What people ask AI to build: 172 real app ideas (2026 data)
We read the first prompt of every real AppX project: 172 app ideas from 124 people in 11 languages. What they ask for, how they write it, what happens next.
Ilyos Olimov · · 10 min read

AppX is an AI mobile app builder, and this is our own data. We read the first message of every project that real users started on AppX between May and 15 October 2026. We publish only totals. No prompt is quoted, and no person or company can be identified. The method and its limits are at the end.
Most writing about AI app builders is about the tools. Very little is about the people using them: what they ask for, how they ask, and what they do next. We have that data for our own product, so we counted it.
Key findings
- 172 app ideas from 124 people. Of 271 people who opened an AppX account, 124 (46%) described at least one app. (AppX, May to October 2026)
- People write specifications, not wishes. The median first prompt a person wrote themselves was 255 words long. 46% were 300 words or longer, 16% were 1,000 words or longer, and only 15% were 20 words or shorter.
- Half of the prompts were formatted like a document. 51% used headings, bullet points or numbered lists.
- People expect a real app, not a mock-up. 54% of prompts mention a backend, a database, an API or accounts, and 41% ask for sign-up or login specifically. 22% say outright that they want a production-ready or fully working app, not a prototype.
- Business tools were the most common idea. 17% of self-written prompts asked for an internal business tool: inventory, CRM, bookings, staff or fleet management. Games were the rarest, at 1%.
- A quarter did not write in English. 26% of prompts were written in one of 10 other languages, led by Uzbek (10%), Portuguese (5%) and Russian (3%).
- Android comes up three times as often as iOS. 29% of self-written prompts mention Android, and 10% mention iOS or the iPhone.
- 15% ask for a copy of an existing app or website, by name or by link.
- Building is a conversation. The median project had 3 build or edit turns. 84% of projects got at least one more message after the first build, and 7% had 10 or more.
- Most people try one idea. 82% of people who described an app started exactly one project.
What people ask AI to build
We sorted every self-written first prompt into one category by hand. Starter ideas that people picked with one click from our dashboard are counted separately below, so they do not inflate a category.
| Category | Prompts | Share |
|---|---|---|
| Business tools and internal systems (inventory, CRM, bookings, staff, fleet) | 25 | 17% |
| Productivity and personal tools (habits, to-do, notes, calendars, timers) | 15 | 10% |
| Education and learning (exam prep, languages, memorization, kids) | 14 | 9% |
| Media, music and creators (players, news, readers, camera, portfolios) | 13 | 9% |
| Marketplaces, services and shopping | 13 | 9% |
| AI assistants and AI-first apps (chat, voice, keyboards, agents) | 13 | 9% |
| Social, dating and messaging | 12 | 8% |
| Money and finance (wallets, expenses, crypto, invoices) | 10 | 7% |
| Health, fitness and safety | 10 | 7% |
| Lifestyle and home | 5 | 3% |
| Transport and logistics | 4 | 3% |
| Games | 2 | 1% |
| Other or unclear | 14 | 9% |
| Total | 150 | 100% |
Three things stood out.
Small businesses are the biggest group. The most common request was not a consumer app but a tool for running a business: a stock and sales app, a booking system for a barbershop, a management app for a farm, a dashboard for a hotel. These people already have a business and want software for it.
Education is bigger than we expected. 9% asked for learning apps: exam preparation, language practice, memorization trackers and apps for children.
AI apps are a category of their own. 9% asked AI to build an AI product: a chat assistant, a voice assistant, a speech-to-text keyboard, an AI tutor.
The remaining 22 projects (13% of all 172) started from one of the five starter ideas on the AppX dashboard. The habit tracker was picked 10 times, the freelance invoice tracker 6 times, the expense splitter 3 times, the portfolio 2 times and the booking calendar once.
How people write their prompts
Prompt length, counted in words, for the 150 prompts people wrote themselves:
| Words | |
|---|---|
| Shortest quarter | 40 or fewer |
| Median | 255 |
| Longest quarter | 641 or more |
| Longest prompt | 7,743 |
| What the prompt contains | Share of prompts |
|---|---|
| Headings, bullet points or numbered lists | 51% |
| A backend, database, API or accounts | 54% |
| Sign-up or login | 41% |
| A named technology, such as React Native, Expo, Flutter or TypeScript | 36% |
| Android | 29% |
| A demand for a production-ready or fully working app | 22% |
| A named app or website to copy | 15% |
| iOS or iPhone | 10% |
| A role for the AI, such as "you are a senior developer" | 9% |
| A request for a website or web app instead of a mobile app | 5% |
Many people arrive with a document they have already written, often a product requirements document with sections, feature lists and screen lists. Others write one line. Both kinds of prompt are common, and a builder has to handle both: a long specification needs to be cut into steps, and a one-line idea needs questions or sensible defaults.
The share of prompts that ask for accounts and a backend matters. A screen-only mock-up does not meet what most people ask for. If you are writing a prompt yourself, say plainly whether your app needs accounts and where its data should live. Our guide on how to describe your app to AI has a template.
