First Users Without an Audience: Test 20 Prompts
A no-distribution product earns early users by making the job clear enough to be suggested in a moment of need.
You have a tool, no audience, and a stranger in ChatGPT who is stuck on a task. The first hundred users can come from a suggestion in that moment, not from a launch you cannot afford.
The unscalable work remains: talk to a few people, sit with their screens, and fix the one sentence that makes the tool obvious. Do not buy a launch. Do not chase a trend. Those tactics assume a scale they do not have. A small product has only a job, and the job has to be said plainly.
The first hundred come from a mid-chat suggestion
Before September 29, 2026, ChatGPT plugin discovery was limited to the plugin directory or naming the plugin. That made distribution a search problem: you had to be found by someone who already knew what to ask for. OpenAI's DevDay on September 29, 2026 started ChatGPT's mid-chat plugin suggestion feature. When a user describes a job, ChatGPT searches its plugin directory and shows matching plugins in a picker that includes an Install button. The entry point is a stuck task, not a category search.
As of October 2026, no paid placement for ChatGPT plugin suggestions had been announced. For a founder with no budget, clarity is the first advantage, speed the second, and a clear no the third. If a stranger's prompt matches your job, the tool can appear at the exact moment the need is felt.
Twenty prompts are your pre-launch test
Run a 20-run prompt test. It is the cheapest way to find out whether your metadata says the job before a stranger does. You are testing whether the tool can be suggested, not traffic.
Say you build a tool that turns messy meeting notes into action items. For the test, use prompts that directly state the task or indirectly describe the need. The third prompt should not match your job.
- The metadata line begins with “Use this when…” and names one painful task.
- The metadata line includes “Do not use this when…” and names one wrong job.
- You have 20 prompts: direct, indirect, and negative.
- Each prompt is written as a user would ask, not as a feature list.
- You can say which field changed after each failed run.
Metadata is the job, not the tagline
A ChatGPT plugin's title, short description, example prompts, and tool descriptions affect whether it shows up in the directory or the in-chat suggestion picker. Since July 2026, ChatGPT plugins have been packages that can include an MCP-connected app, skills, or both. The assistant reads the fields that describe what the plugin does, not your marketing site.
The negative line is the part most founders skip. It tells the picker when to stop suggesting you. A tool that appears for the wrong job loses trust faster than a tool that stays quiet.
Run the twenty prompts in fresh chats. Note which ones fail. Change one field at a time, then rerun the failed prompts. Keep a note of the prompt, the field, the result, and the date. Do this before you ask for a single signup. If the tool appears for the direct and indirect prompts and stays out of the negative prompt, you have a job that can be suggested.
Then send the link to the first person who asked. If they do not understand the job in one sentence, fix that sentence first, then ask again.
