How to Build Your Own AI Assistant

Gen AI can save you time, but once you’re using it frequently, the process of repeatedly uploading the same background files and re-entering prompts for common tasks can really eat into your efficiency gains. That’s why many generative AI platforms allow you to create custom AI assistants: what ChatGPT calls a “custom GPT,” Claude calls a “Project,” and Google Gemini calls a “Gem.” These assistants store elements of a prompt that you might want to use over and over so you don’t have to include them every time you ask the platform to help you with your recurring tasks and challenges.

Creating assistants can help with a wide range of tasks and requires no special technical knowledge or coding skill. For example, I’ve created a marketing AI assistant for myself, stocked with examples of my past newsletters, articles, and social-media posts; I use it several times a week for things like brainstorming content ideas or slicing a newsletter into shareable nuggets for LinkedIn. I’ve also created a strategy sounding board, an AI that answers tech-support questions, a research summarizer grounded in specific subfields of academic and applied research, and a project tracker for my biggest ongoing project. Most of these took me less than 20 minutes to create, and they’ve saved me hours of work.

As a writer and speaker on the digital workplace who has also built my share of websites over the years, I can say that my limited coding skills have been totally irrelevant to the creation of these custom assistants. It’s become clear in my two years of creating many different custom-trained models and assistants, as well as helping many other people to create and refine theirs, that if you can describe a project in plain language or write a job description, you have what it takes to create a valuable custom AI assistant.

Let’s first look at what assistants can help with, and then how to build and use one.

Types of AI Assistants

There’s a wide range of uses for custom AI assistants (as you can see if you search for custom GPTs that people have shared online), but here are four major areas where you are most likely to find them helpful:

• Writing, marketing, and communications. AI assistants can help with writing and content creation, whether that’s drafting social media posts, writing reports, building slide decks or addressing customer complaints. Setting up an assistant—as opposed to working via one-off prompts—allows you to give the bot guidelines and examples to work off, so you don’t have to dig them up every time you need something written or revised. Here’s a great example in action.

• How-tos and troubleshooting. Tell an AI assistant about your tech setup and you’ll have personalized answers at your fingertips anytime anything goes wrong. Same for other fields: You can create financial or tax helpers, data analysts, talking car manuals, equipment repair advisors, and so on. (Though of course gut- and fact-check the advice they give, as gen AI is still prone to hallucination.) This approach can help you provide faster support to customers, too: Responding to a query HBR sent out in an Instagram story about custom AI bots, Roberta, a customer success analyst, described using an AI assistant to automate routine customer inquiries so that she can focus on more complicated questions and provide better responses in less time.

• Productivity and project management. Custom AI assistants sort and prioritize your tasks, extract action items from meeting transcripts, and construct and update project timelines. For example, entrepreneur Sarah Dopp created a Claude Project she uses every morning to guide her through a series of day-planning questions that help her identify and prioritize tasks and reflect on her learning goals for that day. And with the introduction of “agents”—AI systems that can interact with programs and websites on your behalf—chat-based AI assistants will also soon be able to connect directly to note-taking apps, digital calendars, task lists and other productivity programs.

• Strategic advice, coaching, and training. You can create an always-on sounding board with whatever subject-matter expertise you need, configured with your preferred tone of interaction—anything from strict taskmaster to encouraging cheerleader (though, again, keep the potential for hallucinations in mind). Tech VP Ken Romano created a custom GPT that channels the kind of feedback he’s frequently heard from his managers, and used it to fine-tune his product roadmap before sharing with fellow execs.

To get started, you might create one all-purpose AI assistant in each of these four areas. As you work with the bot, though, you may find that you get better results by creating more bots, each with a narrower focus—for example, creating one assistant to help you write outreach emails, and a different AI to help you create pitch decks.

How to Create Your Custom Assistant

Once you have an idea of what you’d like an assistant to do for you, take these basic steps to get it up and running.

Choose a platform.