Languages
| Language | Prompts | Share |
|---|---|---|
| English | 128 | 74% |
| Uzbek | 17 | 10% |
| Portuguese | 9 | 5% |
| Russian | 6 | 3% |
| Spanish | 4 | 2% |
| French | 3 | 2% |
| Arabic, Bengali, Hindi or Urdu in Latin script, Polish, Ukrainian | 1 each | 3% together |
This counts all 172 first prompts. The share of Uzbek and Russian reflects where AppX started: the product is built in Uzbekistan and its interface is in English, Russian and Uzbek. Some people wrote in one language and asked for the app's text in another, for example a prompt in English for an app with Arabic right-to-left screens.
What happens after the first build
| Projects | |
|---|---|
| Projects with at least one build | 161 |
| Median build and edit turns per project | 3 |
| Projects that got at least one more message after the first build | 135 (84%) |
| Projects with 5 or more turns | 35 (22%) |
| Projects with 10 or more turns | 12 (7%) |
Of the 124 people, 102 (82%) started one project and 11 started three or more.
People rarely stop after the first build. They look at it, then ask for changes: a new screen, different colours, a fix. That is why we build bigger apps from an approved plan, in milestones, and why a cheap, quick edit matters as much as the first build. Our separate measurement of how long a build and an edit take has the timings.
What we could not measure
We wanted to answer a simple question: do longer prompts build better? Our records cannot answer it honestly yet. AppX moved to a new build system during this period, and one month of older records marks every first build as a success while the build log for the same month shows about half. A comparison across that change would mislead, so we leave it out. We will publish it when we have three months of records from the current system.
Other limits:
- One product. These are AppX users. AppX is mobile-first, built in Uzbekistan, and has charged from $5 a month since 1 October 2026. Users of a web-first or free tool would ask for different things.
- Small sample. 172 projects from 124 people. Treat a difference of a few points as noise.
- Hand sorting. One reader sorted the prompts into categories, and many prompts could fit two. The keyword shares (login, Android, technologies) come from simple word matching in several languages and will miss some phrasings.
- First prompt only. People often add detail later in the chat. We looked only at how each project started.
Method
- Data: the first user message of every AppX project created by a real user between May 2026 and 15 October 2026, read on 16 October 2026 from the AppX production database.
- Excluded: 6 accounts belonging to the AppX team and its test runs, 2 test or empty prompts, 1 message that was a question rather than an app idea, and 6 projects our team built on behalf of customers. That leaves 172 projects from 124 people.
- Starter ideas: 22 projects began from one of five example prompts on the AppX dashboard. They count in totals and languages, but not in the category table or the prompt-length and content tables, which use the 150 prompts people wrote themselves.
- Turns: a turn is one build or one edit that AppX ran in response to a message.
- Privacy: prompts were read only to sort and count them. This page contains no prompt text, app names or user details.
How to cite this
AppX (2026). What people ask AI to build: 172 real app ideas. https://appx.uz/blog/what-people-ask-ai-app-builder-to-build-2026
For wider numbers on AI-assisted development from surveys and platform reports, see our collection of AI app development statistics.
FAQ
What do people build most with AI app builders?
In AppX data from May to October 2026, the most common request was an internal business tool (17% of self-written prompts), followed by productivity and personal tools (10%). Education, media, marketplaces and AI-first apps were each about 9%. Games were 1%.
How long is a typical prompt for an AI app builder?
The median self-written first prompt on AppX was 255 words. A quarter were 40 words or fewer, and a quarter were 641 words or more. 51% were formatted with headings or lists.
Do people want a prototype or a working app?
Mostly a working app. 54% of prompts mention a backend, database, API or accounts, and 41% ask for sign-up or login. 22% say directly that they want a production-ready app rather than a prototype.
Do people keep editing after the first build?
Yes. 84% of projects got at least one more message after the first build. The median project had 3 build or edit turns.
Keep going
More practical ways to move your app forward.
- How to describe your app to AI: a prompt template
Write a clear AI app-builder prompt with a copyable template, a practical example, acceptance checks, and focused follow-up edits.
- How to turn a client brief into a mobile app prototype
Turn a client brief into a mobile app prototype with a copyable AppX prompt, a practical test checklist, and a focused client-feedback workflow.
- The 5-backtick fence: stopping prompt injection via Markdown
We ship user input into an LLM architect prompt every turn. A reviewer caught that a 4-backtick string from the user could close our 3-backtick fence and run arbitrary instructions on the model. The fix is one weird Markdown trick.
- Investigators before builders: a 2-phase LLM workflow
Most teams dispatch coding agents straight to the build. We run a read-only investigator phase first — and the findings have rewritten our plan in 30-40% of cases. Here's the shape, the brief format, and what to do when findings contradict the plan.
- Step 0: delete dead code before you refactor with an LLM
Every byte you load into an LLM agent's context is a byte it has to keep track of. Dead code looks like signal — and agents preserve, extend, or mimic it. Our rule: a pure-deletion commit before any structural refactor of a file over 300 LOC.
Ready to try your idea?
Start with a clear description. AppX will help you turn it into an app you can test.