Staff accountants should start with Clear Client Data Out of ChatGPT, a 10-minute lesson, because the first problem with AI at a firm is not a wrong total. It is client details already sitting in a chat or a saved memory, and a partner who has just been asked what the firm keeps in AI tools. Once that is under control, this path moves to the work that fills a staff accountant's month: the bank reconciliation, the client P&L review, overdue receivables and the firm's own rules for every AI reply.
Every lesson uses a made-up firm and practice screens, so nothing touches a real client file. The business case is the same throughout: AI does the matching, sorting and first drafts that eat the hours, and you keep every accounting call, every sign-off and every number that goes to a client.
Which lesson should a staff accountant take first?
Take Clear Client Data Out of ChatGPT first. You play Morgan, a staff accountant at Calloway & Reyes CPAs, a made-up five-person firm. The partner wants the Okafor Family Trust's details out of ChatGPT before Friday. You delete the one chat with the client's name and balance, archive a Q3 close checklist chat the firm still needs, and then find the part most people miss: deleting the chat did not delete the memory ChatGPT saved from it.
You delete that one memory under Settings, Personalization, leave the harmless ones, switch off model training and start using Temporary Chat for client work. The lesson is honest about Temporary Chat: a copy may still be kept for up to 30 days, so it is safer, not secret.
This comes first because every later lesson asks you to upload a client file. Know where it goes before you do.
What comes next, and why in this order?
The order follows a month at a firm: reconcile, review, collect, then make the AI follow your firm's rules every time.
Step 2: How do you reconcile the bank with ChatGPT?
ChatGPT: Month-End Bank Reconciliation (12 minutes) has you upload a bank statement and a ledger export for Harbor Lane Studio, a made-up design studio closing September. The request gives both starting and ending balances and says: if a row could be two things, say Needs review. ChatGPT matches 38 of 40 rows and lists four reconciling items: a missing deposit, a bank fee, a duplicate bill and an outstanding check.
Then you check it. You open the ledger and find the duplicate yourself, add the four items up with a calculator, and confirm both adjusted balances come to $54,355.65. The summary table goes into the month-end file, and three of the four items become entries you make in your accounting software.
Step 3: How do you review a client's P&L with AI?
Analyze a Client's P&L Export (11 minutes, ChatGPT) is the client review. Millbrook Accounting, a made-up firm, exports Copper Pot Kitchen's profit and loss detail for January to June. Before any totals, you ask ChatGPT only to check the file. It finds 4 bank feed duplicates worth $3,475.00 and 9 rows with no account.
You decide the fixes, then get a monthly summary with net profit margin, each cost as a percent of sales, a chart and three takeaways. The one that matters: food and beverage cost went from 29.0% to 34.0% of sales, faster than sales grew. The lesson closes with a double-check to do in the books before you present it.
Step 4: How do you chase overdue invoices without the dread?
Collect Overdue Invoices Faster using ChatGPT (12 minutes) starts from an AR aging report with $29,275.40 past due across ten invoices. ChatGPT builds this week's chase list, oldest and biggest first, and holds back the one marked Disputed. You check that the chase list plus the disputed invoice equals every past-due dollar in the file.
The drafts come in three tones: a final notice to the 108-days-late client, a firm reminder and a friendly nudge. No late fees, no legal threats. You match each draft to the invoice number, amount and dates in the file, and you send it yourself.
Step 5: How do you make AI follow your firm's rules every time?
A Claude Project for Your Firm: Fees, Terms and Tone (9 minutes, Claude) moves the firm's rules out of your head and into one place. You create a project for Millbrook Accounting, write instructions for tone and client policies, and add the fee schedule and engagement letter terms as project knowledge. A chat inside the project answers a client's fee question with the right numbers and offers a call instead of tax advice by email.
When the firm adds a rule about the client portal, you change the instructions once and every new chat picks it up. Keep client returns and IDs out of the project knowledge; firm documents belong there.
Step 6: How do you turn a messy export into a workbook?
Messy Export to Clean Excel Workbook (8 minutes, Claude) has Claude's built-in Excel skill turn a messy sales export into a workbook with a Data tab, a Summary tab built with SUMIFS and SUM formulas, and a chart. Claude says exactly what it changed, including two duplicate rows it removed, so you can check them. Because the summary uses formulas, fixing a row updates the totals.
How fast should a staff accountant go?
The six lessons total about 62 minutes. A practical pace is Steps 1 and 2 together, ideally the week before a close, then one lesson a week outside your busiest close or filing weeks. Run each one on a single real client, with you checking every row, before you make it routine.
Costs to plan for when you do this for real:
- ChatGPT lessons: a free account works to practice. For client files, use a business plan your firm approves and turn model training off.
- Claude lessons: a Claude account. Projects are on every plan; check Anthropic's current plans for limits.
- Anything on a firm card: get the partner's OK and check each vendor's pricing page first.
What should you do after the path?
Pick the one task that ate the most hours last month, most often the bank reconciliation, and run it with AI for one full close next to your usual method. Keep a short list of what it got wrong.
Prompts that fit the role: check a P&L export before analysis, draft your firm's project instructions, sort ChatGPT's saved memories, write a bank reconciliation exceptions memo, summarize an AR aging report and build a month-end close checklist. Terms you will meet: bank reconciliation, closing the books, chart of accounts, project instructions and ChatGPT saved memories.
If you also handle client email and receipts, the AI Learning Path for Bookkeepers covers refund replies, meeting recaps, expense categories and receipt capture. The accounting and bookkeeping firms hub collects more examples, and the best AI for accountants and bookkeepers compares the main assistants.
What should a staff accountant not hand to AI?
Do not hand AI the judgment calls or the sign-off. Let it match, sort and draft; you decide and post.
- Accounting calls. Whether a row is a duplicate, which account it belongs to and whether a variance is a trend are yours. The lessons make ChatGPT ask you before it fixes anything.
- Tax advice. Anything about what a client will owe or can deduct goes to a CPA or enrolled agent, in a call, not in an AI-drafted email.
- The arithmetic you sign. Add the reconciling items and the totals yourself, and tie every number back to your accounting software. AI can add wrong.
- Client identifiers. Remove bank account numbers, card numbers and tax IDs before any upload, and follow your firm's policy and engagement letters on confidentiality.
Key takeaways
- Staff accountants should start with Clear Client Data Out of ChatGPT, a 10-minute lesson that also shows a deleted chat can leave a saved memory behind.
- The path follows the month: bank reconciliation, a client P&L review, overdue receivables, firm rules in a Claude project and a formula-built Excel workbook.
- In every lesson, ChatGPT or Claude checks the file and lists problems before it calculates, and you decide the fixes.
- Reconciling items, cost percentages and chase totals are checked against the source file by you, not taken on trust.
- Tax advice, sign-offs and client identifiers stay with the accountant.
Frequently asked questions
- How is this different from the bookkeeper path?
The bookkeeper path centers on client emails, meeting recaps, expense categories and receipts, with n8n builds. This path centers on the month-end work a staff accountant owns: reconciliation, P&L review, receivables and firm-wide AI rules, using ChatGPT and Claude only.
- Do I need to know how to code?
No. Every lesson uses practice screens where you click and paste what the lesson gives you. The Excel lesson writes the formulas for you and tells you what it changed.
- Can I upload real client files after the lessons?
Only on a business plan your firm approves, with model training off, following your firm's AI policy, and after removing account numbers and tax IDs.
- Is this accounting or tax advice?
No. It is practical training on AI tools. Your firm's procedures, your engagement letters and a licensed professional come first.