What AI can do in your job and what it must never touch

If you remember one thing from this page, make it this:

AI does drafts, summaries, and plain-language versions for you, the professional, to review. It does not make clinical, legal, or financial decisions. And patient or client confidential information never goes into a consumer chatbot.

Every use on this page lives inside that rule. Anything outside it isn't on this page, because it shouldn't be in your workflow. Your professions already run on this structure: work product gets reviewed, judgment stays with the licensed human, confidentiality is not negotiable. AI doesn't change any of that. It just adds a very fast drafter at the bottom of the stack, with the same review obligations as anything else that crosses your desk.

One more thing before the useful part, because trust here has to be earned in both directions: AI states wrong things with exactly the same confidence as right things. In your jobs, that fact has to guide how you use it. Everything below is built around it.

For paperwork around the care. Keep the care itself in human hands.

• Turn medical jargon into plain language a patient's family could understand: general explanations of a condition or procedure, never anyone's actual chart or records.
• Draft the non-clinical writing: the staff memo, the training outline, the schedule note, the newsletter.
• Summarize a long guideline or policy update so you know what changed. Then check it against the original before you rely on it.
• Make practice questions when you're studying for a certification.

And the hard lines, plainly: it does not diagnose, it does not dose, it does not make clinical calls, and patient information never goes into a chatbot. The rules that protect your patients protect you here too: keep their data out of it.

Why the dosing line is there and not fine print: AI will state a dosage, a contraindication, a clinical claim with complete confidence and no idea whether it's right. In most jobs a confident wrong answer costs an awkward correction. In yours it can cost a person. That boundary keeps AI on staff memos and prevents it from practicing medicine without a license.

A fast first pass on the words. The law stays with you.

• Draft routine correspondence, such as a status update or scheduling letter, that you review word by word before it goes out.
• Summarize a long document you are allowed to share. Practice on a public filing or an invented case, never a client file. Then compare the summary with the original before relying on it.
• Turn a legal concept into plain language for a client letter you review and own.
• Organize notes into a timeline or checklist, using fictional or public material unless your firm has approved the tool for client work.

And the hard lines: it is not legal research you can cite. AI is known to invent cases and citations that look completely real. It does not give legal advice. And client-confidential documents don't go into consumer AI tools. Privilege matters more than speed.

That invented-citations line deserves a beat, because it's the single best-known AI failure in your profession and pretending otherwise would insult you: lawyers have filed briefs citing cases that do not exist, complete with plausible names, reporters, and page numbers, because AI generated them and nobody checked. The failure wasn't using AI. The failure was treating a drafting tool as a research database. It fabricates with perfect formatting. Anything that will be cited, filed, or relied on gets verified in a real legal research source, every time, no exceptions.

Use AI for drafts and explanations. Keep the numbers and judgment with you.

• Draft the client emails: the document-request list, the deadline reminder, the "here's what I need from you."
• Turn a tax or accounting concept into plain language for a client note that you verify before sending, because the rules change and AI's memory of them can be stale.
• Write up your processes: the onboarding checklist, the month-end steps, the how-we-do-it doc.
• Summarize new guidance so you know what to read closely. Check it against the source.

And the hard lines: it does not decide tax positions, it does not produce numbers you can trust unchecked, and client financial data doesn't go into consumer AI tools.

The specific trap in your field is staleness plus confidence: thresholds, limits, and rates change, sometimes mid-year, and AI will sound equally sure when quoting last year's figure or a plausible number that was never real. Look up any figure from AI before using it. The current-year source is the source; AI is, at best, the thing that reminds you which source to open.

That shared shape is the whole reason this page exists. In a careful job, limit AI to drafts, summaries, or plain-language versions that you review and that use nothing confidential. If a use falls outside those limits, leave it out.

On the rest of this site, the honesty section warns that AI is sometimes confidently wrong and tells you to check its work. On this page, that warning was the opening, the middle, and every hard line above, because in your work, "check it" isn't advice. It's the license.

So here's the honest close instead: used inside the boundaries, AI is genuinely worth your time. The memo that took your evening, the plain-language letter you rewrote three times, the guideline you never got to read: those are tasks where drafting help may be useful. Whether it saves time depends on the task and the review it needs. The boundaries aren't the price of using AI safely. They're the instructions.

Pick one non-sensitive task from your profession's list above, such as a staff memo, scheduling letter, or document-request email. Nothing confidential in, everything reviewed on the way out. Do that one task with AI this week, and hold the review bar exactly where you'd hold it for a junior colleague's first draft.

If it saves you real time, add a second task from the list. If a use ever makes you hesitate on confidentiality or judgment grounds, trust the hesitation. In your line of work, it's usually right.