The AI Product Cold Start Playbook: Week 1 to Week 4
A dated four-week cold start plan for a new AI product: manual outreach, one community, directory submissions, and what to measure at each step.
Published 2026-11-09 · 5 min read
This is a four-week cold start plan for an AI product with no audience and no budget. It assumes the product works and that you know roughly who it is for. If either is untrue, spend the four weeks on that instead — no distribution plan compensates for an unclear product.
The sequence matters more than the effort. Each week builds the thing the next week needs: week one produces conversations, week two produces a reputation somewhere, week three produces durable pages, and week four produces the data to decide what to repeat.
Week 1 — twenty conversations
The goal this week is not signups. It is learning what people actually call the problem you solve, because you will need their words for every subsequent step.
- 1
Days 1-2: write down 25 people who have the problem
Not people who might like the product — people who have demonstrably hit the problem. Look for public complaints, questions in forums, people who built a workaround. Write down where you found each one, because that tells you where more of them are.
- 2
Days 2-4: message ten of them individually
Two sentences. Name the specific thing they posted that made you think of them. Ask whether the problem is real for them, not whether they would try your product. A question about their workflow gets answered; a request to try something gets ignored.
- 3
Days 4-7: onboard three of them by hand
Create the account, import their data, run the first task with them on a call. This is slow and it is the point — five onboarding calls teach you more than five hundred signups, because you see where they hesitate.
- Write down every phrase they use. Their words become your one-liner, your description, your tags and your first posts.
- Note where they got stuck. That is your activation problem, and fixing it matters more than any channel.
- Ask each one who else has this problem. Two weeks of this produces a better prospect list than any tool.
Week 2 — one community, done properly
Pick a single community where your buyers already gather. Spend the week being useful in it before mentioning your product at all.
- Answer questions you can actually answer. Ten substantive replies in your niche do more for you than one announcement, because replies are how a handle becomes recognisable before it asks for anything.
- Read the rules and the last month of posts. Every community has norms, and the fastest way to be removed is to arrive posting a link. Some have a designated showcase thread or channel — use it.
- Share what you learned in week one. A post about what surprised you when you talked to ten people about the problem is genuinely useful and mentions your product only in passing. That is the format that works.
Week 3 — prepare the asset pack, then submit
Directory and platform submissions are the cheapest durable surface available to a new product, and they are mostly a preparation problem. Build the pack once and reuse it everywhere.
| Asset | Spec | Common mistake |
|---|---|---|
| Logo | Square, transparent background, legible at 32px | A white box around the logo — the most frequent asset rejection |
| One-liner | Under 60 characters, says what it does for whom | Abstract: "AI-powered workflows for teams" |
| Short description | 120-160 characters | Copied from the homepage hero, so it includes slogans |
| Long description | 400-800 characters, first sentence defines the product | Leads with a claim nobody can verify |
| Screenshots | Three, showing the product working with realistic data | An empty state, a login screen, or a marketing graphic |
| Demo video | Under 60 seconds, one complete task, captioned | Two minutes of narration before anything happens |
Then submit to two or three places, spaced out. Not ten in an afternoon — five near-identical announcements in one day produce no compounding and make the campaign visible. One or two a week lets you improve the description based on what the first ones taught you.
Sort candidates by relevance rather than size. A small directory read by your exact buyers beats a large one read by everyone, and the smaller one is often easier to be visible in. Check whether the other listings look like your competitors or like noise.
Week 4 — measure and decide
By now you have conversations, a presence somewhere, and a few durable pages. The last week is about finding out which of it worked, before you commit to doing more of anything.
- Set up referrer tracking. Several platforms strip UTM parameters, so segment by referrer as the fallback. Without this you are choosing next month's channels based on which submission felt busiest.
- Identify the one activation step that matters. Not total signups — the action that indicates someone got value. Everything else is a proxy for it.
- Find the drop-off. The step where most people leave is worth more than any new channel, and you now have enough users to see it.
- Talk to everyone who signed up. While there are only a few dozen this is possible, and it is the cheapest research you will ever do.
One channel that produced ten users is worth more than four channels that produced two each. The first is a place to invest; the second is a list of things to stop doing.
What to do in month two
Repeat the week that worked, and drop the rest. If week one produced the best conversations, do more of them. If a directory sent real traffic, find five more like it. If a community ignored you, leave it — attention is not distributed evenly and you do not have to argue with that.
The part most founders skip is that a submission keeps working after the announcement. A page describing your product keeps being read by people searching for a tool like yours months later, which is why the persistent surfaces deserve more maintenance than the one-day ones. Update the description and screenshots when the product changes; a stale listing reads as an abandoned product.
More in Cold start
- The 10 Mistakes That Kill an AI Product LaunchA diagnostic list of the ten failures behind a flat AI product launch, ordered by how often they happen and how cheaply each one is fixed.
- What to Do in the 30 Days After Your AI Product LaunchThe month after a launch is when most founders stop and most of the value is created. A week-by-week plan for converting a spike into something durable.
- How to Cold Start an AI Tool in a Crowded CategoryWhen thousands of products claim the same category, generic positioning is unrecoverable. How to narrow until the competition stops being direct.