Using AI at Work

• Drafting responses to common questions
• Improving clarity in instructions or guidance
• Creating consistent messaging
• Translating complex information into plain language

• Generating conceptual ideas for visuals, layouts, or multimedia assets
• Suggesting design variations based on constraints such as brand, format, or audience
• Helping translate abstract ideas or themes into concrete visual or experiential concepts
• Iterating on creative approaches quickly to explore alternatives

• Summarize meeting notes
• Identify action items
• Create meeting agendas
• Turn notes into follow‑up emails

• Create checklists or timelines
• Outline project plans
• Brainstorm options or approaches
• Organize unstructured information

• Summarize sponsor policies, compliance requirements, or regulatory guidance
• Extract key rules, constraints, or eligibility criteria from complex documents
• Assist with checking proposals against lists of required components
• Drafting initial versions of documentation or guidance on research protocols

• Assist with debugging by identifying likely causes of errors and suggesting fixes
• Translate logic or requirements into working code examples
• Refactor code to improve readability, structure, or performance
• Assist with accessibility checks, such as identifying common WCAG issues in front‑end code

• Draft emails, memos, and announcements
• Rewrite content to be more concise or clearer
• Adjust tone (professional, friendly, plain language)
• Create first drafts for web pages or FAQs

Northwestern Role: Research data analyst

Use Case: Every morning, I needed to manually run four Tableau Prep flows in a specific order before anyone else started their work. If I forgot, was sick, or was on vacation, the team’s data became outdated. Tableau Prep Builder doesn’t include built‑in scheduling unless you purchase an expensive enterprise add‑on, so the process depended entirely on one person running it daily. I described this problem to an AI assistant and asked for a way to automate it using tools already available on my machine. The AI helped me create a simple scheduled script that runs all four flows automatically each day. It also generated setup instructions and documentation to make the solution easy to maintain.

Example Prompt: I have four Tableau Prep flows that need to run every morning in a specific order. There’s no built‑in scheduling available. What is the simplest way to automate this using tools I already have, and how should I document the setup?

Consider: AI tools are especially helpful for stepping back and simplifying over‑engineered solutions. Ask the AI to explain why it is recommending a particular approach and whether there is a simpler option. In this case, AI identified that the problem didn’t require complex tooling and could be solved using lightweight, existing tools rather than adding new systems or licenses.

Northwestern Role: Business administrator managing faculty discretionary accounts

Use Case: I needed to send nearly 20 faculty members a personalized update on their discretionary account balances. The process—screenshotting a financial report, editing the image, composing individual emails, and attaching the correct transaction report for each person—took two days to complete.

Using Microsoft Copilot, I built two components. First, an XLOOKUP-based formula that automatically extracts each faculty member's key account figures from a summary report into a clean spreadsheet tab. Second, a VBA macro that loops through the faculty data and generates a personalized Outlook email draft for each person, with the correct figures calculated and the correct transaction report attached. The macro saves drafts for review before sending and flags any missing attachments.

What previously took two days now takes about an hour.

Example Prompts:
Formula: I have a financial summary report saved as a spreadsheet tab. I need a formula that looks up a faculty member's project number and returns four specific fields—carryforward, revenue, expenses, and cumulative surplus—into a separate summary tab.

Macro: I need a VBA macro that loops through faculty data, generates a personalized Outlook email draft for each person using their name, email, and financial figures from specific columns, and attaches a PDF named for each person from a designated folder. Save each as a draft and log an error if an attachment is missing.

Consider: Specificity is key. Before prompting, build a mock-up of the output you want—a sample email, a sketch of the formula result—so you can describe the goal clearly and evaluate whether the output matches. When something isn't working, describe precisely what's wrong rather than just re-running the prompt. Northwestern community members can access a more detailed write-up of this use case in a Weinberg College Commons GenAI Case Study.

Northwestern Role: Administrative staff member

Use Case: I was getting ready for a conference with more than thirty overlapping sessions across three days, and I needed help deciding which ones to attend. I pasted the full session listing into Claude, which already knew my role and current projects from earlier conversations. Note: any similar AI tool, like Copilot or ChatGPT, would work too.) Using that context, Claude recommended a top pick and a runner-up for each time block, explaining why each one mattered for my work. I also asked for the results in formats I could use on my phone, and ended up with a reference PDF, a calendar file with reminders, and a plain-text backup in my Notes app.

Example Prompt: I'm heading into a conference with more than thirty sessions across three days. A lot of the time blocks have three or four sessions running at once. I've pasted the full session listing below — titles, abstracts, presenters, rooms, and times. Right now, my priorities are a major system upgrade, a couple of new data projects, and evaluating AI tools for our team. For each time block, can you recommend a top pick and a runner-up, and explain why each one is relevant to those priorities?

