Whitepaper · August 2026

New Media. ChatGPT Apps.

A plain-language guide to ChatGPT apps in Indonesia

For brand and agency leaders · No technical background needed · August 2026

A new place has opened, and almost nobody has moved in yet.

Millions of Indonesians now type questions into ChatGPT the way they once typed them into Google. Which flight is cheapest? Can I afford this house? What does this insurance actually cover?

Here is what most people don’t know yet: since late 2025, companies have been able to place their own branded tool inside ChatGPT.[1] When a customer asks a relevant question, your company’s tool can appear right there in the conversation — showing real flight prices, running a real loan calculation, listing real products.

The audience for this is already enormous. The number of companies that have actually done it is still very small. That gap — a huge audience, almost no competitors — is the entire opportunity, and this document explains it in plain terms.

This is not the first time a huge number of people have gathered somewhere new before the companies serving them caught up. It has happened before: a new place opens, people arrive first in large numbers, and the businesses that matter to them follow later — usually much later, and usually only once it’s already crowded. Indonesia is inside that early window right now. Windows like this tend to be short, and they close as more companies notice.

Four numbers that make the case

58M
Indonesians who use ChatGPT every month — roughly 3.5x Google’s Gemini

[5]

~65
Company tools live inside ChatGPT worldwide, as of July 2026 — not in Indonesia, in the whole world

[4]

0
Southeast Asian countries where you can buy advertising inside ChatGPT — it exists in eight countries

[8]

18%
Indonesian internet users who have ever used any AI tool at all — most of the country hasn’t arrived yet

[6]

Why isn’t everyone already doing this?

Fair question, and worth answering honestly before anything else.

Part of it is simple timing. Companies could only start submitting their own tools in December 2025.[2] Most companies take longer than eight months to even finish evaluating something this new, let alone build and launch it.[13]

Part of it is a queue, not a rejection. Companies that do try are reporting waits of six weeks to two months just to get reviewed.[12] OpenAI itself has told developers that submission volume has increased — meaning more companies are trying than the small number of published tools suggests.

And part of it is a real, honest caution worth taking seriously. There is no proven return yet — no company has published numbers showing this pays off.

What this means for you

The caution is real, and it is exactly why the recommendation in this document is a small, cheap, reversible first project — not a big commitment. The early window described above doesn’t stay open by itself; it closes the moment enough companies notice it. Acting small and early is how you get the benefit of that window without betting on an outcome nobody can promise yet.

What “an app inside ChatGPT” actually means

It is not a mobile app. Nobody downloads anything.

Here is how it actually works: a customer is chatting with ChatGPT about a holiday in Bali. ChatGPT recognises that your airline can help, and your airline’s branded flight search appears inside that conversation — real routes, real prices, your logo. The customer never left the chat.

People end up using your tool one of two ways.

01

ChatGPT offers it, mid-conversation

You cannot pay for this. It works a bit like Google’s search results before anyone could buy ads: you earn placement by being genuinely the most useful answer.

02

Someone looks for it directly

ChatGPT lists every company’s tool on a page people can browse or search by name. This is the path you can actually market — “find us inside ChatGPT,” on your website or receipts.

Either way, it only works if your tool does one job well. OpenAI’s own guidance to companies building here says a tool should do its job better because it lives inside ChatGPT, not just repeat your website.[3] A tool that’s really just your homepage gives ChatGPT nothing specific to match against a customer’s question, so it rarely gets used. A tool built to answer one question really well is the one that does.

The honest limitation

This is a discovery channel, not a selling channel — ChatGPT already tried letting people buy directly inside the chat, and removed it. OpenAI built a feature that let people buy things without leaving the chat. In March 2026 they switched it off. Walmart found that customers who tried to buy inside ChatGPT completed their purchase only about one-third as often as customers on Walmart’s own website.[7]

Research explains why: about 22% of people have bought something inside an AI tool, but 50% have bought something after asking an AI tool for advice, and roughly two in three are uncomfortable typing their card details to an AI.[13]

“People are happy to research with AI. They still want to buy on your own website or app.”

So this is not a sales channel. It is a discovery channel — the place where a customer forms an opinion about who to trust, before they go to your website to act on it.

Where it fits, by industry

  • Airline or travel company — show real routes and fares while someone plans a trip
  • Bank or insurer — calculate what a customer can afford, or whether they qualify
  • Retailer — help someone find the right product, then send them to your site
  • Telco — match a customer to the right package for how they actually use data
  • School, hospital or public body — answer common questions accurately and in Indonesian

Worth knowing: education and research is the single biggest reason Indonesians use AI, at 30% of all usage.[6]

If you run a brand, or if you run an agency

The situation is unusual and temporary. A large audience already exists, almost no competitor has claimed a place in it, and there is no auction to bid in. Those three things rarely occur together, and they will not last.

