Lenny is a software engineer turned newsletter writer whose Lenny’s Newsletter and related podcast and job board have made him beloved online for his advice and perspective.
Guest Author: Lenny Rachitsky
It’s become clear to me that everyone should be playing with GPTs (custom versions of ChatGPT) right now. It’s easy to think AI tools are a fad, too complicated, or unlikely to have real impact on your job. But you’re wrong. Consider these recent news items that caught my attention:
- PwC forecasts that AI could contribute $15.7 trillion to the global economy by 2030 from increased productivity.
- When Italy banned ChatGPT, the productivity of coders in the country fell by 50% before recovering. On the flip side, Duolingo reported a 25% increase in developer velocity when using GitHub Copilot, and Shopify has already written over a million lines of code with Copilot.
- Nat Friedman (former CEO of GitHub) and Daniel Gross (former Y Combinator partner) are investing $100 million in a startup called Magic that’s looking to replace engineers. Prominent engineers are worried.
- In the public markets, there’s a growing separation between companies betting heavily on AI and those that aren’t.
- The power of AI models is only accelerating. Even MrBeast is worried. Here’s AI video generation 10 months ago vs. today. Here’s a video generated by Sora (OpenAI’s latest model) from the prompt “Two golden retrievers podcasting on top of a mountain.” 😮
As Chris Dixon put it, “The next big thing will start out looking like a toy.”
I started to get super-curious in particular about leveraging GPTs at work after my recent podcast conversation with Logan Kilpatrick (head of developer relations at OpenAI). He shared a few examples of how people at top-tier tech companies are using custom GPTs to help their non-eng teams be more productive, while also taking a load off their eng teams. I wanted to find more examples—to see how and why people are doing this—so I turned to Twitter and LinkedIn. I asked how people are using GPTs in their day-to-day work now.
The volume (and variety) of responses blew my mind. I got over 300 replies with hundreds of examples—from cooking up copy experiments to extracting feature ideas from sales calls to having conversations with their customer personas. The creativity, and the impact people are seeing, is 🤯.
Below, to help you explore this yourself, I’ll explain what GPTs are and help you build your own custom GPT. Then I’ll share 20 examples of how people are using GPTs today to make their workplaces more productive. When you read this post, I encourage you to pick an idea and try to create your own GPT. See how easy it is, and how good they actually are. You may be blown away too.
A big thank-you to Dan Shipper and Dennis Yang for providing feedback on this post, and to everyone X and LinkedIn who shared their experience.
What are GPTs?
GPTs are custom versions of ChatGPT that can be tailored for very specific tasks. They live at a unique URL, run in the cloud, and anyone can make one in a few minutes.
To set one up, you tell the GPT what you want it to do for you (in plain English), upload docs that “train” it on your specific context (e.g. your roadmap), and then adjust the capabilities it can leverage to complete your task (e.g. browse the web, use Zapier to take actions, etc.).
Normally, straight-up ChatGPT is more than enough for most use cases, but if you want to upload proprietary datasets or customize how the chatbot interacts, and then share this tailor-made GPT with colleagues, custom GPTs are the way to go.
How to create your own GPT
It’s so easy. Go here, sign up, click “Create” in the top right corner, and tell the AI what you want it to do. Here’s one I created in a few minutes that helps you think more strategically based on advice from the writings of Hamilton Helmer, Michael Porter, Annie Duke, Richard Rumelt, and Geoffrey Moore.
Once you get the initial version set up, you can then upload “knowledge”—decks, PDFs, spreadsheets, and any other documents that give the GPT more information on your specific context and company. For example, you can upload your roadmap, your company values, your career ladders, etc. The more you give it, the better it does. Think of it like onboarding a person—what context would you want them to have to do their job well?
To delve deeper into GPTs:
- Read this very short guide from OpenAI on how to create a GPT
- Browse examples of the most popular public GPTs
- Get a ChatGPT Plus account. You’ll get faster response times and avoid hitting usage limits. ← Get your company to pay for this; it’s very high ROI for $20/month
- Watch Logan Kilpatrick build a custom GPT step-by-step live
- Learn how LLMs work (also this)
- Add Zapier powers to your GPT in order to let it take actions for you
20 ways to use custom GPTs at work
If you discover an idea here that could benefit your team, go ahead and try to build it yourself as a learning exercise. And if you’ve found another use case for GPTs at your workplace, I’d love to know. Just leave a comment below.
