A useful Revit AI prompt names the model, the objects, the approved source, the exact change, and the checks that will establish whether it worked. For a small batch update, a six-field brief is usually easier to review than a long conversational request. Start with a descriptive field such as instance Comments, then compare the resulting values with the approved input.
How to write a Revit AI prompt with a clear task and scope
When your task is precisely framed, an AI teammate can apply edits and produce verification materials inside Revit. You provide the brief and the artifacts; it does the mechanical steps you request.
• You define the task, sources, and exact selection. The IDs and rules describe the intended scope; the operator must verify what the tool actually changed.
• You attach context (screen recording, annotated screenshots, PDFs, spreadsheets). With ArchiLabs Desktop, teams can pass those assets and have the agent update native model elements and generate views/schedules for review. Integration availability depends on application version, plan, accounts, and permissions.
• You get named deliverables for approval. Treat them as requested team procedures, not guaranteed product UI defaults.
For packaging context and selections consistently, see the practical patterns in Revit Model Context Protocol.
Revit AI prompt example: update Comments on 12 doors
This hypothetical brief shows the minimal fields that let an AI agent act safely and produce reviewable output.
1) Task goal
Update the existing instance parameter “Comments” for twelve Level 2 office doors to match Revision E text. Do not change door Types, Type Width/Height, Fire Rating, Mark, From Room, To Room, swing, host, or phase data.
2) Sources to follow (name exact sheets/columns)
• PDF: “A‑801 Door Revision Schedule – Rev E” (use: Door No., Comment Text)
• Excel: “L2_Office_Door_Comments_RevE.xlsx” (Sheet: L2‑Offices; columns: Element ID, Door No., Target Comments). The coordinator has approved this mapping; hold any row that disagrees with the referenced PDF.
• Screenshot: Marked‑up Level 2 plan (PNG) circling the 12 doors
• 30‑second screen recording noting that Door No. is a tag/display value, while Element ID is the unique target identifier
3) Scope selection (the only elements to edit)
• Confirm the project document and saved version first. The following IDs belong only to that document; do not reuse them in a different model or assume they survive deletion and recreation.
• Edit only these Element IDs: 846321, 846339, 846352, 846367, 846380, 846395, 846402, 846416, 846427, 846441, 846458, 846469.
• For each ID, confirm it appears in the Excel mapping before changing anything. If an ID is missing in Excel, skip and report as “skipped.”
4) Exclusions and constraints (what must never change)
• Do not modify: Type, Type Width, Type Height, Fire Rating, Mark, From Room, To Room, Phase Created/Phase Demolished, swing, instance offsets, or host location.
• If the Target Comments cell is blank, leave the current Comments value unchanged and report as “unchanged.”
5) Rules to apply (exact edits and fallback)
• Set the instance parameter “Comments” to the Excel “Target Comments” value for each listed Element ID.
• Trim leading/trailing whitespace; do not append prefixes/suffixes.
• No other parameters are to be created, deleted, or edited.
6) Acceptance criteria and deliverables (how we’ll verify)
• Scope tally: Provide counts of changed, unchanged, and skipped; their sum must equal 12.
• Before-change record: Export a CSV named “QA_Doors_RevE_L2_before.csv” listing the 12 Element IDs with current Comments (Element ID, Door No., Comments).
• After-change record: Reread the same twelve objects into an external comparison register using a verified extraction method. Include document identity, element ID, Door No. and Comments. A native door schedule can supplement this where its available fields support the review; do not assume Element ID is a standard schedulable field.
• Visual check: Create “L2 Office Plan – After” showing tags; cloud or color the 12 targeted doors for visibility only.
• Exceptions: Add a text note “Door_Update_Errors_RevE” listing any skipped IDs with reasons (e.g., not in Excel mapping).
How to clarify ambiguous Revit automation instructions
• Ambiguous: “Update the office doors per the latest schedule.” Actionable: “Edit only instance ‘Comments’ for these 12 Element IDs [list] using ‘A‑801 – Rev E’ and ‘L2_Office_Door_Comments_RevE.xlsx’. Do not change Type, size, Fire Rating, Mark, From/To Room.”
• Ambiguous: “Fix door notes where needed.” Actionable: “For each listed Element ID, set instance ‘Comments’ to the Excel ‘Target Comments’. If the cell is blank, leave the value unchanged and report it as ‘unchanged’.”
• Ambiguous: “Make sure you don’t break tags.” Actionable: “Leave Mark and From/To Room untouched. Supply before/after external records for the same model and twelve IDs so we can compare Comments by object identity.”
Check Revit parameter changes before accepting the AI result
• Use verified IDs, not display marks: Element IDs identify elements within a particular document; Door No./Mark is an editable display value. Preserve the document/version context and verify that each identifier still resolves to the intended object.
• Rely on a before-change value record: The exported “QA_Doors_RevE_L2_before.csv” is your evidence of prior Comments for the 12 IDs. A duplicated plan view is convenient for visibility, but it is not a frozen parameter snapshot.
• Confirm the scope math: Review the changed/unchanged/skipped counts—together they must equal twelve. “Unchanged” includes rows where Target Comments was blank or matched the existing value.
• Compare values, not just counts: review every before/after row against the approved mapping, including skipped and unchanged items. If you use a native schedule as supporting evidence, document its filters and itemization separately; the export does not automatically preserve those settings.
• Check wider effects: compare relevant non-target doors with a captured baseline using the same extraction method. A current-state scan alone cannot establish what changed. Investigate any unexpected difference before accepting the batch.
• Optional warning gate: Ask the agent to list any new post‑edit Revit warnings associated with those 12 IDs for triage. This surfaces issues but is not engineering validation; see ideas in Use AI to Find and Fix Revit Warnings Automatically.
For your first prompt, choose a small set of objects whose current values you can inspect completely. Prepare the source mapping and before record before requesting the edit; those two artifacts make the result much easier to evaluate.
Learn about how Model Context Protocol (MCP) is changing the way we use tools like Autodesk Revit and the future of AI agents and architecture
Keep Revit models healthy on autopilot. Use AI to spot, prioritize, and fix warnings fast—saving time, errors, and rework.