You can build a practical, shared library by writing small, reusable task cards that standardize how your team runs common Revit edits. Start with one: a short card that updates the Description on selected equipment from a CSV. The workflow is simple—preflight, human review, apply, post‑check—and AI helps match rows to elements, produce a full diff, and summarize exceptions.
Why Revit teams need a shared AI prompt library
A handful of focused cards turns “tribal know‑how” into steps anyone can run. Each card defines purpose, inputs, model context, steps with checks, and acceptance criteria. This is a team practice, not a built‑in product feature. Use it with the assistant or automation tool your team prefers, and keep the cards in your wiki or repo for versioned reuse.
For packaging the model context so an assistant knows exactly which elements, views, and schedules to use, see the practical guidance in Revit Model Context Protocol.
A reusable Revit prompt template for equipment descriptions
This hypothetical card shows the level of specificity that makes a library reliable without turning into a spec sheet.
• Title: Update Equipment Description from CSV
• Version: v0.2.0
• Card owner: BIM Standards (role)
• Purpose: Update a single instance text parameter named exactly “Description” on targeted Mechanical/Electrical Equipment in the active model.
• Scope: Instances in the active model on editable worksets; no writes to links; skip elements where “Description” is absent or not an instance text parameter.
• Model context: Run on a saved selection named “Equip_Desc_Targets.” If a schedule is used for previewing, it’s for inspection only; writes are made to the selected elements. Note: ElementId values in previews are for verification context and can change after deletion/recreation; use them only with the correct document/version context and verify that they still resolve to the intended objects.
• Input schema (CSV UTF‑8):
• AssetTag (text, required) — primary, verified‑unique match key within the targeted selection.
• Description (text, required) — the new instance value to set.
• Optional: Notes (text) — ignored by the write step; reported in exceptions when present.
• Tiny sample input:
• Header: AssetTag,Description,Notes
• Row: CH‑014,"Chiller, east plantroom",Q2 label update
Run steps with embedded checks
• Preflight
• Validate CSV columns and UTF‑8 encoding.
• Confirm that AssetTag values in the CSV are unique; if duplicates exist, list them and mark those CSV rows as “skip: duplicate key.” No fallback keys.
• Match rows to elements by AssetTag within “Equip_Desc_Targets.” Report CSV row numbers that do not match any element.
• For each matched element, verify that a writable instance text parameter named exactly “Description” exists; otherwise mark “skip: parameter not found or not an instance text parameter.”
• Produce a complete change preview (no truncation) listing, per matched element: AssetTag, ElementId (for reference), Family, Type, CurrentDescription, NewDescription, and a Reason column when skipping.
• Human review (team‑enforced)
• A reviewer inspects the full preview and confirms scope and values before applying.
• Apply
• Write NewDescription only when it differs from CurrentDescription; never overwrite with blanks.
• Skip elements on read‑only worksets or in linked models and record reasons.
• Post‑check
• Re‑read “Description” on all candidates and output a verification list showing FinalDescription.
• Report counts that distinguish: CSV rows, rows matched, cells proposing changes (non‑blank and different), elements updated, no‑ops (value identical), and skips (with reasons).
Acceptance criteria before apply
• The complete preview is generated and reviewed by a human.
• The number of updated elements equals the count of non‑blank Description cells that differ from current values, minus skips noted in the preview.
• No row is applied unless matched uniquely by AssetTag within the targeted selection.
How to run and review a saved Revit AI prompt
• Prepare context
• Save a selection “Equip_Desc_Targets” that includes only the equipment you intend to edit. If you also create a schedule for inspection, remember a schedule export does not automatically include its filters or model settings.
• Provide inputs
• Supply the CSV following the schema above. Confirm the AssetTag column is unique inside the targeted selection.
• Ask your assistant to execute the card
• AI is well‑suited to matching rows to elements, generating a full proposed change set, flagging duplicates/missing keys, and producing exception reports. It can then carry out batched parameter edits and generate a verification list after you confirm the preview.
• Review, then apply
• Your team enforces the review step. Do not proceed until the complete diff is inspected and accepted.
If your team uses Windows and wants a tool that can create and update native Revit elements in addition to this kind of data edit, see ArchiLabs Desktop. Integration availability depends on application version, plan, accounts, and permissions.
Handle failed tasks and retire outdated prompts
Concrete failed‑input example
• CSV row: CH‑099, ,“attempt to clear”
• Failure: “skip: Description blank; clearing not supported by this card.”
• CSV rows with duplicate keys: CH‑014 appears twice with different Description values
• Failure: “skip: duplicate AssetTag in CSV (rows 8, 19).” No fuzzy fallback writes.
• CSV row: P‑200,"Pump 200",— but AssetTag P‑200 not in selection
• Failure: “skip: match not found in Equip_Desc_Targets (row 22).”
Required reporting
• CSV rows total
• Rows matched by AssetTag within selection
• Cells proposing change (non‑blank, unequal)
• Elements updated
• No‑ops (same value)
• Skips, by reason category (duplicate key, no match, read‑only, link, parameter missing/not instance text)
Retire/supersede policy
• When the input schema or checks change, bump the version (e.g., v0.1.0 → v0.2.0) and move the prior prompt to a “Retired” folder with a one‑line pointer: “Superseded by v0.2.0; added duplicate‑key detection.” These are team artifacts, not product features.
For help choosing tasks that work reliably with today’s tools, review What AI can and can’t do in Revit today.
Test prompt revisions against a known Revit input
• Create or pick a known element in your sandbox model with AssetTag TEST‑E‑001 and Description “OLD.”
• Prepare CSV: TEST‑E‑001,NEW
• Run Preflight only and confirm the preview shows a single proposed change with no skips.
• Apply, then Post‑check should report: elements updated = 1; cells proposing change = 1; no‑ops = 0; skips = 0. Re-run the same CSV value “NEW” to verify the card reports a no-op instead of rewriting. Changing the source back to “OLD” would request another update.
Write this Description card into your repo, designate an owner, and run it on a small, controlled selection using the full preview → human review → apply loop; once it’s stable, use the same pattern for your next four cards.
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
Understand what Revit AI can do, where model access and source data limit results, and how to review geometry, parameters, and documentation.