1. Refine new product copy
“We built a UX GPT Writer trained on all of our writing, brand guidelines, and product nomenclature. Designers rough out what they need it to say, send the GPT a screenshot from Figma, and it will rewrite it to be concise, using our guideline and brand voice standard. This means anyone on our team can write well, whether they’re a product manager, designer, or engineer.”
—Diego Zaks, VP of design at Ramp
2. Have conversations with your personas
“Everyone has ‘personas.’ I literally took our persona documents, made it a PDF, uploaded it to the GPT (as knowledge), and made a GPT to talk to our personas. It helps me refine roadmap ideas and really empathize better with my customers.”
—Dennis Yang, PM at Chime
3. Learn from past user research findings
“We built a GPT that indexes all of our user research decks, and then people can ask questions [about past research findings]. It’ll answer the questions and link to the decks.” —Anonymous
4. Create a source of truth for product requirements
“We created a custom GPT that has our DB schema, company overview, and some of our product requirements in markdown format. We are in the middle of an intense push over the next six weeks to deliver a fairly substantial feature set. This custom GPT is fully armed to act as both a source of truth for requirements and brainstorming technical approaches.
If you are a PM and want to take [your work] to the next level, look at delivering not just the usual PRD, but also an internally shareable link to a custom GPT that has those same requirements added to its context. (Don’t forget to disable the training toggle in setup.) Your devs will love you.”
—Matt Burch, VP of engineering at Making Space
5. Figure out internal ownership and technical dependencies
“We have a GPT that searches the internal employee databases and org structures and component libraries, and answers queries like ‘Who is the eng lead for X” and “What is the Jira component for Y?’ etc.” —Anonymous
6. Dream up copy experiment ideas
“I use this for experiment analysis and copy:
- Train a custom GPT on brand guidelines and examples.
- [I feed] Google Sheets → Zapier → GPT connection. When I enter in my control copy, GPT will create two variants, explaining the hypothesis for each.
- GPT → Zapier → Google Sheets connection. When I enter in experiment results, GPT will hypothesize why X variant won.
- Upload a CSV of all experiment results to custom GPT again, so GPT can learn from winning experiments.
Challenges: When analysing experiment results, it lacks context on the product and other external factors, but I can’t feed that to the GPT out of data privacy issues.”
—Quintin Au, growth marketing manager at Spotify
7. Create a personal exec coach for yourself
“I’ve made a CEO coach GPT that I’ve trained on podcasts and books from my fav business leaders so I can ask it questions and advice. Works well!”
—Alex Zaccaria, CEO of Linktree
8. Help your team set annual goals
“I built a custom GPT to assist my team with setting their annual goals. It used to be a real struggle to get people to do it, and when they did, it often took hours and multiple meetings to refine.
I took the company values, team principles and values, company scorecards, and our progression model and used that as a knowledge base. I then had my team share where they wanted to be in five years, what they are currently interested in, and what their current role in the company is.
The GPT then returns custom, detailed goals using the SMART framework that takes all the values above into consideration. It gives people their annual goal and then it shares quarterly goals based on the annual ones. They now have goals that are unique to them and their career aspirations while still aligning with what the expectations are at a company level. I’ve been told countless times how much they love the workflow and how much easier it is to set goals. I’ve got people adding goals on time for the first time in seven years without having to track them down.”
—Andrew Kelly, Designer
9. Grade the relevance of search results
“At Faire we use GPT to grade the relevance of our search results. For reporting, we used to have a manual operation grading the relevance of search. Now automated through GPT. For model offline evaluations, we used to manually look at examples, now automated through GPT.
[In terms of impact on reporting], it’s much cheaper to label with GPT vs. humans. [On model evaluation], an increased experiment win rate. We have a better idea of what our models are actually doing to relevance before launch and a better sense of what works vs. not.”
