Claude AI at Work: Two Habits That Save Real Time

Most people I meet have already tried Claude AI at work. They opened it, asked it something, got a perfectly good answer, and then went back to doing the job the way they always did it. Nothing was wrong with the answer. It just did not change anything.

I see that pattern in almost every class I teach. The tool does fine on a one-off question and falls apart as a habit, because every new chat starts from zero. You re-explain who you are, what the company does, what the report is for, who reads it. By the time you have typed all that, you could have written the thing yourself.

Two changes fix most of it. Neither one is clever. Both take about ten minutes to set up, and then they keep paying you back every week.

Claude AI at Work Starts With a Project, Not a Chat

In the left-hand sidebar, click Projects, then New project. Name it after a recurring job rather than after a tool — “Monthly board report”, “Client proposals”, “Quarterly headcount summary”. Then fill in two things:

• Project knowledge. Upload the reference material you would otherwise paste in every single time: last quarter’s report, your style guide, the blank template, two examples you were happy with.
• Custom instructions. The standing rules, written once. Something like: I am the operations manager at a 40-person engineering firm. Write in plain English, short sentences, no marketing language. Our fiscal year starts in July. If you are not sure of a number, leave it blank and flag it instead of estimating.

From then on, every chat you start inside that project opens with all of that already loaded. You type “draft this month’s report from the attached numbers” and skip the preamble entirely. In class, this is the step that turns the tool from a novelty into something people actually use on a Tuesday afternoon.

One gotcha: keep project knowledge tight. Three to five focused documents beat a whole folder. Load too much and Claude starts searching the pile instead of reading all of it, so the answers get vaguer rather than better. If you cannot say in one sentence why a file is in there, take it out.

Tell It What Not to Do

The biggest jump in output quality does not come from a longer prompt. It comes from one constraint. Paste in your rough draft and type something like this:

Below is my draft email to a client who missed a deadline. Give me three versions: one neutral, one firmer, and one that proposes a new date. Keep each under 120 words. Do not add any facts that are not in my draft — no dates, numbers, or names I did not give you.

You get three usable drafts in about fifteen seconds and you take the best lines from each. But the sentence that matters is the last one. Leave it off and you will get back a polished, confident email containing a delivery date nobody ever agreed to — and it reads so naturally that you will send it without noticing.

The same move works on data: Summarize what this spreadsheet shows. If a column heading is ambiguous, ask me what it means instead of guessing. That one instruction to ask back has saved me from more bad summaries than any clever prompt ever has.

One gotcha: a constraint holds best in the message you put it in. Twenty messages into a long conversation, the output starts drifting back toward its old habits. When you notice that, restate the rule, or start a fresh chat inside the project — your custom instructions carry over, the drift does not.

Plenty more on the calendar

If you would rather watch this done live, Learn to Use Claude AI Today – Transform Your Role & Efficiency runs on Thursday, October 8. We set up projects and custom instructions from scratch, work on prompts that hold up under pressure, and go through the everyday jobs — email, summaries, cleaning up long documents, first drafts — where Claude earns its keep in the first week. You can ask questions as we go.

The full current schedule is on the front page of PCWebinars.com, and new dates go up as they are confirmed. Register for this one, or browse the others and pick whatever fits your week.

About your trainer

I am Tom Fragale. I have been a computer trainer and consultant for more than 30 years, and I have taught well over 30,000 people how to get more out of the software they already own — Excel, Access, Power BI, Word, Outlook, and now the AI tools that sit on top of all of it: Claude, ChatGPT, and Microsoft Copilot. My background is applied, not theoretical. Alongside the teaching, I build real databases and reporting systems for clients in Excel, Access, and Power BI, so the examples I use in class are the problems I ran into last week, not textbook exercises. I also publish free tutorials on YouTube and longer courses online.

Come see the full schedule and the rest of the free material any time at PCWebinars.com.