Clinical Documentation Integrity · AI-Driven Queries · Clinical Validation
The Query That Never Went, and the One That Got Answered in Four Seconds
The 2026 query standard draws one line through both failures. Most query tools sit on the wrong side of it.
Under-querying understates how sick the patient was. Over-querying manufactures a diagnosis the record cannot carry. The standard released on September 1 treats both as failures, and the fix for both is the same test.
A CDI (clinical documentation integrity) specialist reads a chart on day three. The nursing assessment describes a patient who cannot be roused for most of a shift. The labs are abnormal. The physician's notes say nothing about it. Her checklist says not to query, because the indicator came from nursing and a query must not create a diagnosis. She closes the chart. The record goes out understating how sick the patient was.
Sixty days after a different patient went home, a query lands on a physician's screen. An AI platform combing the closed chart flagged an isolated white count and suggested sepsis. She is between patients. The interface offers a checkbox. She checks Yes. The record system logs that a 200-page chart was opened and answered in four seconds.
Neither clinician was careless. Each did what her tool asked. Both cases are composites, usually discussed as opposites: one program too timid, the other too aggressive. They are the same failure. In each, the decision to query turned on where the indicator came from, or on what the tool suggested, instead of on what the record supports.
That is the line the AHIMA and ACDIS Guidelines for Achieving a Compliant Query Practice, 2026 Update (AHIMA, the American Health Information Management Association, and ACDIS, the Association of Clinical Documentation Integrity Specialists) now draw. Released September 1, 2026, it supersedes every prior version, including the 2022 Update most query tools were written against. It names the queries a program is expected to send, and makes the organization answerable for every query it delivers.
What the standard actually says
Clinical indicators may be sourced from anywhere in the encounter, including nursing and ancillary assessments. Where providers conflict, the attending is queried. And process owners retain accountability for every query delivered, including those generated by automated or AI-assisted processes.
01
What Changed on September 1
The standard movedThe 2026 Update applies across inpatient, outpatient, ambulatory, and professional-fee settings, and it says plainly that an organization should not assume a vendor-supplied or EHR-integrated (electronic health record) tool produces compliant queries by default; templates are audited regularly. A checklist that still carries the 2022 prohibitions is an item an auditor can ask about.
Both directions matterA tool that lists only prohibitions guards against one error and produces the opposite one. Under-documentation weakens risk adjustment, quality measurement, and the medical necessity defense on the short stays that draw the most scrutiny. Over-capture books a diagnosis no one managed. The standard sets out when a query is required as carefully as when it is barred, and adds that high query volume usually signals an education need, not a need to keep querying.
The Query That Never Went, and the One That Got Answered in Four Seconds
September 16, 2026
02
The Rules That Reversed, and the One That Replaced Them
What the old list got wrongFour common prohibitions are contradicted or narrowed by the standard, and three suppress queries it requires. Nursing documented it is a valid source of indicators. Another provider documented it splits in two: with no conflict there is nothing to query, but where providers conflict a query to the attending is required. Only an indicator is present is narrowed, not barred, since a supported, reportable, undocumented condition is the leading reason to query. And a technology suggested it changes nothing, because the organization is accountable either way.
The construction testThe constraint that survives is about how the query is built, not where the indicator came from. Present the value as it appears and do not interpret it: a heart rate of 120 is reported as a heart rate of 120, never as tachycardia; a hemoglobin of 10 g/dL is never written as anemia. Quote the chart and source every indicator. Offer only clinically relevant options, including that the condition is not present. Never let reimbursement, quality measures, or denial mitigation appear in the query or motivate it. The four-second query fails on every line; the unsent day-three query would have passed.
| Feature | A query the standard expects | A query that will not survive validation |
|---|
| Trigger | A supported, reportable condition not yet documented; or documentation that is unclear, incomplete, or conflicting | An isolated lab or vital shift with no management; a consultant's added specificity with no conflict; anything chosen because it raises the DRG (diagnosis-related group, the payment class) |
| Timing | Concurrent, or shortly after discharge and before billing, under a written policy on when post-bill queries are permitted | RiskWeeks to months after discharge or billing, with no policy governing it |
| Clinical basis | Indicators quoted from the record, plus the treatment and decision-making a real diagnosis carries | A pattern match: elevated creatinine becomes acute kidney injury; a transient vital sign becomes sepsis |
| Format | Neutral, multi-option, including "not clinically supported" and "other" | Leading, binary Yes/No checkbox for a post-hoc addition |
| Documentation | A narrative addendum that carries the provider's reasoning into the record | A checkbox signed with no change to the clinical notes |
The table is the standard's line. What puts a query on the wrong side of it, on either side of the claim, is almost always the same mistake, and it is the subject of the next page.