Consider: Getting personalized results depends on the AI tool already having context about your role and priorities. Without it, you'll likely get a generic schedule — accurate but not weighted toward what matters most to you. If you're starting from scratch, spell out your current priorities directly in the prompt.

Northwestern Role: Staff manager who regularly produces email communications for a department.

Use Case: Each quarter, I need to produce an HTML email newsletter for our department — but the available tools produced unreliable code, and writing HTML by hand required specialized knowledge our small team didn’t have. I used Claude to generate table-based, inline-styled HTML email code from a structured prompt that named our color palette, font choices, layout, and technical constraints. The result was a reusable, professional-quality newsletter template built without writing a line of HTML myself.

Example Prompt: Build an HTML email newsletter template with a Northwestern purple (#4E2A84) header background and white Georgia serif headline, a thin teal accent bar (#007FA4) beneath the header, and three to four content sections with images on the left and text on the right, each with a labeled section badge.

In the content sections, include image placeholders with the required pixel dimensions, a table-based layout with inline styles, and a maximum width of 600px. Use Georgia for headlines and Verdana for body text.

Also, create a text-only version as a separate file for the plain-text email alternative.

Consider: The technical constraints — table-based layout and inline styles — are what you need to specify for the output to be email-compatible; the AI handles the implementation once you name them. After generating the HTML, open a second chat and paste the code with a prompt asking the tool to review it for Outlook and Gmail compatibility.

Building the template and inserting content work best as two separate steps: get the structure right first, then add real content in a follow-up prompt. Northwestern community members can access a more detailed write-up of this use case in a Weinberg College Commons case study.

Northwestern Role: Research core administrator

Use Case: I had a spreadsheet with location data that was entered in a form. I forgot to specify the format for the location entries on the form, so it had a combination of just cities, cities and states, and just states. The entries were in different formats as well. I have a Microsoft Copilot add-on license, so I was able to have Copilot in Excel clean up the data, creating two new columns with clean city and state values.

Example Prompt: I want to clean up the location column. I need to convert it into two columns for city and state and make the formatting consistent.

Consider: Copilot, and other AI tools, make assumptions when processing data. After any data cleaning steps, ask it to be explicit about how it made any decisions and what assumptions it made. For example, in my file, it saw that most of the cities were in Illinois, so it assumed locations were in Illinois when the state was missing and the city name could be in Illinois. To follow-up on this, I asked Copilot to add a new column marking rows where it made an assumption about the state.

Northwestern Role: Research faculty

Use Case: We built a new software package as a part of a research project and wanted to create a logo for the package.

Example Prompt: Create a logo for a Python package that helps people process family tree data. The name of the package is [package name]. The logo should be inside a triangle. Use purple and green as the primary colors.

Consider: Some of the options the AI tool generated looked very similar to other logos. Make sure to check any images to make sure they aren’t copying others’ work. We didn’t have anyone on our team who felt they had the graphic design experience to create the logo, but if you do, letting a person create it may be a better choice.

Northwestern Role: Department staffer who responds to questions, provides guidance, or supports services

Use Case: When I need to respond to common questions or explain a process, AI can help draft a clear, consistent response. I’ll paste in an existing reply, instructions, or technical description and ask the tool to simplify the language or adjust the tone for a general audience. This helps me respond more efficiently while improving clarity and reducing confusion for the person receiving the information.

Example Prompt: Here’s a response I often send to people who ask about [topic]. Rewrite it so it’s clear, friendly, and easy to understand for someone who isn’t familiar with the technical details. Keep it accurate, but use plain language.

Consider: AI is helpful for clarity and consistency, but I still review everything carefully before sending. I make sure the information is correct, up to date, and appropriate for the situation. I also personalize my responses when needed so they don’t feel automated or impersonal.

Northwestern Role: Department human resources representative

Use Case: Our team uses behavioral interview questions for staff job openings. We ask each member of the interview committee to assess different skills and experience related to the job. We used to find ideas from online lists of interview questions, but AI helps us find relevant questions faster, and we can also use AI to modify the questions to fit our team or the job better.

Example Prompt: I am interviewing a candidate for a [job title] position in the [department or group name] at Northwestern. I’ve uploaded the job description. Please generate a list of 15 behavioral interview questions I can ask to assess their user support skills and experience working across teams.

Consider: AI tools can also generate lists of things to look for in candidate answers, but these are often overly general. We choose interviewers because we value their experience and judgment, and we don’t want to replace that with AI output. We don’t use AI to evaluate candidates.

