If you have been through a behavioral health audit, you have heard “golden thread.” People nod. Then they go back to writing notes that list a PHQ-9, a diagnosis, and “continue current treatment.”
The thread is simple. A stranger who only has the chart should be able to answer: what is going on, what you did about it today, and why this still needs a licensed clinician. If any of those is missing, the note may be true and still fail.
I am not a psychiatrist. I sit with the people who get the recoupment letters. This is the pattern in those letters.
The chain, in order
Start with the problem as it showed up today. Not the lifetime diagnosis dump. “Sleep still broken, missing two days of work, sertraline 50 mg for six weeks with partial response” is a problem. “MDD, recurrent, moderate” is a label. You need both. The label without today is how you get “not medically necessary.”
Then the assessment has to do work. Why this, still, now. What you are ruling out. What would change your mind. A sentence of actual thinking is worth more than a copied criteria list.
Then the intervention has to match. If you billed psychotherapy time, something psychotherapeutic happened and you can say what. If you changed a dose, the note says why this dose, not just the new number. If you did nothing, “nothing” needs a reason: the last change was eight days ago, side effects are settling, they have a therapy intake Friday.
Then the plan has to point at the problem you opened with. Follow up in four weeks is not a plan if you did not say what you will look for. “Recheck sleep and work function; if no movement, discuss augmentation” is a plan.
That is the thread. Problem, thinking, what you did, what happens next. A reviewer pulls one visit and walks it. They do not owe you credit for the excellent work in your head.
Where notes usually break
The assessment is last month’s paragraph with a new date. The plan is a standing sentence. The PHQ-9 dropped four points and nobody mentioned it. You billed 90833 and the body of the note is meds and “supportive conversation.”
Therapists break it a different way. Beautiful process notes, thin medical necessity. “Client processed anger toward mother” can be true and still not tell a payer why this is treatment rather than a good talk. The treatment-plan goal and how they are functioning have to show up in the same note as the intervention.
Psychiatrists break it by writing a perfect med list and a canned MSE, then leaving the “why this visit” implied. Implied does not survive a reviewer who is paid to doubt you.
Scores do not replace the thread
I like scales. Practices that never record a PHQ-9 or GAD-7 look sloppy. Practices that paste the score and skip the sentence about what it means also look sloppy. “PHQ-9 is 11, down from 16, sleep item still a 3, that matches what she described” is a thread. “PHQ-9: 11” is a number.
Same for risk. “Denies SI” on every note, including the one where they talked about giving the pills to a friend “for safekeeping,” is how you lose credibility for the whole chart.
What this has to do with AI drafts
A draft can lay the sections out and still miss the thread. Fluency is not a through-line. If the visit never connected the problem to the plan, the draft will be a tidy version of that failure.
Read the note the way an auditor will: start at the chief complaint, jump to the assessment, jump to the intervention, jump to the plan. If you have to invent the connection, edit until you do not.
We build Nextvisit to draft from the visit and to use the template you actually bill against. That helps with completeness. It does not bless a note that never makes an argument. You still have to make the argument. The point of a faster draft is that you have time left to do that before the next knock on the door.