Your choice of platform may be determined by which AI tools your organization provides (or allows you to use), but if you’re able to choose where to create your assistants, begin by thinking about your needs. For example, ChatGPT is, as of this writing the only platform where you can interact with a custom assistant via voice, as a conversation (and not just using voice dictation). If you’re using the bot for writing, know that Claude seems to have the best grasp of writing style (you can compare writing samples here) but unlike Gemini and ChatGPT, doesn’t have a live connection to the web (so it can only work from its own training data). And if you have Google Docs or Gmail threads you want to summarize or analyze, a Gemini Gem has the edge, because it allows you to integrate these assets easily. AI responses can also be affected by what each platform includes in its training data and what safety or moderation provisions it has in place; platforms and plans also vary in how else they make use of your chats, so look into these differences before choosing a platform.

Start by experimenting.

The easiest way to create a new AI assistant is with a normal chat session in which you give a prompt that includes any background information or persona instructions you think the AI platform will need to answer a question or complete a task you have assigned (like “you are a marketing copywriter”). Even experienced AI users find that it can take some trial and error before you get a useful response; speed that up by providing lots of explicit feedback. (“Bullets 2, 3 and 6 are great, but point 1 is too wordy and points 4 and 5 violate the brand guidelines I showed you.”) Once you start getting decent results, consider the prompts you used and the feedback that steered the AI in the right direction, and synthesize these items into a first draft of your new assistant’s custom instructions. (You can even ask the AI platform to do that for you: “Take the results of this chat and write a set of custom instructions for a marketing copywriter that reflect all the feedback I provided in this session.”)

Write custom instructions.

The heart of any custom assistant is the guidance you draft that tells the bot who it is, what its purpose and tasks are, and how you want it to go about its work. Write these instructions in the second person (“you will do this”), and sketch a persona: “You’re a detail-oriented data analyst working for a really difficult boss who has no tolerance for errors,” or “You’re an editor with a slightly whimsical style, so you specialize in coaxing a little quirkiness from the contributors to your corporate blog.” Describe the kinds of outputs you want it to create (“You clean datasets and return them as CSV tables; you provide bullet-form advice on research methodologies.”) Give your AI assistants goals and notional KPIs: “Your goal is to identify cost savings” or “You want to acquire the maximum number of inbound sales leads from every piece of content.” And specify the tone of the answers you want by giving some guidance on voice.

Depending on the platform, you may have an upper limit on the number of characters you can fit into your instructions, so it pays to be efficient—and the best way of drafting great instructions is often by asking a non-custom AI for help. (“You are an AI consultant and prompt engineer helping an inside sales professional create a custom AI that will draft pitches, follow-up emails and talking points; please draft custom instructions that tell the AI to…”)

Provide knowledge or background files.

Custom assistants become more powerful when you equip them with reference files that they can draw on as examples, guidance, or contextual information. For example, the virtual project manager I created to support a complex tech project is backed by a diagram of our prototype and documents that provide background context on the project vision and roadmap as well as some detailed technical information.

Put Your Assistant to Work

Your AI assistant won’t be perfect right away; you will likely need to experiment with a few different prompts, update its instructions, and give it additional files for context before you get consistently good results. Give it lots of feedback on what it’s getting wrong, how you want it to get better, and what it’s getting right, adding additional files that may be helpful as examples or context. Once you see what improves its performance, revise the assistant’s core instructions—you can even ask it to write its own revisions based on how your session has unfolded.

Investing in the steady improvement of your AI assistant’s accuracy and effectiveness will yield big returns in time savings and response quality. When you work with AI via one-off sessions, it can be tempting to revisit and add to a thread many times to take advantage of the knowledge you’ve given it. But you’ll probably find that if the thread gets long enough, the platform starts “forgetting” things you’ve already told it or gives you a warning. But when you have a custom assistant already loaded up with your files and instructions, you can start a new thread for each work session.

Once you have an effective, custom AI assistant that supports your work, you’ll spend less time wrestling with AI prompts, and more time using AI to accomplish your core tasks.

Copyright 2025 Harvard Business School Publishing Corporation. Distributed by The New York Times Syndicate.

Explore AAPL Membership benefits.