From side project to a product AI can name

I turned an app built for my mom into nine products in five sessions. The sixth step most vibe coders skip: making it something an assistant can name.

By Dasha. Updated November 1 2026. https://heardofyou.io/guides/side-project-to-product-ai-can-name

Most side projects die because nothing is at stake, not for lack of time. They come alive when built for a real person with a real problem. They become a product when the parts specific to that person turn out to be thin.

I learned this by building a small business app for my mom. Over five working sessions, it turned into nine products.

Here is that process. And the sixth step I did not know about then: making the product something ChatGPT, Claude, Gemini or Perplexity can name when someone asks.

It started as an app for my mom

My mom makes dumplings for a living. She ran the business out of notebooks and chat messages.

I built her an app: receipts, invoices, freezer inventory, and an assistant she could talk to in Russian. It worked because she used it every day and told me when it did not.

It became a product the day I noticed almost nothing in it was about dumplings. Receipts, invoices, inventory and a chat assistant are universal. Only the branding and a few hard-coded words were hers.

The five sessions

1. Remove the hard-coded parts. Every reference to her business became a setting. One session, mostly search and replace, plus a settings screen.
2. A setup wizard. Seven steps on first run: business name, what you sell, currency, tax, language, the basics. The wizard turns "my mom's app" into "an app".
3. Teach the assistant from the database. Her business was written into the AI prompts. Now they read the business settings at run time. The AI features got better, because the context was explicit instead of assumed.
4. Feature flags. With features behind switches, one codebase could split into smaller products that each do one job.
5. Package. A README, a setup script, a build script and a clean bundle for each product.

One codebase, nine products. Every bug fix lands in all nine. That part still feels like a trick.

Five tests before you productise anything
- It solves a workflow problem, not a curiosity.
- Someone already uses it and would notice if it stopped.
- The domain-specific parts are thin.
- The tech stack is one other people use.
- You can explain the value in one sentence.

That last test matters more than I knew. It is not just a pitch test. It is the sentence an assistant needs.

The sixth step: make it nameable

The five sessions gave an AI assistant no reason to know the products existed.

Assistants decide what to recommend (/guides/how-ai-decides-what-to-recommend) by searching the web, reading what other people wrote, and breaking ties for names they already know. A freshly packaged product has none of that.

So here is the sixth step, in the order I would do it now.

Pick the category word. Buyers ask for "an invoicing app" or "a freezer inventory tool", never for your clever name. If your product does not sit in a category a buyer would type, the assistant has no question to match you to. Choose the plain word, even if it feels small.

Write the one sentence, and put it at the top. "[Name] is a [category] for [who], that [one thing]." On the homepage, first paragraph, as text, not an image. That is the sentence an assistant quotes. Our guide on why AI doesn't recommend your product (/guides/why-ai-doesnt-recommend-you) has a test for it.

Make the site readable. A page drawn entirely by JavaScript, or blocked to crawlers, is invisible to an assistant. Our AI crawler check (/tools/ai-crawler-check) tells you in a minute.

Get three honest mentions.
- A review from the real person you built it for.
- A thread where you answer, under your own name, a question your product solves.
- One list or comparison page that names you.

Three is not many. It is the difference between zero third-party pages and some. What works and what doesn't (/guides/what-works-and-what-doesnt) shows why this beats anything on your own site.

Then measure. Ask each assistant five buyer questions, three times, and count how often you are named. A rate, not a rank. The method is on how we measure (/how-we-measure). The do-it-yourself version is how to check if ChatGPT recommends you (/guides/check-if-chatgpt-recommends-you).

Expect zero the first time. The point is a number to beat at day 30.

What I would tell the me with the dumpling app

Build it for one person. Notice when the specific parts are thin. Split it only if the split is clean.

And before you spend a weekend on packaging, spend an hour on the category word, the sentence and the first three mentions. Packaging makes a product real to you. Mentions make it real to an assistant.

A first version of this appeared on my studio blog, dashbuilds.dev.

What to do next

Is your side project already a product? Find out whether any assistant knows it exists. The free check (/check) runs 5 buyer questions, 3 times each, on ChatGPT, Gemini, Claude, Perplexity and Muse. It shows who was named instead and which sources they cited.

Built it with AI coding tools and wondering what comes next? The vibe coders (/for/vibe-coders) page is written for you.

Questions

My side project has no users yet. Is it too early for any of this?

Too early to pay anyone, yes. Not too early to pick the category word and write the one-sentence description. Those decide which questions you could ever answer. Do those two, then find your first real user.

What counts as a real person with a real problem?

Someone who would notice if the thing broke. My mom ran her business out of notebooks and messages, and the app replaced them. Those are the stakes a side project needs. It is also your first honest review.

Do I need nine products?

No. I got nine because one codebase happened to split cleanly. One product with a clear category, a real page and three mentions beats nine that nobody can name.

Can AI assistants find a product with no reviews at all?

Rarely. They lean on pages other people wrote. A product with zero third-party mentions is usually invisible to them, even with a perfect site. So the first real user matters more than the launch.