What We Learned Launching on an AI Directory

The listing

GlitchBot is now listed on Dang.ai, a directory of AI tools. Directory listings are a standard early-stage distribution move and it would be easy to write a paragraph about being honoured and leave it there. More useful is what the traffic actually taught us, because the answer was not what we expected.

Directory traffic behaves differently

Visitors arriving from a directory have a distinctive profile. They did not search for "Gantt chart software" — they were browsing a list of AI tools and clicked something that looked interesting. They arrive with low intent, high curiosity, and almost no context about what the product is for.

That combination is brutal on onboarding, and it is exactly why it is worth having. Search traffic is forgiving: someone who searched for an AI Gantt chart generator will tolerate a confusing first screen because they already know what they want. A directory visitor will not. They give you roughly fifteen seconds to make the value obvious, and if you fail, they are back on the list.

What broke

Two things, both embarrassing in hindsight.

The first was that our landing page assumed the visitor knew what a Gantt chart was and cared about having one. For someone browsing AI tools generally, that is a large assumption. The people who bounced fastest were not uninterested — they could not tell within a few seconds whether this was for them.

The second was that our first-run experience asked for a project description before showing anything. That is fine when someone has a project in mind. Directory visitors mostly do not; they want to see the thing work before investing effort. Requiring input before demonstrating output inverted the order those users needed.

What we changed

We added a set of pre-filled example projects on the landing page — pick "ship a mobile app" or "run a product launch" and watch the agents build the plan, with no typing and no account. That single change did more for directory conversion than any copy revision, because it moved the demonstration before the ask.

We also rewrote the opening line to lead with the outcome rather than the mechanism. What the product does is generate project plans; what it is for is not having to build one by hand. The second framing survives a reader who has never used a Gantt chart.

Whether directories are worth it

Honestly: the direct traffic is modest and does not compound. What made this one worth doing was the diagnostic value. A directory sends you a stream of users with no context and no patience, which is the harshest possible test of whether your first screen works. Ours did not, and we would probably not have found out as quickly any other way.

If you are considering directory listings for an early-stage product, that is the framing we would suggest: treat them as cheap usability testing that happens to also send traffic, and instrument the first sixty seconds carefully. You can see the result on our home page, and the reasoning behind the plan-generation flow in the AI Gantt chart generator guide.

Instrumenting the first sixty seconds

The change that made this listing useful was deciding, before the traffic arrived, exactly what we wanted to learn. Without that, a traffic spike produces a satisfying graph and no information.

We tracked four events: landed, watched an example plan generate, started their own plan, and edited a generated plan. Four events is deliberately few. The temptation is to instrument everything, which produces a funnel with fourteen steps that nobody analyses. Four steps can be read at a glance and each one maps to a decision we could actually make.

The drop-off was concentrated between the first and second events, which is what pointed at the landing page rather than the product. Had we only tracked sign-ups, we would have concluded the product was the problem and spent a month on the wrong thing.

Why low-intent traffic is the useful kind

There is a counterintuitive argument for seeking out audiences who are not looking for you. High-intent traffic — someone actively searching for AI Gantt chart software — masks onboarding problems, because those users will push through friction to get the thing they came for. They convert despite your first screen, not because of it.

Low-intent traffic has no such tolerance. Every point of friction converts directly into a bounce. That makes it a much more sensitive instrument: problems that high-intent users absorb silently show up immediately and unambiguously in a directory cohort.

The practical implication is that early-stage teams should probably seek out one source of low-intent traffic deliberately, and read it as a diagnostic rather than a channel. The traffic itself is close to worthless. The signal is not.

What we still have not solved

In the interest of not writing a post where everything worked: the fix improved the first session and did essentially nothing for week-two retention. Users who generate a plan, edit it, and find it useful still frequently do not come back, because a single project plan is not something you need daily.

That is a real product question and not an onboarding one, and we do not have a confident answer yet. The current hypothesis is that the recurring use case is agents doing work against the plan rather than the plan itself — which is where our agents work is pointed — but that is a hypothesis, and directory traffic will not tell us whether it is right.