# The AI feature only feels smart after the first useful minute > Why setup wizards, starter prompts, model routing, async jobs, ratings loops, and segment benchmarks decide whether an AI feature earns activation or just curiosity. - Canonical HTML: https://growth.iangoh.com/blog/the-ai-feature-only-feels-smart-after-the-first-useful-minute/ - Published: 2026-05-26 - Updated: 2026-05-26T15:40:00Z - Categories: ai products, activation, product UX - Niches: AI products, SaaS, developer tools, support software, creator tools ## On this page - Setup has to feel smaller than the question - A blank AI box is heavier than most teams realize - Speed often gets mistaken for intelligence because users have so little else to judge - The rating button matters because it tells you what broke the second chance - Usage is not proof if the wrong segment is doing all the playing - Where this cluster is most useful ## Start with these related tactics - [AI install wizard for 90-second setup](/growth-ideas/ai-install-wizard-for-90-second-setup/): Turn AI setup into a short guided wizard so a user can get from curiosity to first answer in about 90 seconds instead of ten wandering minutes. - [Suggested prompts in the empty AI state](/growth-ideas/suggested-prompts-in-empty-ai-state/): Seed the blank AI box with realistic starter prompts so users can borrow a good first move instead of freezing at an empty input. - [Task-based model routing for AI speed](/growth-ideas/task-based-model-routing-for-ai-speed/): Route lightweight jobs to smaller fast models and reserve larger models for harder reasoning so the product feels quick without giving up depth where it matters. A lot of AI product teams think they are selling the answer. Usually they are selling the first useful minute before the answer. If that minute is slow, blank, or confusing, the model never gets a fair trial. That is why I like this PostHog cluster. It is practical in the unfashionable way. The lessons are not about sounding magical. They are about getting a real user from curiosity to one clean win fast enough that the product can earn a second try. ## Setup has to feel smaller than the question The clearest move is [AI install wizard for 90-second setup](/growth-ideas/ai-install-wizard-for-90-second-setup/). If a user came in with one product question, they should not have to fight a ten-minute configuration ritual before they learn whether your assistant is useful. This fits well beside [layered context injection for AI answers](/growth-ideas/layered-context-injection-for-ai-answers/). The product has to do more of the setup work up front if it wants the user to ask a smaller, more natural question. ## A blank AI box is heavier than most teams realize The next fix is [suggested prompts in the empty AI state](/growth-ideas/suggested-prompts-in-empty-ai-state/). An empty box quietly asks the user to invent the use case, the syntax, and the confidence all at once. Starter prompts are useful because they turn the first session into choosing, not inventing. That is a much better bargain for anxious first-time users, especially in SaaS, support software, and creator tools where the user already has a real job in mind. ## Speed often gets mistaken for intelligence because users have so little else to judge That is where [task-based model routing for AI speed](/growth-ideas/task-based-model-routing-for-ai-speed/) earns its keep. A simple question should not feel like a board meeting between four models. For heavier work, [async AI workflows with cached retries](/growth-ideas/async-ai-workflows-with-cached-retries/) is the honest companion. If a job is genuinely long, say so, keep the user posted, and avoid making them pay the latency cost twice when a retry happens. This also reinforces [uncertainty, source, and progress cues in AI UI](/growth-ideas/uncertainty-source-and-progress-cues-in-ai-ui/). The product feels smarter when it explains what kind of work it is doing instead of pretending every request should look instant. ## The rating button matters because it tells you what broke the second chance I would not ship an AI surface without [AI response rating with follow-up context](/growth-ideas/ai-response-rating-with-follow-up-context/). The first bad answer does not only create disappointment. It tells you where trust will disappear if nothing changes. A thumbs-down with one short follow-up field gives the team something much better than vague AI skepticism. It gives a concrete miss to inspect, replay, and fix. ## Usage is not proof if the wrong segment is doing all the playing The hardest tactic here is probably [AI vs non-AI activation benchmark](/growth-ideas/ai-vs-non-ai-activation-benchmark/). Raw usage can flatter a team into believing the assistant is working when it is mostly attracting tourists. For AI products and developer tools, I would compare the AI path against the normal activation path and split it by user segment early. If your real buyers are not the ones getting the value, the roadmap problem is bigger than prompt quality. ## Where this cluster is most useful This batch is most useful for AI products, SaaS copilots, support software, and creator tools where the first session has to prove value before the user bounces back to ChatGPT or the old manual workflow. The lesson is simple. The model can be strong and the feature can still lose if the first minute asks too much. If you want help turning that first useful minute into a better activation system, [Ian Goh advisory](https://iangoh.com/advisory) is the clearest next step. ## Related GrowthDex tactics - [AI install wizard for 90-second setup](/growth-ideas/ai-install-wizard-for-90-second-setup/) - Onboarding, Activation, Product - [Suggested prompts in the empty AI state](/growth-ideas/suggested-prompts-in-empty-ai-state/) - Onboarding, Activation, Retention - [Task-based model routing for AI speed](/growth-ideas/task-based-model-routing-for-ai-speed/) - Product, Activation, Retention - [Async AI workflows with cached retries](/growth-ideas/async-ai-workflows-with-cached-retries/) - Product, Retention, Operations - [AI response rating with follow-up context](/growth-ideas/ai-response-rating-with-follow-up-context/) - Feedback, Retention, Product - [AI vs non-AI activation benchmark](/growth-ideas/ai-vs-non-ai-activation-benchmark/) - Analytics, Activation, Positioning ## Essay chronology - [Newer essay: The request count is usually the wrong number](/blog/the-request-count-is-usually-the-wrong-number/) - support-led growth, product signal, brand trust - [Older essay: The first customers should leave tracks for the next ones](/blog/the-first-customers-should-leave-tracks-for-the-next-ones/) - early-stage growth, founder-led sales, brand trust ## Keep reading - [The help page starts earning when it can finish the job](/blog/the-help-page-starts-earning-when-it-can-finish-the-job/) - support-led growth, seo, activation - [The Product Hunt launch should stay usable after the spike](/blog/the-product-hunt-launch-should-stay-usable-after-the-spike/) - community-led growth, activation, SEO - [The integration should feel like your product, not a detour](/blog/the-integration-should-feel-like-your-product-not-a-detour/) - product-led growth, activation, technical SEO ## Continue through the blog - [SaaS](/blog/#path-saas) - 3 essays in this path - [AI products](/blog/#path-ai-products) - 3 essays in this path - [developer tools](/blog/#path-developer-tools) - 3 essays in this path ## Sources - [PostHog Newsletter: What we've learned about building AI-powered features](https://newsletter.posthog.com/p/what-weve-learned-about-building) · [GrowthDex source hub](/sources/posthog-newsletter-what-we-ve-learned-about-building-ai-powered-features/) - [PostHog Newsletter: What we wish we knew before building AI products](https://newsletter.posthog.com/p/what-we-wish-we-knew-before-building) · [GrowthDex source hub](/sources/posthog-newsletter-what-we-wish-we-knew-before-building-ai-products-news/) ## Editing notes - Kept the essay on one concrete test about the first useful minute instead of drifting into generic claims about AI transformation. - Used plain objects like setup rituals, blank boxes, retries, and rating buttons so the argument stays tied to work teams actually ship. - Let the PostHog operating details carry the proof rather than padding the piece with abstract language about intelligence or delight. - Ended on a segmentation check that gives the reader a hard diagnostic instead of a soft future-looking conclusion. ## Advisory If you want help turning this into a growth system, Ian Goh offers advisory at https://iangoh.com/advisory.