The Opportunity

The opportunity

ChatGPT has 800M+ monthly users. Apps will appear directly in chat conversations - no separate app store to browse, no download screens, no install friction. Just natural conversation where the right app surfaces at the right moment.

Think about what that means for distribution.

When someone's drafting a legal contract in ChatGPT and says "I need a lawyer to review this," your app can appear right there. When they're writing tweet copy and want to make it visual, your app shows up in that exact moment of need. Zero clicks to discover. Zero friction to use.

This is fundamentally different from every app distribution model before it. The App Store required browsing, searching, reading reviews. Google required you to find the website first. Facebook apps lived in a separate section most users never visited.

ChatGPT apps appear in the flow of work. The user doesn't think "I should go find an app for this." They just express what they want and if your app matches that intent, you get the conversion.

OpenAI’s Apps SDK just dropped in preview in October 2025. The app directory doesn't exist yet - that's coming later this year.

The opportunity to win in this new distribution channel is wide open.

The four phases of distribution

We have seen this before. Major technology and distribution shifts follow a repeating four-step cycle. ChatGPT is now at a pivotal inflection point within this cycle. This framework - drawn from examples like Facebook's and Google's platform cycles - shows exactly where we are and what's coming.

Phase 1: Market conditions met

There's broad consensus that a massive new category or platform is emerging, but no definitive winner has surfaced yet.

Several major players (OpenAI, Google, Anthropic, Meta) are competing intensely, each looking for the edge that leads to future monopolies or duopolies.

For LLMs, this started in 2022-2023. Everyone knew AI was huge but nobody had won yet.

Phase 2: Identifying the moat

One player begins to distinguish itself by identifying and building a defensible moat. For OpenAI, their ChatGPT product has become synonymous with AI for the majority of the consumer market. They've hit 800M+ monthly users. They're winning the race.

To scale quickly, the leading platform recognizes it needs third-party developers or content creators and starts constructing an ecosystem.

This is where platform thinking kicks in. You can't build everything yourself. You need other people creating use cases you'd never think of.

Phase 3: Platform opening ← WE ARE HERE

The leader opens up to third-party integrations, offering distribution and incentives (such as access to users or new monetization avenues) to developers and businesses.

This triggers a "gold rush" where companies can experience hypergrowth if they move early.

This is the phase we just entered. Apps SDK released in preview. Developer Mode is live. No app directory yet but that's coming. The gates just opened.

The window for outsized growth will be brief. Distribution cycles are shortening over time. The iPhone App Store gold rush lasted maybe 2-3 years. Facebook Platform maybe 18 months. This one might be 12 months or less.

Companies must move quickly to avoid being left behind.

Phase 4: Platform closing & monetization (coming 2026+)

Eventually, the platform restricts or "closes" the ecosystem for reasons like monetization, growth, or competitive protection.

Strategies include:

• Building their own first-party products to absorb popular use cases
• Suppressing organic distribution
• Shifting towards paid distribution for third parties
• Taking larger revenue cuts
  

This cycle is visible in major platforms like Facebook, Google, and Apple historically. They all did this. OpenAI will too.

Where this leaves you: Platform strategy

The brutal risks:

They control your distribution completely. OpenAI can feature your app and you get 100k users overnight. They can bury your app in search results and you get nothing. They can change the discovery algorithm tomorrow and your carefully optimized metadata stops working. Zero control.

They might just clone you. This happened constantly in the App Store. Company builds successful app, Apple looks at it and thinks "that's a great feature... let's just build it natively in iOS." App is now worthless because everyone has the native version. OpenAI will do the same thing. Your success actually increases the risk of being Sherlocked.

Revenue model is completely unclear. Monetization details are "coming later this year." That could mean revenue share (probably), but what percentage? 30% like Apple? 15%? 50%? Who knows. They might limit certain business models entirely. You're building without knowing the unit economics.

How to play smart:

Build distribution outside ChatGPT from day one. Email list. Web version. Social following. Other distribution channels. You need to be able to survive if OpenAI changes the rules or shuts you down.

This isn't pessimism - it's just smart business. Every time you get a user through ChatGPT, capture their email or get them to create an account on your platform. Give them a reason to come back to you directly, not just through ChatGPT.

Move fast, build moat, be ready to pivot. First mover advantage is real in platforms like this. The apps that launch early get more usage, more feedback, better optimization. Network effects compound. Users learn which apps they like and stick with them.

But also be ready to pivot if the platform changes. Your moat can't just be "we're the ChatGPT app for X." It needs to be "we're the best solution for X, and ChatGPT is one distribution channel."

