The short answer
To generate a fixed asset depreciation schedule with AI, give the model the asset details (cost, placed-in-service date, class life, method, and convention) and ask it to produce a year-by-year table using MACRS, straight-line, or your book method. The prompt below does this in one pass, but you must verify the class life, convention, and any Section 179 or bonus depreciation limits against IRS Pub 946 before relying on the numbers.
AI is excellent at laying out the arithmetic and formatting a clean schedule. It is unreliable at picking the correct recovery period, applying the mid-quarter test, or knowing the current-year bonus percentage. Treat it as a fast draftsman, not a tax authority.
A ready-to-use base prompt
Paste this into your model and fill in the brackets:
You are a tax accountant. Build a MACRS depreciation schedule for the following asset. Show a year-by-year table with columns: Year, Beginning Basis, MACRS Rate, Depreciation Expense, Accumulated Depreciation, Ending Basis.
- Asset: [description]
- Cost basis: [$amount]
- Placed in service: [date]
- Property class / recovery period: [e.g., 5-year, 7-year]
- Method: [200% DB, 150% DB, or straight-line]
- Convention: [half-year or mid-quarter]
- Section 179 election: [$amount or none]
- Bonus depreciation: [percentage or none]
Apply Section 179 first, then bonus depreciation on the remaining basis, then MACRS on what is left. State every assumption you make. Do not invent the recovery period or bonus percentage—if I did not provide it, ask.
The last line matters. Without it the model will silently guess a bonus rate, which is where most errors start.
Prompts for specific tasks
Reconcile book vs. tax depreciation
Given this asset (cost, date, class, method), produce two parallel schedules: one for tax using MACRS 200% DB half-year, and one for book using straight-line over [X] years with no bonus. Then show a third column: the annual book-tax timing difference and the cumulative difference. Flag the year the difference reverses.
This is useful for deferred tax workpapers and for spotting Schedule M-1 items.
Test whether the mid-quarter convention applies
Here is a list of assets placed in service this year with cost and quarter placed in service. Calculate the total basis of assets placed in the last quarter as a percentage of total basis for the year. If it exceeds 40%, tell me the mid-quarter convention applies to all assets and re-run the schedule accordingly.
AI handles the aggregation math well; you still confirm which assets are excluded from the test (residential rental and nonresidential real property, listed property, and assets expensed under 179).
Draft a client-facing explanation
Explain in plain English, at a 10th-grade reading level, why this piece of equipment is depreciated over 5 years instead of expensed all at once, and what Section 179 changes. Keep it under 150 words.
What AI gets wrong—and your checklist
Before you drop an AI schedule into a return or a workpaper, verify these:
| Item | Why AI misses it |
|---|---|
| Recovery period | Models often assume 5- or 7-year without checking the asset class in Pub 946 Table B-1/B-2 |
| Bonus percentage | The phase-down rate changes by year; models pull outdated figures |
| Section 179 limit and phase-out | Annual dollar caps and the investment threshold update yearly |
| Mid-quarter test | Requires knowing which assets are excluded |
| Listed property limits | Autos have separate caps AI rarely applies |
| Land | Not depreciable—must be split from building basis |
| Short tax year | Requires proration the base prompt won't do unless told |
A worked example you can adapt
Say a client buys $30,000 of office furniture (7-year property, 200% DB, half-year convention), no 179, no bonus. A correct MACRS schedule looks like this:
| Year | Rate | Depreciation |
|---|---|---|
| 1 | 14.29% | $4,287 |
| 2 | 24.49% | $7,347 |
| 3 | 17.49% | $5,247 |
| 4 | 12.49% | $3,747 |
| 5 | 8.93% | $2,679 |
| 6 | 8.92% | $2,676 |
| 7 | 8.93% | $2,679 |
| 8 | 4.46% | $1,338 |
When you run the prompt, check that the model uses these published table percentages rather than computing declining balance from scratch—both methods should agree, but the tables are the authority. If the AI's Year 1 figure isn't 14.29% of basis, something in the convention or class is off.
Guardrails for using AI on real files
Keep a few habits:
- Never paste identifiable client data into a public model. Strip names and EINs, or use a tool with a business-associate or confidentiality agreement.
- Spot-check one full asset by hand the first time you use a new prompt, then trust the pattern.
- Save your verified prompt as a template so junior staff produce consistent output.
- Ask the model to state assumptions every time, so you can catch a wrong recovery period in one glance.
AI won't replace your judgment on property classification or elections, but it will cut the schedule-building and reconciliation time to minutes.
If keeping up with shifting Section 179 limits and bonus phase-downs is the hard part, DayLift's 5-minute morning briefing flags the tax-law changes that affect your depreciation workpapers before you sit down to them.