Resource, September 2026

AI and the Nursing Home Survey

NELS Summit Archive. Published 18 September 2026. Written for administrators and emerging leaders in long-term care; not affiliated with CMS, any state survey agency, or any vendor listed.

Software that promises "survey readiness" is multiplying, and much of it now says it uses AI. Before judging any of it, it helps to be precise about what the survey actually is, what surveyors actually cite, and what changed in September 2026. This page puts those facts in one place, from primary sources, and then lays out what the current tools say they do.

1. How the standard survey works, and how often

Every Medicare- or Medicaid-certified nursing home is subject to a standard survey conducted without prior notice. The Social Security Act makes tipping off a facility about a survey date punishable by a civil money penalty of up to $2,000.¹

Frequency. The statute requires each skilled nursing facility to receive a standard survey no later than 15 months after its previous one, and the statewide average interval between standard surveys may not exceed 12 months.¹ A standard or abbreviated survey may also be run within 2 months of a change of ownership, administration, management, or director of nursing.¹ Complaint investigations happen on top of this, using abbreviated portions of the survey process.²

Extended surveys. A facility found to have provided substandard quality of care on a standard survey must receive an extended survey, immediately afterward or no later than 2 weeks after the standard survey is completed.¹

Backlogs are real. The federal Office of Inspector General found that as of 31 May 2021, 71 percent of nursing homes (10,913 of 15,295) had gone at least 16 months without a standard survey, with state backlogs ranging from 22 to 96 percent.³ In July 2026 CMS wrote that the federal survey budget has not increased since 2015 while state agencies' complaint-survey obligations rose more than 20 percent, and that this has produced a backlog of homes that have not received their statutorily required standard survey.² In practice, "annual survey" is a rule of thumb, not a guarantee of timing.

What surveyors do on site

State surveyors follow the Long-Term Care Survey Process (LTCSP), which CMS implemented in 2017.² The current LTCSP Procedure Guide (effective 23 June 2026) sets out the steps:⁴

1. Offsite prep. The team coordinator reviews the facility's information before arrival, including complaints and facility-reported incidents that may be folded into the survey, and makes the mandatory facility task assignments.
2. Entrance. On arrival the team coordinator asks the administrator for the information needed immediately, from the entrance conference list, and surveyors go to their assigned areas.
3. Initial pool. Surveyors screen residents room to room without staff and observe, interview, and do a limited record review. The guide's own example: with a recommended team of four surveyors, each has about eight residents, an initial pool of about 32.
4. Sample selection and investigations. The team finalizes the sample (including closed records) and investigates each sampled resident's care areas in depth.
5. Facility tasks. These include dining, infection control, SNF beneficiary notification review, kitchen, medication administration, medication storage and labeling, the resident council interview, sufficient and competent nurse staffing, personal funds, environment, resident assessment, binding arbitration agreements, and QAPI/QAA review.
6. Deficiency determination and exit. The team decides what to cite, holds an exit conference with the facility, and creates the citations, which appear on the survey report (Form CMS-2567).

The interpretive guidance surveyors apply to each requirement, including the F-tag definitions, is Appendix PP of the State Operations Manual, published on CMS's nursing homes guidance page along with the current survey resources.⁵

2. The risk-based survey that began 8 September 2026

On 16 July 2026 CMS issued memo QSO-26-14-NH, directing state agencies to begin using a Risk-Based Survey (RBS) at qualifying nursing homes nationwide from 8 September 2026.² The key facts, from the memo itself:

• The RBS is a modified LTCSP standard survey: a streamlined review of all required areas with fewer activities and a smaller resident sample, conducted in roughly half the time with fewer surveyors.²
• It was tested from 2023 in 22 states and over 100 facilities, and CMS reports its findings were comparable to the traditional LTCSP.²
• All facilities remain subject to a standard survey at least every 15 months.²
• To qualify, a facility must not have: less than a 5-star overall rating; less than a 3-star staffing rating; any actual harm, immediate jeopardy, or substandard quality of care citation in the last survey cycle; more than 18 months without a standard survey; a staffing waiver; a failed Payroll-Based Journal or MDS audit; a health inspection score above the state 50th percentile; two or more residents aged 65 or older coded with schizophrenia after admission without that diagnosis; a change of ownership since the last standard survey; or Special Focus Facility candidate status.²
• Using June 2026 data, CMS counted 1,560 of 14,682 facilities (12.01%) as qualifying. The single most common exclusion was an overall rating below 5 stars (11,639 facilities, 78.39%), followed by a staffing rating below 3 stars (5,692, 29.43%).²
• A listed facility is disqualified before the survey starts if, among other things, it picks up a harm, immediate jeopardy, abuse (at any level) or substandard quality of care citation on an intake investigation, or has an intake pending triaged at immediate jeopardy.² CMS's guidance page adds that if concerns about resident safety arise during an RBS it is immediately expanded, and that the RBS does not apply to complaint surveys.⁵
• Qualifying facilities get an icon on Nursing Home Care Compare, and the qualifying list is to be published on the Provider Data Catalog beginning 30 September 2026.²