Watch for signals the game is changing. Policy changes, revenue structure announcements, feature restrictions, apps getting removed. When OpenAI starts tightening control, you need to already have your backup plan in motion, not scramble to create one.

Why you can't opt out (even if you want to)

Here's the uncomfortable truth: you don't get to sit this one out.

When a new platform opens, staying out feels principled. "We're not going to depend on another platform." "We'll focus on our own channels." But it's strategically untenable.

Competitors who integrate early gain compounding advantages. They get distribution. They learn faster. While you're "staying focused," they're reaching escape velocity. Users experience faster workflows through their apps, and suddenly your product feels slower and clunkier in comparison.

User expectations shift fast. If someone's drafting a contract and ChatGPT connects them to a lawyer through Upwork's app but not yours, they're not thinking "Fiverr chose not to build a ChatGPT integration for strategic reasons." They're thinking "Upwork works in ChatGPT and Fiverr doesn't. I'll just use Upwork."

This is a prisoner's dilemma. Even with the downsides (and we covered the risks), defecting alone leaves you worse off. Others will play and reap the benefits while you watch from the sidelines.

Opting out doesn't freeze the game. It just sidelines you. Competition moves. User expectations move. The platform's economics move. Standing still isn't strategic patience - it's strategic suicide.

This experience will become an EXPECTATION. As workflows move inside ChatGPT, if your brand doesn't have an app in someone's workflow, they'll use a competitor that does. You won't even know you lost them.

Key takeaways from this framework:

Historically, new distribution channels are rare but create huge opportunities for growth, especially before they become saturated.

Startups benefit most by moving fast to leverage these openings—late adopters will be at a severe disadvantage.

The imperative is to integrate early, as customer expectations and competitive pressure will force everyone into the game eventually.

What should you build?

A good ChatGPT app should answer "yes" to most of these questions:

Does this task fit naturally into a conversation? Booking, ordering, scheduling, quick lookups. Things people already express conversationally. "Book me a table at 7pm" or "Create a gig for this contract" feel natural. "Browse our entire catalog" doesn't.

Is it time-bound or action-oriented? Short or medium duration tasks with a clear start and end. Create a gradient tweet. Post a job. Generate a diagram. Not "explore our full product line for 30 minutes."

Is the information valuable in the moment? Users can act on it right away or get a concise preview before diving deeper. A freelance gig card they can edit and post immediately. A tweet graphic they can download now. Not a 50-page document they need to study.

Can it be summarized visually and simply? One card, a few key details, a clear CTA. The ChatGig gig card shows budget, deadline, description - everything needed to make a decision. Not trying to cram an entire dashboard into chat.

Does it extend ChatGPT in a way that feels additive or differentiated? It does something ChatGPT can't do alone. Connecting to external services, creating specific visual formats, handoffs to humans. Not just rewording what ChatGPT already does well.

What works well & what to avoid

What works well:

Fixed templates with clear outputs Gradient tweets, diagrams, charts, forms. Anything where the format is predictable and the output is clear. Everyone knows what a gradient tweet looks like—they just need to customize it. The chat interface is perfect for "I want X, make it look like Y" workflows.

Workflow handoffs to external services This is the big one. Natural transition points from "working in ChatGPT" to "need an external service." ChatGig fits here - you're drafting work in ChatGPT and need to hand it off to a human expert. The key is preserving context across that handoff.

Eliminating context switching Users are already in ChatGPT, in flow state. Anything that keeps them there instead of forcing them to open another tab is valuable. Even if your app isn't dramatically better than the web version, removing the friction of switching can be enough.

What to avoid:

Long-form or static content better suited for a website If users need to scroll through pages of content or spend 20 minutes reading, that's not a chat experience. Build a webpage. The chat window is for quick, actionable information.

Complex multi-step workflows If your workflow requires 10 steps with conditional branching and lots of back-and-forth decisions, the chat interface probably isn't optimal. Users get frustrated. You're fighting against conversational UI constraints. Wait for better primitives or just use a webpage.

Heavy visual browsing Shopping, design exploration, anything where users need to see 20+ options at once and compare them visually. You can't show a grid of 50 products effectively in chat. The window is too limited.

Ads, upsells, or irrelevant messaging Don't use the chat space for marketing. Users came to accomplish a task, not see your promotional content. This will destroy trust and may get you removed from the platform (at least in the early stages, I could see this happening later down the line).

Duplicating ChatGPT's system functions Don't try to rebuild text analysis, summarization, basic research. If ChatGPT can already do it natively, you're not adding value. Find the gaps where ChatGPT needs external data, external services, or specialized functionality.

By following these principles, your app will feel like a natural extension of ChatGPT rather than a bolt-on experience.