Worth knowing when queries surface in an audit
A provider's response to a clinical validation query represents that provider's clinical judgment at the time of documentation. Under the 2026 Update it is not stand-alone grounds for post-payment recovery, and not evidence of fraudulent billing, absent other findings. That protects a well-built query. It does nothing for one whose only basis is the checkbox.
4 of 8
Prohibitions corrected
Contradicted or narrowed by the standard; three would have suppressed required queries
$5.7B
Health care share of FCA recoveries
Of $6.8 billion in Department of Justice False Claims Act recoveries, fiscal year 2025
Jan 1, 2026
California AB 489 in effect
AI tools may not imply that a licensed human is providing the assessment; each use is a separate violation
The Query That Never Went, and the One That Got Answered in Four Seconds
September 16, 2026
03
The Category Error Underneath Most Denials
An indicator is not a diagnosisNearly every manufactured-diagnosis dispute traces to one confusion: treating a clinical finding as a disease state. Retrospective logic is linear. It matches a value to a diagnosis and trusts the match. An elevated creatinine is an indicator. Acute kidney injury is a physician's reasoning connecting that indicator to an etiology, a severity, and a treatment response. Without the treatment linkage, the diagnosis fails the criteria for reporting additional diagnoses and is stripped on validation.
Where it shows upThree conditions account for most downcoding and recoupment in this category. Sepsis, challenged where sustained organ dysfunction is absent and vitals resolved after a modest fluid bolus. Acute metabolic encephalopathy, challenged where mild lethargy or baseline dementia is treated as acute organ failure. Severe malnutrition, challenged where dietitian intervention and body-mass evidence are missing. In each, the dispute is never about the code. It is about whether anyone managed the condition.
| Feature | Defensible | High risk |
|---|
| Clinical basis | Documented indicators, active medical decision-making, therapeutic intervention | An isolated lab or vital shift with no active management anywhere in the record |
| Construction | Indicators sourced to the chart, values reported as found, multiple choice with a way to say something else | Interpretation supplied for the signer; a binary agreement box for a new diagnosis |
| The response | A narrative addendum that gives the reasoning and how it affected care | A checkbox signed with no change to the clinical note |
| Who is asked | The provider who delivered direct face-to-face care | A physician advisor, medical director, or other administrative role standing in for the treating provider |
| Repeat querying | A follow-up only where new clinical indicators have come to light | A second query for the same condition sent solely to override the first answer |
| The addendum | Identifies itself as a delayed entry, dated and attributed, original content preserved | An entry that cannot show when it was made or what the record said before |
What This Asks of Both Sides
For a health system, three things follow. Set a written timing policy and hold to it, since the standard leaves timing to you and an unwritten practice is the one that looks improvised in an audit. Configure the tools so a query for sepsis, encephalopathy, or malnutrition cannot be generated without documented indicators and treatment in the chart. And track the reversal rate of diagnoses captured this way, because a tool whose suggestions are routinely stripped on validation is not a revenue source, it is an audit exposure with a dashboard attached.
For a payor, the obligation is symmetrical and less often stated. Where a high-weighted DRG rests on a diagnosis with no management footprint, the recovery is correct and will survive appeal. Where the clinical pillars are present, the claim is a correct payment and denying it manufactures appeal cost, provider abrasion, and improper-denial exposure. A recovery overturned on appeal is phantom savings. It consumed review time and damaged a relationship without protecting a dollar.