What to do: commission one small, genuinely useful tool. Choose the single question your customers ask most often, and answer it better inside ChatGPT than anyone else does. Then tell people about it directly — don’t just wait for ChatGPT to surface it on its own.

How to judge it: count people who used the tool, finished what they came to do, and then arrived on your website. Do not count sales inside the chat — there won’t be many, and that isn’t the point.

The trade: being early costs you one small project. Being late means paying to reach an audience a competitor already reaches for free.

There is nothing to buy here, so this is not media buying. It is a build-and-own service — closer to how you sold website work and search optimisation than how you sell advertising. Project fees and retainers, not commission on spend.[11]

Four places the work sits:

01

Getting found

Two skills: naming and describing a client’s tool so ChatGPT offers it on its own is genuinely new and needs a retainer. Telling customers directly where to find it is familiar campaign work.

02

Designing the conversation

Deciding what the tool does and how it feels is design work, not programming work — where agencies are already strong.

03

Building it

The tool is built on an open technical standard, so the same work can be reused if other AI services matter later.[1]

04

Proving it worked

Nobody has agreed what good looks like yet. The agency that defines sensible measurement for Indonesian clients will own the conversation.

The risks — and what to do about each

No official report proves this works.

No company or independent body has published data showing that a tool inside ChatGPT actually pays off. → What to do: treat the first project as a small experiment you measure yourself. Build in your own before-and-after numbers from day one.

You cannot buy your way in.

No advertising, no guaranteed placement, no promised audience size. → What to do: never promise a client a specific number of people reached. Measure only what actually happened.

There is no money in it directly.

OpenAI takes no cut and pays none — any revenue comes from customers who arrive on your own website. → What to do: value it against what you currently pay to attract that same visitor through search advertising.

Indonesian data law is strict.

Fines up to 2% of annual revenue, and prosecutions have already happened.[9]What to do: make the first version answer general questions only — no logins, no personal data leaving your systems. Start the legal conversation before you start building.

OpenAI might build your feature themselves.

If they do, your tool becomes worthless overnight. → What to do: build on something only you have — your prices, your inventory, your terms.

The AI might say something wrong next to your logo.

Customers may blame you for it. → What to do: make your tool return checked facts, not AI-written prose. Agree in advance what triggers you to switch it off.

The audience is thinner than it looks.

58 million users sounds large, but only 18% of Indonesians online have tried AI at all. → What to do: base your business case on what the pilot actually delivers, not on how big the market might become.

The recommendation

Do one small project. Make it read-only and anonymous in the first version, so the legal review stays simple. Agree before you start what success looks like and when you would walk away. Done that way, none of the risks above are serious. Done as a big transformation programme, every one of them becomes a problem.

A short glossary

ChatGPT
the AI assistant made by OpenAI that people chat with to ask questions or get help.
An app inside ChatGPT
a branded tool from a company that appears within the chat when relevant. Nothing to download.
OpenAI
the American company that makes ChatGPT and sets all the rules for this.
Open standard
a shared technical method any company can use, so work built once can be reused elsewhere rather than being locked to one provider.
Pilot
a small, deliberately limited first project, run to learn whether something works before committing further.

Sources

  1. [1]Introducing apps in ChatGPT and the new Apps SDK — OpenAI
  2. [2]Plugins in ChatGPT and Codex — OpenAI Help Center
  3. [3]Monetization, Apps SDK — OpenAI Developers
  4. [4]State of ChatGPT Apps 2026 (Mid-Year) — ChatGPTAppsRank
  5. [5]Daftar Aplikasi AI Terpopuler di Indonesia pada Mei 2026 — DataIndonesia.id
  6. [6]AI Adoption in Indonesia Remains at 18.2% — Intimedia
  7. [7]OpenAI Ends Instant Checkout, Pivots to Retailer App Integration — Stellagent
  8. [8]Where ChatGPT Ads Are Available: Country List — LocalAds
  9. [9]Data Protection & Privacy 2026, Indonesia — Chambers and Partners
  10. [10]What Happened to ChatGPT Plugins? — Drag
  11. [11]The Apps In ChatGPT Playbook — Fantasy
  12. [12]ChatGPT App Review Pending for 6+ Weeks — OpenAI Developer Community
  13. [13]ChatGPT Apps SDK Business Opportunity: Build Now, Wait, or Skip? — Arsum

Thinking about a pilot for your own brand or clients?

One email is enough to start a plain conversation about whether this makes sense for you, and what a small first version could look like.

Prefer to write it yourself? me@irvinesoftware.dev · Call or text +1 949 346 7578 · Leading a team? See the technical-leader page