Northwestern Role: Department staff with communications responsibilities

Use Case: When I’m having trouble getting started writing something, I use an AI chat tool to help me get something on the page. If I have relevant information in a different format already, I upload those documents and tell the tool what I want to write. If I don’t have anything to work from, I just start writing without worrying about grammar or getting the right tone, and then ask the AI tool to help me structure and format it into what I need. I have it ask me questions if it needs more information, which helps me fill in the details. Once I have something to work from, I can rewrite and edit.

Example Prompt: I need to write a three- to four-paragraph article on our team’s recent project on [fill in details] for our department newsletter. The audience is faculty and staff in the [department name] department at Northwestern. I’ve uploaded a document with the project summary. I want to give a high-level overview of the impact of the project. Make the tone professional but engaging. If you need additional details to write a draft, ask me.

Consider: AI writing can all sound similar after a while. You don’t want to lose your voice. I try to avoid copying and pasting large chunks of AI-generated text into my document. Seeing the output in the chat tool is enough to help me get started. I can pick and choose what I want to incorporate and put things into my own words as I write.

Northwestern Role: Administrator or research support staff

Use Case: I had a data file with hundreds of records compiled from multiple sources, and I suspected there were duplicate or redundant entries—cases where the same item appeared twice under slightly different names, or where a sub-program was listed separately from its parent. Reviewing everything by hand would have been tedious and easy to get wrong. I asked an AI assistant to go through the file and flag potential duplicates, giving it a concrete example of what I meant. It returned a prioritized list organized by confidence level. I then went through the list, told it what to merge, remove, or keep, and it applied the changes directly to the file.

Example Prompt: Can you review this file and identify any duplicate or redundant entries? For example, [describe a specific case from your data]. Give me a plain-text list of potential duplicates so I can decide what to keep or remove.

Consider: The AI can surface candidates efficiently, but the final calls belong to you—it doesn't know your subject matter well enough to decide what counts as truly redundant. The most effective workflow is to review the AI's list and give explicit feedback on each item. Explaining why something should be merged rather than just saying it should be helps the AI apply consistent logic to similar cases across the file. Always spot-check the result to confirm your changes took effect.

Northwestern Role: Manager of an IT team

Use Case: We need the tone and writing of our team’s user documentation to remain consistent even though over a dozen people are contributing to it. We have a team style guide that we used to generate a series of AI prompts team members can use to review their drafts. These cover proofreading, checks for clarity and consistency, audience alignment, and whether the tone matches a handful of characteristics.

Example Prompt: Please act as a professional senior-level copy editor and review the text in the uploaded document for spelling, grammar, and punctuation errors. Provide detailed corrections and explanations for each issue found. Specific areas to check: [fill in details].

Consider: Across different AI tools, we’ve seen them apply our guidelines too strictly. For example, when asking it to find areas where writing can be more concise, they sometimes will suggest we use a slightly shorter word just because it’s shorter. We remind the team that they need to use their judgment in evaluating the AI suggestions.

Northwestern Role: Department staffer who coordinates projects, events, or operational tasks

Use Case: When I’m juggling multiple tasks or starting a new project, I use an AI chat tool to help me organize my work. I’ll paste in notes from emails, meeting summaries, or a rough list of ideas, and ask the tool to turn that information into a checklist, timeline, or simple project plan. If I’m not sure where to start, I describe my goal and constraints and let the AI suggest an initial structure. I can then refine priorities, adjust timelines, and add context based on what I know about the work and the teams involved.

Example Prompt: I’m coordinating a small internal project that involves [brief description]. I have notes from emails and meetings, which I’ve pasted below. Help me organize this into a clear checklist with key steps and suggested timing. If you need more information to make this useful, ask me questions.

Consider: AI can help make things feel more manageable, but it doesn’t know what’s most important in my environment. I review and adjust the output to reflect real deadlines, stakeholder expectations, and dependencies. The tool gives me a starting point, not a final plan.

Northwestern Role: Web developer

Use Case: We help different groups in our school create custom websites and applications. The people we are creating a site for often want to see alternative layouts. We used to draw them, or in more formal meetings, we might mockup options using Adobe Illustrator. Now, we can build a basic layout in the framework we’re using for the project and then use a coding agent to make changes during a discussion.

Example Prompt: We ask the coding agent to get familiar with our codebase, documentation, and project plan first. We only give it access to a copy of our code so that we always have the original as a backup. When we ask it to make changes, we include both what we want to change and what should stay the same. If the outcome doesn’t match what we wanted, we have it revert to the previous code and try again.

Consider: This works well for prototyping, but once we have agreement on a direction, we rework the code ourselves. The code AI agents generate is usually well-documented and looks polished, but we’ve found that it can miss edge cases, make strange assumptions, and be organized in a way that doesn’t work with the rest of our application.

