PDPM Section GG documentation 2026: AI scribe for SNF self-care and mobility coding

May 8, 2026 · 7 min read

Section GG of the MDS 3.0 drives PDPM reimbursement in skilled nursing. The 18 self-care and mobility items use a 6-point coding scale, and inaccurate documentation costs real money — CMS audits in 2024–2025 found average downcoding losses of $20–$60 per resident per day in mid-stay reassessments. Across a 60-bed facility, that's $400k–$1.3M/year lost.

AI scribe tools designed for primary care don't handle Section GG. Here's what works for SNF in 2026.

Section GG quick refresher

Items split into Self-Care (eating, oral hygiene, toileting hygiene, etc., 7 items) and Mobility (sit-to-stand, transfers, walking, stairs, etc., 11 items). The PDPM function score is computed across these and feeds the Nursing and PT/OT case-mix groups.

Where downcoding happens (real audit findings)

• "He can do it himself, the staff just helps because it's faster." Documented as 06 (independent) when the resident actually requires touching assistance → should be 04. Loss: ~$25–$45/day in nursing CMG.
• Toileting hygiene scored only at SOC. If status changes, mid-stay GG reassessment must capture it. Most SNFs don't reassess proactively.
• "Refused" coding without 88-reason documentation. Audit finding: refused without documented attempt = full points lost.
• Walking 50 feet vs Walking 150 feet. Two separate items. Coding both at the same level when resident clearly slows on the longer distance is common over-coding (audit risk).
• Roll left and right scored to the resident's stronger side only. Should be averaged or scored to weaker side per CMS guidance.

Why generic AI scribes break on Section GG

Primary-care scribes summarize visits into SOAP-style narratives. GG documentation is structured: each of 18 items needs a discrete numeric code with a supporting evidence note. The narrative format produces text like "patient required moderate assist with toileting" — but doesn't generate the score 03 nor flag whether it's GG0130C (toileting hygiene) vs GG0170A (roll left/right).

What you actually need: a scribe that listens to therapy/nursing observations, identifies which of the 18 GG items each statement maps to, suggests the score, and surfaces missing evidence for items the encounter didn't address.

2026 vendor reality for SNF Section GG

DIY Section GG prompt template

You are a Section GG documentation assistant for a skilled nursing facility.

Below is a transcript of a caregiver-resident interaction during ADL care.
Identify which of the 18 GG items the observations map to.
For each, suggest a score (06/05/04/03/02/01) with the supporting quote.
If an item is mentioned but evidence is ambiguous, mark it FLAG.
For items not mentioned in this interaction, list them as MISSING.

GG items to consider:
- GG0130A Eating
- GG0130B Oral hygiene
- GG0130C Toileting hygiene
- GG0130E Shower/bathe self
- GG0130F Upper body dressing
- GG0130G Lower body dressing
- GG0130H Putting on/taking off footwear
- GG0170A Roll left/right
- GG0170B Sit to lying
- GG0170C Lying to sitting on side of bed
- GG0170D Sit to stand
- GG0170E Chair/bed-to-chair transfer
- GG0170F Toilet transfer
- GG0170G Car transfer
- GG0170I Walk 10 feet
- GG0170J Walk 50 feet with two turns
- GG0170K Walk 150 feet
- GG0170L Walking 10 feet on uneven surfaces

Output JSON:
{
  "scored": [{"item": "GG0130A", "score": "04", "evidence": "..."}, ...],
  "flagged": [{"item": "...", "reason": "..."}],
  "missing": ["GG..."],
}

Transcript:
{transcript}

Pair this with Whisper transcription of caregiver narration during care delivery. Caregivers narrating "I'm helping you sit up... now we'll roll to the left side... let's get those legs over..." gives the scribe enough to score 5–8 items per care episode.

The PDPM assessment windows

The 5-day window is where AI scribe ROI concentrates — that's when GG drives 30+ days of payment. Sloppy GG at 5-day costs the entire stay's revenue.

Cost vs revenue math

• Average SNF stay: 27 days under PDPM
• Average daily rate: $580 (2026 national)
• 1-point downcoding on 3 GG items: ~$45/day lost × 27 days = $1,215 per stay
• 60-bed facility, 200 admissions/year: ~$243k/year if even 50% of admits have a 1-point GG error

An AI scribe correctly catching 80% of those errors saves ~$195k/year. Tool cost (Apricot, Olli, or DIY): $5k–$30k/year. ROI is 5–40×.

Audit risk to track

• OIG and CMS focus on Section GG in 2024–2026 audits.
• Pattern flags: many residents with identical GG scores at SOC; sudden status improvements without therapy notes; refused-coding without 88-reason.
• An AI scribe that randomizes scores to look "human" is a worse problem than over-coding. Don't add randomization — document the actual evidence.