—Minh Pham, PM at Faire
10. Create data pipeline documentation
“[I] created a tool for my data engineers which almost completely off-loaded the work they absolutely hated to do (and thus saved for last/never): documentation. Before, documentation of our data pipeline was horribly incomplete and would quickly grow stale. Even though it isn’t perfect, the tool saved so much time that adoption by the team was assured.”
—Jason B. Hart, director of global analytics and data operations at Springboard
11. Generate takeaways from raw survey results
“I’ve got a custom GPT to analyze survey responses from the prior week/month/quarter to create a report for the team each week.”
—Adrianne Stone, PM at Big Cartel
12. Generate takeaways from sales conversations
“For me, GPT revolutionizes sales. Our most impactful application of GPT lies in summarizing customer interviews in sales. We used to spend a ton of time conducting interviews and synthesizing findings to improve our sales process, find innovative feature ideas for R&D, and enhance marketing collateral.
It’s all history now.
With GPT, we:
↳ Identify recurring pain points across all interviews.
↳ Find compelling customer testimonials for marketing materials.
↳ Pinpoint nuanced critique points that we may have missed.
↳ Generate action items for every individual customer interaction to hand over to CSM/AE.
The benefits to our customer experience (CX) efforts are gigantic.”
—Adam Egger, co-founder of SKLLD
13. Score customer leads
“We created a chatbot to nurture/score potential customers. This way we would only process the ones that would be more likely to become customers.”
—Javier Ortiz, head of marketing/growth at Taxfix
14. Avoid vendor customer support
“When a team is frequently working with a vendor with complicated support (e.g. they don’t have a chatbot, long wait times, etc.), we make a GPT for that vendor. We upload all of the vendor documentation and API docs and point it to the website and support docs, etc. Now when the team has a support question, they first ask the GPT the question. We find they often get faster, better answers than from the vendor itself.”
—Dennis Yang, PM at Chime
15. Enrich new-user signups
“GPT + web scraping enriches every product signup with important metrics.”
—Yash Tekriwal, community and growth at Clay
16. Write optimized product details copy
“We have a GPT writing SEO-optimized e-commerce product info copy in five local languages simultaneously. So far, we’ve seen a 95% price drop and ~5-10x productivity gain compared to the previous manual process. Huge for our translation costs.”
—Rickard Liljeros, CPO at Nordiska Galleriet Group
17. Draft release notes
“I built a custom GPT to write product updates. Here’s a real example.”
—Michael Rumiantsau, founder of Narrative BI
18. Search through your wiki in Slack
“We have a GPT that is trained on all of our Atlassian wiki pages and built a Slack bot to query it.” —Anonymous
19. Track competitors
“We are doing competitor and category tracking in social with LLMs.”
—Noam Cohen, co-founder of Tetrix
20. “Clone yourself” to reduce meetings
“I have made myself a clone [GPT] bot that is trained on how I like to write project proposals, PRDs, roadmaps, and more, from the PM perspective [that has] saved me endless hours of work.
It provides a lot of autonomy for my marketing team. I do not have to write content for them. For my product team, we can quickly take client conversations and rapidly return a PRD and/or proposal. This saves me countless hours of middle management.”
—Traci Levine, CEO and founder of makeitMVP
If you’ve found other interesting uses for custom GPTs at work, please share in the comments!
Note: You need ChatGPT Team or ChatGPT Enterprise to restrict access to your GPT to just your company. Otherwise, you need to rely on keeping the URL secret (which is probably fine in most cases).
Bonus: Publicly available GPTs for product managers
- ChatPRD by Claire Vo: An on-demand chief product officer that drafts and improves your PRDs, while coaching you to become an elite product manager.
- PM Product Sense interview prep: Tailored for Big Tech product-sense-type PM interviews, this GPT provides ideal responses to generated or your own questions, with a focus on AI-powered solutions (and provides a sample mock).
- Landing Page Optimizer: Find actionable changes in landing page copy to improve conversion, by submitting a URL.
- Release Notes Writer: Customer-centric copywriter for software release notes.
Know of others? Leave a comment!