None of this is an argument for abandoning the tools. It is an argument for putting a human expert back in the loop structurally rather than symbolically: every suggested diagnosis ratified by someone who confirms it appears in the physician's reasoning and management plan before the claim drops, and queries that ask for a brief clinical rationale rather than a binary agreement. That is the same move on both sides of the claim, away from industrial-scale retrospective capture and away from reflexive denial, toward validation done while the record can still speak for itself.
The Query That Never Went, and the One That Got Answered in Four Seconds
September 16, 2026
04
The Exposure, and the Numbers That Hide It
The provider's exposureA query response is an attestation, not paperwork. The Department of Justice reported more than $6.8 billion in False Claims Act settlements and judgments in fiscal year 2025, over $5.7 billion of it involving health care, and a diagnosis added after discharge with no clinical footprint is the kind of claim enforcement reads as knowing. Exposure runs through the False Claims Act (31 U.S.C. § 3729) and civil monetary penalties (42 U.S.C. § 1320a-7a). California's AB 489, effective January 1, 2026, adds a narrower risk: an AI tool whose interface or marketing implies a licensed human is providing the assessment violates the statute, the deployer is liable with the developer, and each use is a separate offense.
The numbers reported upwardQuery agreement rate is equally consistent with excellent indicator selection and with leading construction, which makes it a screening signal and never a finding. Case mix index moves with service-line volume. Capture rates move when the cohort gets sicker. None of those numbers can tell you whether a diagnosis was managed, which is the only question that decides the claim, on either side of it.
An algorithm can surface a diagnosis in seconds, and another can deny it as fast. The audit trail shows whether anyone thought about it for longer than four, and whether anyone asked when the record said they should. Ask whether you would still believe the diagnosis if there were no financial consequence at all.
Where MHMDAA Comes In
MHMDAA (Michael Hill, MD and Associates) is a revenue cycle consulting and payor-provider dispute resolution firm, retained by payors and providers alike, which is why this reads as method rather than argument. A query that cannot be defended on the record it cites is a problem for the hospital that sent it and for the reviewer deciding what it proves. The work is unglamorous: pull the query tool your teams actually use, set it against the 2026 Update line by line, and write down what you found and changed.
| Phase | What the engagement does |
|---|
| 1. Tool review | Checklist, templates, and any AI or EHR query generator reconciled against the 2026 Update; review documented |
| 2. Construction audit | Sent queries scored on the construction test; unsent charts scored for required queries that were suppressed |
| 3. Late-query governance | A written post-discharge and post-bill query policy, which the standard requires the organization to define, with approval steps and a hard boundary on new high-severity diagnoses once billed |
| 4. Metrics realignment | Agreement rate and case mix expansion replaced as program targets by construction accuracy, validation survival, and query volume by provider read as an education signal |
Michael Hill, MD & Associates
Revenue cycle consulting and payor-provider dispute resolution · (877) 464-4556 · info@mhmdaa.com
Neutrality. This brief is educational commentary; it is not legal, coding, or clinical advice, and it is not a determination about any specific claim, payor, provider, or matter. Scenarios described as composites are illustrative of method only and are not findings about any party. MHMDAA is an independent revenue cycle consulting and payor-provider dispute resolution firm; its principals provide dual-perspective analysis and independent expert services to payors and providers alike, and nothing in this publication constitutes advocacy for either side of any dispute or prejudges any question on which they may serve as independent experts. Authorship. This document was drafted with human authorship and may include AI-assisted formatting or summarization. All analysis, conclusions, and opinions are solely those of Dr. Michael Hill.
Sources
- ACDIS and AHIMA. Guidelines for Achieving a Compliant Query Practice, 2026 Update, released September 1, 2026; supersedes all prior versions, including the 2022 Update.
- U.S. Department of Justice. False Claims Act Settlements and Judgments Exceed $6.8 Billion in Fiscal Year 2025, January 16, 2026. 31 U.S.C. § 3729; 42 U.S.C. § 1320a-7a. California AB 489 (Ch. 615, Stats. 2025), effective January 1, 2026.
- ICD-10-CM Official Guidelines for Coding and Reporting, Section III. MHMDAA. CDI 2.0 Pre-Query Clinical Appropriateness Checklist, v2.1, September 3, 2026, and internal papers on late sepsis queries and physician liability for AI-generated diagnoses.
