Why focus on monetization instead of user growth for ChatGPT?

With 800M MAU, ChatGPT has massive scale but faces ongoing monetization challenges. The real opportunity lies in improving conversion from free to paid and increasing ARPU through vertical-specific solutions that command premium pricing—similar to how Cursor, Harvey, and other specialized AI tools have proven developers and professionals will pay for domain expertise.

Why choose developers over other segments like healthcare or education?

Developers have proven willingness to pay ($10-25/month for Cursor, GitHub Copilot, Replit), use coding tools daily, and represent a high-frequency use case. While healthcare has high willingness to pay, it faces compliance barriers. Education has mixed willingness to pay. Developers offer the fastest path to revenue with existing proof points.

How is this different from 'How would you improve Google?' type questions?

Both are broad improvement questions, but ChatGPT requires understanding the AI competitive landscape—model layer vs. application layer dynamics, vertical solution opportunities, and the moat-building challenge. You need to demonstrate strategic thinking about market positioning while still getting into specific product design, not just high-level strategy or narrow feature additions.

Why prioritize ChatGPT Canvas Pro over ChatGPT Code Studio?

Canvas Pro leverages existing assets (Codex, Canvas interface, enterprise distribution) with minimal partnership risk. Code Studio requires deeper VSCode forking and creates tension with Microsoft's GitHub Copilot. Canvas Pro targets non-developers who prototype then hand off to developers—a natural enterprise workflow with faster time to market.

OpenAI PM Interview: How Would You Improve ChatGPT?

Welcome to the 15th edition of PM Interview Prep Weekly! I’m Ajitesh, and today we’re tackling one of the most commonly asked questions in OpenAI PM interviews—one that seems simple but has layers of complexity beneath the surface.

I’ve been receiving requests to cover more OpenAI PM interview questions, which gives a glimpse into the kinds of questions that are popular in AI PM roles today.

One of the most commonly reported questions is: “How would you improve ChatGPT?” This question seems straightforward in theory, but it’s deceptively broad. Almost as complex as “How would you improve Google?” If you talk about high-level strategy without getting into design specifics, you’ll come across as someone who isn’t in the weeds and doesn’t understand the product craft. But if you jump straight into feature building or focus too narrowly on one aspect, you miss the chance to demonstrate strategic thinking and moonshot ideas—which are also a big part of why you’re asked these questions.

The interesting wrinkle here is the market dynamics in the AI field. There’s intense competition across the model layer, application layer, and infrastructure, with vertical solutions emerging everywhere. Anyone recruiting for an AI PM role should understand this landscape. What’s the winning strategy at different layers? How do you build a sustainable moat? You need to bring that perspective into the interview. In some ways, this question gives you a great opportunity to showcase your understanding of the AI space and come up with creative, strategic solutions.

Important clarification: ChatGPT in this context refers to the consumer product (chat.openai.com), not the APIs. It’s used by individual consumers and also by businesses through Team and Enterprise subscriptions, where it can connect to internal data sources and knowledge bases.

The Case

Interviewer: “How would you improve ChatGPT?”

The Interview Approach

Note: This framework is one approach, but feel free to take your own path in solving this problem.

Clarify Goals & Constraints - Understand what “improve” means and set a clear goal (adoption, engagement, monetization, retention)
Identify Customer Segments - Think beyond “people who use ChatGPT” and pick ONE segment based on monetization potential
Define the Problem - Map the customer journey and identify core problems that block conversion or limit value capture
Design Solutions - Create 3 solutions with increasing complexity
Define MVP & Success - Pick one solution, scope the MVP, and set clear metrics tied to your goal

Here’s how I would approach this case, focusing on the monetization opportunity that OpenAI is clearly pursuing.

My Approach

Goal Setting & Business Context

First, I’d clarify with the interviewer what “improve” means. Given ChatGPT’s 800M MAU but ongoing monetization challenges, I’d propose focusing on revenue growth as the primary goal.

Goal: Increase ChatGPT revenue by improving conversion from free to paid and increasing ARPU (average revenue per user) within 12-18 months.

I’d also clarify a few questions with the interviewer—for example, is it okay to focus on the U.S. market for now? And just to confirm, my understanding is that ChatGPT here refers to the consumer product, so the API is out of scope for this question

Before jumping into solutions, I’d share the business context that shapes my thinking:

Foundation models are becoming super competitive (OpenAI, Claude, Gemini all strong)
Application layer where consumer/business have higher propensity to pay; vertical-specific solutions command premium pricing
Competition from specialized AI tools (Cursor for coding, Harvey for legal, etc.) shows willingness to pay for domain expertise

This context tells me that when thinking about improving ChatGPT for monetization, the real question is: which application layer should we as a product team invest in? Where can we build vertical-specific value that commands higher pricing?

