewhere you control, avoid pasting the full running log into a general AI chat repeatedly (a short recent excerpt is usually enough for organizing), and see what ChatGPT remembers, sees, and shares for what stays in a chat history by default.
Choosing what counts as an entry
Not every log needs the same granularity, and getting this wrong in either direction undermines the record. Logging every minor fluctuation produces so much noise that a real pattern gets buried in it; logging only the days that felt notable produces gaps that make frequency impossible to judge later. If your clinician did not specify a threshold, ask AI to help you turn a vague instruction into a concrete one you can apply consistently - the decision itself should still be yours.
My clinician asked me to track [what you were asked to observe] but
didn't specify exactly when something counts as worth logging. Help
me think through two or three concrete threshold options I could set
for myself (for example: log anything noticeable vs. log only when it
affects an activity) - as options for me to choose between, not a
recommendation for which one is correct.
Whichever threshold you choose, apply it consistently for the whole tracking period, and note the threshold itself at the top of your log so your clinician knows what “logged” actually means in your specific record.
Common pitfalls
- Changing the format partway through. Decide the columns once, before logging starts, so two weeks of entries stay comparable.
- Writing a conclusion into the raw entry. “Probably the new medication” belongs in a separate notes section you discuss with your clinician, not in the factual observation field.
- Asking AI to spot patterns or estimate severity. That is exactly the judgment your clinician is positioned to make with the full picture; a model reading a log out of context is not.
- Paraphrasing clinician-given red flags. Copy them exactly, and follow them exactly - do not let a model’s rewording introduce ambiguity into an instruction that matters.
- Letting the log replace calling for care when something red-flag matches. The log is for pattern-spotting between visits, not a substitute for acting on instructions you were already given.
Start this week
Design your log format once using the prompt above, start entering observations in the moment rather than reconstructing them later, and organize the raw entries into a clean table before each appointment. Use the symptom impact log to keep the structure consistent from the first entry to the last.