How to get discovered

This is critical. You could build the best app but if it isn’t found or brought into people’s conversations, it may as well not exist. Now, what we’re about to go through isn't speculation. These are learnings from directly reading what OpenAI have published in their docs and from our early testing.

In short, when you type something into ChatGPT, it reads your app's metadata and makes a decision about whether to invoke your app.

Here's exactly what it's looking at:

Understanding the algorithm

Name structure matters way more than you think Use domain.action format. calendar.create_event is good. myCalendarApp is bad. ChatGPT uses this to understand what your tool does semantically.

Description clarity is everything Start with "Use this when..." and include "Do not use for..." scenarios. Be brutally explicit. "Use this when the user wants to create a social media graphic with a gradient background" beats "Creates images" by a mile.

The description does two jobs: tell ChatGPT when to call you, and when NOT to call you. Most people skip that second part.

Parameter documentation with examples Describe every parameter with actual examples. If you have background_color, show: "#FF5733", "rgb(255, 87, 51)", "coral". Use enums for constrained values. The more specific you are, the better ChatGPT understands how to fill parameters from natural language.

Negative prompts prevent disasters Tell ChatGPT when NOT to use your app. Without this, you get called at weird times. Example: "Do not use for general questions about note-taking apps. Only use when the user explicitly wants to create, read, or modify content in their Notion workspace."

Before you write any code: the golden prompt set

Define three categories of prompts. This becomes your entire testing framework.

Direct prompts - Users explicitly name your app:

• "Use Notion to create a task called X"
• "ChatGig, create a gig for this contract"

Indirect prompts - Users describe outcomes without naming your tool:

• "I need to organize my notes" (should maybe trigger Notion)
• "Help me find a freelancer to finish this contract" (should trigger ChatGig)

Negative prompts - Cases where your tool should NOT activate:

• "Tell me about the best project management tools" (don't activate Notion)
• "What's the history of the gig economy" (don't activate ChatGig)

Write down expected behavior for each. This is your testing baseline.

Writing your metadata

For each tool:

Name it with domain + action pairs. gig.create not createGig.

Write descriptions that convert:

• "Use this when [specific scenario with details]"
• "Do not use for [specific anti-patterns]"
• Examples for every parameter

Add readOnlyHint: true for tools that never change state.

At app level:

Polished description affects discovery ranking. Keep it short, clear, action-oriented.

Icon that stands out. You're competing visually with dozens of apps.

Starter prompts showcase your best use cases. "Create a gradient tweet from my latest blog post" is great. "Use Gradient Tweet" is lazy.

Sample conversations show real examples. Users learn by seeing.

The testing & iteration loop

Setup: You need ChatGPT Plus ($20/mo) for Developer Mode.

Settings → Apps & Connectors → Advanced Settings → enable Developer Mode.

Use ngrok for local testing or deploy your MCP server publicly.

Process:

1. Link your connector - Add custom connector, paste MCP URL, complete OAuth if needed
2. Run your golden prompt set - Test every prompt in all three categories
3. Track two metrics:
   • Recall: Did ChatGPT call your app when it should?
   • Precision: Did it only call when appropriate?
4. Change ONE thing at a time - Modify one metadata field, retest everything, log results
5. Repeat - What doesn't work today might work tomorrow as the algorithm learns

Keep a spreadsheet: Prompt → Expected → Actual → Pass/Fail.

Log everything with timestamps: "Changed description to include 'use when creating visual content' - 3pm Oct 15 - recall 60%→75%, precision stayed 90%."

This feels tedious but it's how you actually optimize. No magic formula. Just iteration based on real results.

Production monitoring:

Once live, you need:

• Weekly tool-call analytics (which prompts trigger you, any failure patterns)
• Spike monitoring (sudden increase in "wrong tool" confirmations means something changed)
• User feedback loops (capture issues immediately)
• Periodic prompt replays (retest your golden set every few weeks)

Final Thoughts

A few things to remember:

Success isn't just about building a great product. It's about understanding the metadata game, the discovery algorithm, the platform dynamics.

Never depend entirely on this platform. Build your own distribution, capture emails, have an exit plan. Platforms always consolidate control eventually.

The apps we're building today are the worst ChatGPT apps will ever be. The platform will get better UIs, better auth, better everything. What's hard today will be easy in 6 months. But the early distribution wins will already be locked in.

Find natural workflow handoffs. Don't try to rebuild ChatGPT or create complex flows the chat interface isn't suited for. Find the moments where users are already in ChatGPT and would normally leave to use another tool. That's your app opportunity.

Join communities of other builders. We're all figuring this out together. The people sharing what they learn will move faster than the people building in isolation.