For administrators, the practical point is that the RBS turns ordinary data (ratings, PBJ staffing, MDS accuracy, citation history) into a survey-intensity lever. Any tool that claims to help with "the survey" in 2026 should be judged partly on whether it helps with those inputs.

3. The most-cited F-tags nationally

Vendor blogs routinely publish "top F-tag" lists without saying where the numbers came from. We computed ours directly from CMS's Health Deficiencies dataset on the Provider Data Catalog (file NH_HealthCitations_Aug2026.csv, processing date 1 August 2026). The file lists every health citation from each facility's three most recent inspection cycles: 419,479 citations at 14,627 facilities, across 267 distinct F-tags, with survey dates from 23 March 2017 to 23 July 2026. It includes both standard-survey and complaint citations.⁷

Three things stand out:

• The list is broad, not concentrated. The top ten tags together account for 35.3% of all citations. Readiness built around a single "top tag" misses most of what gets cited.
• Infection control reaches almost everyone. Four in five facilities in the file (81.7%) have at least one F880 citation across their last three cycles.
• Standard surveys and complaints cite differently. 68.7% of citations are marked as standard-survey only, 24.3% complaint only, and 7.0% both. Among citations from standard surveys (317,371), the leaders are F880 (6.3%), F812 (5.9%), F761 (3.9%), F689 (3.6%) and F656 (3.5%): kitchen and medication storage matter more on the recertification visit than the all-citation ranking suggests.

4. Scope and severity: where citations land on the grid

Every citation carries a letter from A to L. Severity runs from level 1 (potential for minimal harm) to level 4 (immediate jeopardy); scope is isolated, pattern, or widespread.⁶ Substandard quality of care (SQC) is defined in SOM Chapter 7 as deficiencies under specific resident rights, abuse, quality of life, quality of care, behavioral health, pharmacy, administration and infection control requirements that reach immediate jeopardy, a pattern of or widespread actual harm, or widespread potential for more than minimal harm.⁶ Counts below are our tally of the same CMS file.

Level 2 accounts for 92.3% of all citations, with D (isolated, no actual harm) alone at 63.1%. Actual harm (G to I) is 3.2%, and immediate jeopardy (J to L) is 2.3%, or 9,480 citations. Those 9,480 immediate jeopardy citations are spread across 4,404 facilities, 30.1% of the facilities in the file. They concentrate in a few tags: F689 accidents and supervision (28.2% of all immediate jeopardy citations), F600 freedom from abuse (14.8%), and F684 quality of care (7.9%). A level A deficiency is not written on the CMS-2567, which is why the file has none.⁶

Why this matters for the RBS: a single G-level or higher citation, or any SQC, in the last survey cycle disqualifies a facility from the risk-based survey.²

5. AI and software tools for survey readiness, and what each claims

The tools below market themselves for skilled nursing survey readiness or compliance and describe AI in their own materials. Everything in this section is the vendor's claim, as stated on the linked page when we read it in September 2026; we have not tested these products, and inclusion is not an endorsement. We found no independent, published evaluation of any of them against actual survey outcomes.

The free baseline every tool is built on

Most "survey intelligence" starts from public CMS data that any administrator can download: the Health Deficiencies file used above, the rest of the nursing home datasets on the Provider Data Catalog, and each facility's profile on Care Compare. A spreadsheet of your own state's citations by tag and severity is a reasonable benchmark to hold any vendor's risk score against.

