To check AI-assisted BIM work, define the expected result before the edit and verify it independently afterward. For a small equipment-documentation package, reconcile the intended object identities and values, inspect the affected geometry and sheets, and record unresolved issues separately from accepted work. The assistant’s own completion report is a starting point for review.
Define acceptance criteria for AI BIM model changes
Acceptance is a documented decision for a very specific deliverable: one closed set of 24 equipment instances with agreed unique asset IDs, and two sheets that present them. The acceptance relies on two kinds of evidence that catch different errors:
• Machine-readable comparisons for completeness and identity (the ID set in the defined task scope exactly equals your provided list).
• Native-property and geometric inspection against supplied placement criteria, plus a visual review of the sheets.
Do not rely on row counts alone. Schedule rows vary with itemization and filters; accept only when the set of project Asset_ID values exported from the defined scope equals your CSV list, and there are no extras. A schedule export also does not automatically carry your filter context—capture the applied filters separately in the acceptance note and produce the comparison export yourself.
What AI model checking can flag and what reviewers must decide
What AI can do in this workflow
• Create or update native model elements and write instance parameters from your mapping.
• Build the equipment schedule and assemble two sheets.
• Generate a draft verification bundle (counts, ID lists, and flags) to reduce what you must inspect.
What requires human judgment
• Legibility and presentation on the sheets (a machine report does not prove readability).
• Any geometry flags or unexpected placements. Do not accept unexplained exceptions—either fix them or record the approved rationale.
If your team wants an AI that works alongside Revit or AutoCAD for this scope, see ArchiLabs Desktop. Integration availability depends on application version, plan, accounts and permissions.
For background on QA/QC thinking and sheet automation patterns, you can borrow ideas from Avoiding Costly Mistakes: Automating QA/QC in Revit Models and Auto‑Build Schedules and Sheets with AI this Weekend.
How to run a BIM model acceptance check
• Define acceptance, not just scope
• Provide the authoritative list of 24 Asset_ID values (one per instance, no duplicates).
• Name one exclusion set: explicit instance IDs the automation must not modify.
• State the two sheets to deliver and the schedule filters you expect.
• Prepare inputs the agent can consume
• Family/type to use, level/phase/workset rules, and a parameter mapping.
• A CSV (e.g., asset_list.csv) with Asset_ID, Equipment_Tag, Service and any known values.
• An exclusion file (e.g., do_not_touch_ids.txt) listing the instance UniqueIds to skip.
• Execute, then verify in two passes
• Agent performs placement/updates, builds schedule and sheets, and emits a draft report.
• Reviewer independently exports the schedule from the model, compares the Asset_ID set to the CSV, and performs a short visual review of the two sheets. The agent’s own report is not sufficient on its own.
BIM checking example: verify 24 equipment IDs and two sheets
Use this as an illustration and replace names to fit your project.
Scope and deliverables
• Goal: Ensure exactly 24 equipment instances exist with the provided Asset_ID set and are documented on two sheets.
• Family/Type: the approved equipment family/type listed in the project mapping (example placeholders).
• Level/Phase: Place on Level 3 and Phase = New Construction.
• Exclusion set (explicit IDs to skip): {UNIQUEID_006, UNIQUEID_011, UNIQUEID_019}.
• Sheets to produce: EQP‑301 Equipment Plan – Level 3; EQP‑302 Equipment Schedule – Level 3.
• Schedule filters to apply: Level = Level 3; Phase = New Construction (state them in the acceptance note).
Machine-verifiable checks (identity over counts)
• ID set equality: The set of Asset_ID values in the defined task scope equals the 24 IDs in asset_list.csv; no missing and no extras. This is the primary completeness check.
• Uniqueness: No duplicate Asset_ID values in the model or schedule.
• Schedule context: After applying the stated filters and itemization, the schedule rows correspond to those same 24 Asset_IDs. Do not accept row count alone.
• Parameter fill: Required approved data fields exist on all 24; inspect their actual instance/type bindings and native phase properties separately. If some fields are allowed to be blank (e.g., two pending administrative notes), state the allowance explicitly.
• Exclusion enforcement: Compare the relevant properties of the excluded instances with a real before record, and check for wider shared-type effects. A filtered schedule alone cannot prove this.
Visual review (short, targeted)
• Plan sheet EQP‑301: Tags are placed without overlaps and point to the correct instances; the view reads cleanly at the project’s standard scale and template.
• Schedule sheet EQP‑302: Title and columns match the template; sorting/grouping match the stated rules; any intentional blanks are clearly noted (e.g., “TBD”).
• Geometry sanity: No obvious hard overlaps with major elements as seen in plan; any flagged locations are either corrected or documented with an approved note before acceptance.
Evidence bundle for acceptance
• Independent schedule export (CSV/XLSX) produced by the reviewer from the model with the stated filters.
• The agent’s draft report (useful but not authoritative) listing model instance counts, the observed Asset_ID set, and any geometry flags.
• PDFs for EQP‑301 and EQP‑302 at the correct page size.
What to include in a BIM acceptance checklist
Use a single record with three statuses per check: pass, blocked, or excluded, and name the reviewer.
Example (hypothetical)
• Task: Deliver 24 equipment instances (by Asset_ID) on Level 3 with two sheets.
• ID set equality: pass — all 24 IDs present; no extras.
• Uniqueness: pass — no duplicate Asset_ID values.
• Schedule context: pass — reviewer export reflects the same 24 IDs with stated filters.
• Required parameters: pass — all required fields populated; 2 administrative note fields marked TBD as allowed.
• Exclusion review: pass only after comparing {UNIQUEID_006, UNIQUEID_011, UNIQUEID_019} with the saved baseline.
• Plan sheet EQP‑301: pass — tags legible and properly associated on visual review.
• Schedule sheet EQP‑302: pass — fields and order match template; notation present for TBDs.
• Geometry flags: blocked — one overlap at Grid C/7; corrected and rechecked before acceptance.
• Verdict: pass (after correction). Reviewer: [Name], [YYYY‑MM‑DD].
Write an AI request with model checking requirements
• Provide: asset_list.csv (24 Asset_IDs with any known attributes), do_not_touch_ids.txt (instance UniqueIds to skip), family/type and placement rules, schedule field/filters, and annotated screenshots indicating typical locations.
• Ask for: placement/updates, the two sheets, a draft report listing the Asset_ID set discovered in the model, parameter fill rates, and any geometry flags.
• Verify: export the filtered schedule yourself, compare the ID set to asset_list.csv, then perform the targeted visual read.
Set up a first run with a 24‑ID CSV and this acceptance record on a copy of your project; if you want an AI assistant that works with native Revit/AutoCAD elements for this scope, start with ArchiLabs Desktop.
Avoid costly mistakes when doing Revit QA/QC. By doing so, you can reduce project cost and timeline overruns and improve your team's performance.
Upgrade your weekend workflow: use AI to auto-build schedules and sheets, save hours, reduce errors, and launch projects faster.