Two Launches in a Week: What Actually Moved the Needle

Two listings, one week

GlitchBot went live on Startuups.com and Startup Fast within a few days of each other. Because the timing was close and the audiences overlap, it became a reasonable natural experiment in what these listings actually do — and we tracked it properly rather than eyeballing the analytics.

The numbers, honestly

Both produced a traffic spike lasting roughly 48 hours, followed by a return to baseline. Neither produced sustained referral traffic. Anyone claiming directory listings are a growth channel in isolation is, on our evidence, wrong.

What they did produce was a concentrated burst of first-time users over a short window, which is genuinely valuable but for a different reason than the one usually advertised. A hundred unfamiliar users in two days surfaces onboarding problems that a hundred users over two months will not, because you can watch the whole cohort hit the same wall at the same time.

What the cohort showed us

Two things, both concerning the first ninety seconds.

First, a meaningful share of users generated exactly one plan and never returned. Not a failure of the generator — the plans were fine — but a failure to establish what to do next. Producing a Gantt chart is impressive once; it is not by itself a reason to come back tomorrow.

Second, users who edited a generated plan were dramatically more likely to return than users who only viewed one. Obvious in retrospect, and it reframed the onboarding goal: the job of the first session is not to demonstrate generation, it is to get the user to change something. Ownership starts at the first edit.

What we changed

We stopped optimising the generation step, which was already good, and started optimising the first edit. The generated plan now opens in an editable state with an obvious affordance to drag a task, rather than presenting as a finished artefact to admire. Small change, and it moved second-session return more than anything else we tried that quarter.

We also added the pre-filled example projects on the home page, so the demonstration happens before the sign-up rather than after it.

Would we do it again

Yes, with adjusted expectations. Not as a growth channel — the traffic does not compound and the referral tail is negligible. As a way to compress a month of onboarding feedback into a week, they are cheap and effective, and we would recommend timing them to land right after you have shipped something you want tested rather than whenever the form gets filled in.

The broader point: early distribution is mostly a research instrument. Treating it as a research instrument makes you instrument it properly, which is where the actual return is.

Why the first edit matters more than the first result

The finding that a user's first edit predicts return is worth unpacking, because it generalises well beyond our product.

A generated artefact — a plan, a draft, an image — arrives as someone else's work. It can be impressive without being yours. Viewing it is a passive act, and passive impressions decay fast. The moment a user drags a task, renames a phase or changes a date, the artefact becomes partly theirs, and the psychology shifts from evaluating a demo to maintaining a thing they own.

This is a known effect in other contexts and we had not connected it to onboarding. For any product where AI produces a first draft, the design implication is direct: the goal of the first session is not the best possible generated output. It is the smallest possible distance to the user's first modification.

Those goals can actively conflict. A more polished generated plan is harder to start editing, because it looks finished and correcting it feels like vandalism. We now deliberately leave generated plans looking editable rather than final.

Measuring a two-day spike properly

A methodological note, since spike measurement is where most launch write-ups go wrong.

Comparing spike-week numbers to the previous week tells you almost nothing — the cohorts are different in kind, not just size. What we did instead was tag both cohorts and follow them for four weeks, comparing retention curves rather than volumes.

That produced the useful finding: directory users retained substantially worse than search users at week one, and roughly the same at week four conditional on having made an edit. In other words the population was not lower quality, it was more thinly filtered — the same proportion of genuinely interested users, wrapped in a much larger number of browsers. If we had looked only at aggregate retention, we would have concluded directory traffic was worthless and stopped.

What we would tell someone planning a launch week

• Decide what you want to learn first. A spike with no hypothesis produces a graph and nothing else.
• Instrument four events, not fourteen. You will actually read four.
• Tag the cohort and follow it for a month. Week-one numbers from a low-intent cohort are misleading in both directions.
• Time the launch after a change you want tested, not whenever the submission form gets completed.
• Expect no lasting referral traffic. If you get some, treat it as a bonus rather than the plan.

None of this makes directory listings a growth strategy. It makes them a cheap, fast source of the kind of feedback that is otherwise slow and expensive to obtain, which for an early product is arguably more valuable than the traffic would have been.