Questions to ask any vendor

• What data does the model actually read: the 2567 history, the live EHR, MDS, PBJ, or only what staff upload?
• Does it predict citations, or classify findings you already have? Ask for evidence tied to real survey outcomes, not a demo.
• Does it reflect the current LTCSP Procedure Guide and Appendix PP, and how fast does it update when CMS revises them?
• Who reviews AI-drafted plans of correction before they go to the state agency?
• How does it handle PHI, and will the vendor sign a BAA?

6. A survey-readiness workflow that uses the data

This is a practical sequence built from the survey process and the citation data above. Each step names where software, AI or otherwise, can reasonably help, and where it cannot replace a person.

1. Pull your own history the way the surveyors will. Offsite prep reviews complaints and facility-reported incidents before the team arrives.⁴ Download your facility's rows from the Health Deficiencies file and list every tag cited in your last three cycles, with severity. Repeat tags are your first priority.
2. Know your RBS position. Check your overall and staffing star ratings, PBJ and MDS audit status, and any G-level or higher citation in the last cycle against the QSO-26-14-NH criteria.² Staffing and data-accuracy tools belong here.
3. Work the national top tags as standing audits. Infection control (F880), accidents and supervision (F689), food safety (F812), medication labeling and storage (F761) and care planning (F656) lead the standard-survey citations. Each maps to a facility task or care area surveyors always review.⁴ Observation-based audits (hand hygiene, kitchen temperatures and labeling, med cart checks) need eyes on the floor; software can schedule, record and trend them.
4. Close documentation gaps continuously. F656 and F684 citations turn on whether the record matches the care. This is where AI summarization and chart-scanning tools make their strongest claims; treat their flags as a queue for a nurse to review, not as findings.
5. Rehearse immediate jeopardy scenarios. Accidents and supervision (F689) and abuse (F600) produce 43% of immediate jeopardy citations nationally in our count. Drill fall response, elopement, and abuse reporting timelines (F609) with staff on every shift.
6. Run a mock survey on the LTCSP sequence. Use the procedure guide's steps: initial pool screening without staff, sampled-resident investigations, and the facility tasks.⁴ Mock-survey software can structure this; the value still comes from someone independent of the unit doing the observing.
7. After the survey, feed the 2567 back into QAPI. AI-drafted plans of correction can save time, but the plan is a commitment your team has to carry out and the state agency has to accept. Review every sentence.

7. Method and sources

F-tag and severity analysis. We downloaded NH_HealthCitations_Aug2026.csv from the CMS Provider Data Catalog Health Deficiencies dataset (r5ix-sfxw; CMS processing date 2026-08-01) on 18 September 2026. Every row is one health citation; all 419,479 rows carry the F prefix. We counted rows by "Deficiency Tag Number" and by "Scope Severity Code", counted distinct CCNs per tag for the facility shares, and split standard and complaint citations using the file's "Standard Deficiency" and "Complaint Deficiency" flags. Shares are of all 419,479 citations unless stated. The file covers each facility's three most recent inspection cycles, so survey dates span 2017 to 2026 and differ by facility; it is a snapshot of current public records, not a fixed calendar period, and CMS updates it monthly.

1. Social Security Act §1819(g)(2), Standard survey, Frequency, and Extended surveys: ssa.gov
2. CMS, QSO-26-14-NH, "Nursing Home Risk-Based Survey National Implementation," 16 July 2026, including Appendices A, C and D: cms.gov (PDF); CMS press release: cms.gov
3. HHS Office of Inspector General, "States' Backlogs of Standard Surveys of Nursing Homes Grew Substantially During the COVID-19 Pandemic," OEI-01-20-00431, July 2021: oig.hhs.gov
4. CMS, Long Term Care Survey Process (LTCSP) Procedure Guide, effective 23 June 2026: cms.gov (PDF)
5. CMS, Nursing Homes guidance page (Appendix PP, survey resources, RBS overview): cms.gov
6. CMS, State Operations Manual Chapter 7, Survey and Enforcement Process for Skilled Nursing Facilities and Nursing Facilities (Rev. 244, issued 26 June 2026): definitions, severity and scope matrix, and §7535.1: cms.gov (PDF)
7. CMS Provider Data Catalog, Health Deficiencies (r5ix-sfxw), August 2026 file: data.cms.gov

Also from the same publisher: a history of the National Emerging Leadership Summit, the leadership summit for early-career long-term care administrators, which now continues under ACHCA.