Check every chart, not a sample
Most chart audits review a sample, often weeks after the visit. Agents can read every chart updated each day against your own checklist and give each clinician a short list of fixes by morning. This article covers why documentation carries so much weight, what agents can check, what the early evidence shows, and the safeguard that matters most: agents are configured never to write or change a clinical note.
Documentation already takes clinicians' time
In a time-and-motion study of 57 physicians in ambulatory practice, physicians spent nearly two hours on EHR and desk work for every hour of direct clinical time with patients (Annals of Internal Medicine, Sinsky et al., 2016). The study covers four specialties and is now ten years old, but it shows how little spare time clinicians have for going back over old notes.
That is one reason we keep the work on the clinician's side small: a short list of specific fixes on the chart they worked on yesterday, while they still remember the visit.
Why documentation matters for payment
CMS estimated the Medicare fee-for-service improper payment rate for fiscal year 2024 at 7.66%, or $31.70 billion. CMS says the majority of those improper payments fall into two categories: insufficient documentation, and documentation that did not show the billed service was medically necessary. It also states that improper payment estimates are not fraud rate estimates (CMS, 2024).
For a provider, most of that risk sits in ordinary gaps: an unsigned note, a missing date, a plan review that slipped, a note that does not support the units billed. These are the gaps a nightly check finds.
The trouble with samples
A typical review pulls a handful of charts per clinician or per month, checks them by hand, and records findings in a spreadsheet. Before a survey or payer audit, staff pull more charts in a hurry. Two things follow. Most charts are never reviewed, so the real deficiency rate is unknown. And the findings that do surface reach the clinician weeks after the visit.
HHS-OIG's General Compliance Program Guidance, issued in November 2023, describes auditing and monitoring as one of the elements of an effective compliance program (HHS Office of Inspector General, 2023). Checking every updated chart is one way to meet that expectation with evidence you can show.
What agents can check
Each night an agent reads every chart updated that day against the checklist your QA lead approves, built from your state licensure rules, payer policies and accreditation standards. Typical checks:
- signatures and dates on every entry
- plans of care reviewed on schedule, and reviews coming due
- required consents present
- assessments completed on time
- notes that support the billed service and units
The result is a short list per clinician and a summary by site for leadership. A pre-bill check can hold a claim for review, and your billing and compliance staff decide what happens to it.
What the early evidence shows
The evidence that language models can catch documentation errors is early. In one study reported in 2026, researchers injected errors into synthetic oncology cases. Gemini 2.5 Pro detected 97.8% of them and GPT o4-mini-high 87.8%, against a mean of 47.8% for oncology specialists (JCO Clinical Cancer Informatics, via Targeted Oncology, 2026). It is a single study in one specialty using synthetic cases, so it says little about how a model performs on your charts. That is why we test the checks on your own records, with your QA lead confirming findings, before anyone relies on them.
The safeguard
Agents are configured never to write or change a clinical note. When a check finds a gap, the agent flags it with the evidence, and the clinician who wrote the note corrects it. Your QA lead confirms findings and handles disputes. Judgment about whether a plan of care is clinically right stays with your clinical director. Every flag is logged with the chart, the rule it failed and the text the agent relied on.
The same logic applies in billing. Agents can point out a note that does not support a code, and the decision belongs to your coders. See the revenue cycle page for how code changes are approved, the compliance page for how findings feed your compliance program, and the human review page for how people supervise agents.
What we measure
- Share of updated charts reviewed
- Deficiency rate by site and by rule
- Days from flag to correction
- Pre-bill holds and how they were resolved
We start from a dated baseline. Where no reliable count exists today, the first weeks of checks establish one.