Why longitudinal context may be one of the most important next steps for AI documentation in psychiatry
By Allison Sikorsky, DNP, PMHNP-BC
Founder & CEO, PMHScribe
Most AI scribes are very good at one thing: listening to the visit happening right now. Which is very useful. But psychiatry rarely happens in one isolated visit.
We treat people over time. We look for patterns. We compare today with last month. We remember what changed after a medication adjustment, whether sleep improved, whether panic attacks became less frequent, whether a patient followed through on something we discussed, or whether the same concern is coming up again.
So I think one of the most important questions in AI documentation is becoming: Should an AI scribe understand what happened at the last visit too? I think the answer is yes, but carefully.
A follow-up visit only makes sense in the context of what came before it. If a patient says, “I’m sleeping a little better,” the clinically useful question is: better than when? If they say, “The medication is helping,” we need to know what symptoms we were targeting. If anxiety is still present, is it unchanged, improving or worsening?
This is basic psychiatric care, but it highlights an important limitation of documentation tools that treat every encounter as if it exists in a vacuum. A good note should tell the story of change over time.
That is especially true in medication management, where the reason behind a treatment decision often depends on the trajectory of symptoms across multiple visits.
When an AI scribe only knows what was said today, it has to build the entire note from the current conversation. That can work well for a new patient evaluation.
For follow-up care, it can create extra work. The clinician may still need to remember or manually pull forward:
what medication was changed last time
which symptoms were improving
what side effects were being monitored
whether a safety concern was new or already known
what goals were set at the prior appointment
what the patient was supposed to follow up on
Without that context, the note may be accurate for today but less useful clinically. That is an important distinction. Documentation should not only record what happened during an encounter. It should help explain what changed.
There is an obvious risk here too. An AI tool with access to prior notes should not simply pull old information forward automatically.
That is how outdated information becomes permanent. A symptom present six months ago may no longer be relevant. A medication may have been discontinued. A historical concern may not belong in every note forever.
Clinicians have dealt with this problem for years through copy-forward and templated documentation. AI should not recreate the same problem at greater speed. The goal is not unlimited memory.
The goal is useful context.
If AI documentation is going to become more useful in psychiatry, I think prior-visit context should help answer a few simple questions:
What changed since the last visit?
What stayed the same?
What was the clinician monitoring?
What decision was made previously, and what happened after it?
That is far more useful than automatically importing an entire previous note. For example, if a medication dose was increased at the last visit, the AI should recognize that the clinician is now assessing response and tolerability.
If sleep was the main concern previously, it should help identify whether sleep improved.
If a safety plan was discussed, the clinician should be able to see whether that issue remains active without searching through multiple old notes.
That kind of continuity can make documentation both faster and more clinically meaningful.
There is another reason this matters in behavioral health. Psychiatric language is nuanced.
A patient may describe symptoms differently from visit to visit. They may reconsider something they said previously. A diagnosis may evolve. A concern that seemed significant at one appointment may turn out to be situational.
An AI system should not assume that something documented previously remains true simply because it appeared in the chart. Prior information should provide context, not dictate the current assessment. The clinician still has to decide what is relevant today. That is where human review remains essential.
One of the most repetitive parts of psychiatric documentation is rebuilding context that the clinician already knows.
We often spend time documenting:
“Last visit, we increased…”
“Since the previous appointment…”
“Patient previously reported…”
“Symptoms have improved from…”
These details matter. But they are also exactly the kind of structured continuity that technology should be able to support intelligently.
If done well, longitudinal context could make follow-up notes clearer, not longer. The note could become more focused on the actual clinical story:
Here is where the patient was.
Here is what we changed.
Here is what happened.
Here is what we are doing next.
That is the kind of documentation I want to read.
The first generation of AI scribes focused heavily on capturing the current conversation. That made sense. But as the technology matures, I think the more important question will be whether these tools can support continuity of care.
Not by replacing the clinician’s memory.
Not by blindly copying prior notes.
But by helping surface the right historical context at the right time.
Psychiatry is built on patterns, progression and change. Our documentation tools should understand that too.
Should an AI scribe have access to previous psychiatric notes?
Potentially, yes. Prior notes can provide useful clinical context for follow-up visits, but access should be handled carefully and should not result in outdated information being automatically copied into new documentation.
Why is longitudinal context important in psychiatric documentation?
Psychiatric treatment often depends on changes over time, including medication response, symptom progression, side effects, safety concerns and treatment goals. Longitudinal context helps clinicians document those changes more clearly.
Can AI scribes compare current symptoms with previous visits?
Some AI documentation tools are beginning to support more contextual workflows, but capabilities vary. The most useful systems should help clinicians identify relevant changes without replacing clinical judgment.
What is the risk of AI using prior notes?
The biggest risk is carrying forward information that is outdated, inaccurate or no longer clinically relevant. Prior documentation should be used as context, not treated as automatically true.
How can AI improve psychiatric follow-up notes?
AI can help surface prior treatment decisions, symptom trends, medication changes and unresolved concerns so clinicians can focus documentation on what has changed and what matters now.
Will AI replace the clinician’s role in reviewing patient history?
No. AI can help organize and surface relevant information, but the clinician still determines what is accurate, current and clinically meaningful.
Allison Sikorsky, DNP, PMHNP-BC, is a board-certified Psychiatric Mental Health Nurse Practitioner and Founder & CEO of PMHScribe. Her work focuses on behavioral health documentation, clinician workflow, responsible AI and building technology that supports thoughtful, efficient psychiatric care.