Northwestern Role: Department administrator

Use Case: We run an orientation event for students in our department and ask for their feedback afterward. The survey includes several free response questions. In the end, we need to generate a summary of the responses.

Example Prompt: The responses in the uploaded document are answers students at an orientation event gave to the question [survey question]. Can you tell me what the themes are across answers and pull out direct quotes from the responses as examples of each theme?

Consider: The first time I asked Copilot to summarize the responses was in a conversation where I had given it other information about the program and what I thought the survey responses might say. The initial summary of the responses matched closely to what I expected. When I started a new chat a few days later and asked it to summarize the responses again, the output was different. Your chat history can influence what the output is.

Northwestern Role: Administrator

Use Case: I needed to map a specific type of service or program offered across dozens of peer institutions—information useful for benchmarking what Northwestern offers and identifying gaps or opportunities. Done manually, this would have taken weeks of searching across multiple websites and units at each institution.

Instead, I described the task to an AI assistant: what I was looking for, the specific details I needed for each program, and the list of institutions I wanted covered. The AI systematically searched each institution’s website in batches, looking across multiple schools and departments rather than stopping at the first result. It saved findings to a structured file as it worked, then compiled everything into a searchable spreadsheet and a formatted report.

Example Prompt: I need a report on [type of service or program] at the top 50 U.S. national universities. For each university, search across all relevant schools and units. For every program you find, record: the name, the university, the unit housing it, what it offers, who is eligible, the cost, one web link, and any other relevant details. Compile everything into an Excel spreadsheet and a formatted HTML report.

Consider: AI can miss programs that are buried deep in a website or described in unusual terms, so spot-check a sample of entries before relying on the data for major decisions. It also helps to have the AI work in batches and report back after each one—this makes it easier to catch issues early without losing progress.

Northwestern Role: Administrator or research support staff

Use Case: I had a structured data file with hundreds of records—names, descriptions, categories, and URLs—that I needed to turn into a usable resource for colleagues. Building something like that manually would have taken days and required web development skills I don't have. I described what I wanted to an AI assistant and it built an interactive HTML file directly from my data, complete with summary charts, a searchable listing organized by category, and a filtered view showing only the record types most relevant to my audience. When my needs changed, I asked it to swap out a tab or adjust how records were categorized, and it updated the file accordingly.

Example Prompt: This file contains records with the following fields: [list your fields]. Please create an HTML file with a summary tab showing charts of the data broken down by [category] and [cost], and a second tab listing all records organized by [grouping field] with a live search bar.

Consider: Be specific about your audience and what they need to find quickly—that shapes which views and filters are worth building. The AI will make judgment calls about how to categorize ambiguous records, and those won't always match your intent. Plan to review the output, give feedback on specific cases, and ask for adjustments. The more concrete your examples ("this type of record should appear in tab 2 but not tab 3"), the faster you'll converge on something accurate.

Northwestern Role: Instructor or workshop facilitator who teaches data analysis or coding skills

Use Case: I had an existing Jupyter notebook teaching data visualization in Python. The second half used a specific dataset for worked examples and student exercises, and I wanted to swap in a new dataset more relevant to my current course while keeping the structure, exercise numbering, and teaching progression intact. I also wanted each example plot to briefly explain what the visualization reveals, not just how to produce it.

I attached the notebook and shared a link to the new dataset, then asked the AI to leave the first half unchanged, replace all data-related examples, preserve the exercise structure, research the dataset well enough to write accurate plot interpretations, and suggest a new open-ended exercise using something interesting in the new data. It produced a revised notebook where every plot type used the new data with appropriate preparation code, a clear explanation of what the plot shows, and exercises in the same format as the originals.

Example Prompt: I need to replace the dataset in a Python notebook, which I will attach. The new dataset is here: [URL]. It is a subset of the data found here: [source URL]. Please only use the columns and rows found in the subset, but you can get information from the internet about what the columns are and what they mean. The notebook contains a lesson in coding in matplotlib. The first half does not use a dataset, so it should remain exactly as written. When the dataset is introduced, it uses simple examples, and you should do the same. But the examples should also state what the plot says, which is why you will need to do some research. At the end of the notebook, there is a chance for participants to ask their own question. Please think of and use something in the new dataset that participants can similarly play with.

Consider: For datasets with non-obvious column names or specialized subject matter, provide a data dictionary or link to source documentation so the AI can write accurate interpretations rather than generic ones. Review every plot description the AI produces — it may capture the pattern but miss a nuance that matters for your students. Finally, run the revised notebook from top to bottom before sharing it; across a long file, the AI can introduce subtle naming inconsistencies that a full test run will catch.