Customer Segmentation

Rather than thinking of ChatGPT users as one monolith, I’d segment by use case vertical and willingness to pay:

Healthcare professionals - doctors, nurses, researchers using ChatGPT for clinical documentation, diagnosis support, literature review, patient communication
- High willingness to pay (healthcare budgets are substantial)
- Compliance/accuracy needs create barriers (HIPAA, medical accuracy)
- Current Plus tier doesn’t address their specific needs
Education segment - students using for homework help, teachers for lesson planning, tutors for personalized learning
- Mixed willingness to pay (students low, teachers moderate, institutions high)
- High usage frequency (daily during school year)
- Institutional purchasing power through schools/districts
Developers & technical users - software engineers using ChatGPT for coding help, debugging, architecture decisions
- High willingness to pay (evidenced by Cursor, GitHub Copilot success)
- Face strong specialized competition (Cursor, Copilot, Replit Agent)
- Current Plus tier lacks coding-specific features (repo context, IDE integration)
Enterprise knowledge workers - analysts, consultants, marketers using for research, content, presentations
- Moderate-to-high willingness to pay (employer-paid)
- Need company data integration, security, compliance
- Partially addressed by Team/Enterprise, but pricing/packaging may miss smaller teams

Chosen segment: Developers & technical users

Why this segment?

Proven willingness to pay: Cursor ($20/month), GitHub Copilot ($10-19/month), Replit ($25/month) show developers pay for coding AI
High frequency: Developers code daily; this isn’t occasional use

Problem Deep Dive

Let me map the complete workflow of a software engineer to understand where value is created and captured: The process spans from product strategy formulation and feature brainstorming through PRD writing and design doc creation, followed by coding (including test cases), PR reviews, QA testing, and finally deployment.

Current state with ChatGPT: OpenAI recently launched Codex, an AI coding agent included in Plus, Pro, and Enterprise plans. Codex integrates into IDEs (VS Code, Cursor), terminals, and GitHub, and can handle multi-step autonomous workflows—like migrating codebases, updating tests, and generating documentation. This is a significant leap from just chat-based coding help.

However, most developers still don’t know Codex exists or haven’t adopted it. The primary interaction remains the web interface at chat.openai.com—used sporadically for debugging (step 4), occasional brainstorming (step 1), or doc writing (steps 2-3). These are one-off queries with no workflow continuity.

The opportunity: Given what Cursor and others have proven, the real monetization opportunity lies in the code-writing and test-writing space. Windsurf (formerly Codeium) - a late entrant that OpenAI tried to acquire for $3 billion in May 2025 before the deal collapsed—demonstrates both the market size and OpenAI’s own strategic interest. Developers love Cursor but aren’t locked in. The question: how do we build what OpenAI couldn’t buy?

Brainstorm three distinct solutions

ChatGPT Code Studio — A dedicated code editor for ChatGPT subscribers with usage-based pricing beyond the $20/month cap. VSCode-compatible, it brings ChatGPT’s full ecosystem—memory, web browsing, canvas—into the IDE. Developers can brainstorm features, search the web, and leverage Codex as an agentic layer, all within one unified workspace. Think: Cursor, but powered by ChatGPT’s broader toolkit.

ChatGPT Canvas Pro for Enterprise — A Lovable-style visual builder powered by Codex, targeted at enterprise teams already using ChatGPT. Non-developers build prototypes quickly using the Canvas interface, then hand off to developers who refine the code in their existing tools. This taps into ChatGPT’s existing enterprise penetration and creates a seamless workflow from ideation to production.

Model Layer Leap — Invest heavily in training and research to close the quality gap with Claude. Anthropic is widely considered superior for coding tasks—both in benchmarks and real-world performance. This is less a feature and more a foundational bet: can we retrain our models to match or exceed Claude’s coding capabilities? It requires researcher-led investment and may face fundamental constraints if Anthropic has unlocked a unique training advantage.

Prioritize and pick an MVP

ChatGPT Canvas Pro for Enterprise wins first:

Fastest path to market: We already have Codex and the Canvas interface; this is primarily an integration and packaging play.

Leverages existing distribution: ChatGPT already has enterprise penetration. Many Plus subscribers are non-developers who would benefit from a visual builder. This becomes a natural upsell.

Clear handoff workflow: Non-developers prototype → developers refine. This maps cleanly to how enterprise teams already work, reducing adoption friction.

Lower partnership risk: Unlike ChatGPT Code Studio, which could create tension with Microsoft (given GitHub Copilot), this targets a different user segment and workflow.

The Code Studio path requires deeper VSCode forking, Microsoft alignment, and competes directly with Copilot—making it a second-phase move once we’ve proven the enterprise visual-builder market.

The Model Layer investment is critical but orthogonal. As a PM, I’d advocate for parallel research investment, but acknowledge this requires a researcher perspective to validate feasibility. If Anthropic has a fundamental advantage we can’t replicate, we shouldn’t waste resources chasing parity—focus instead on differentiated features where our ecosystem (memory, browsing, Canvas) creates unique value.