Your first 1,000 app downloads without ads: the brief you give the agent

Paid acquisition only works once you know what a paying user is worth, and you cannot know that before you have users. The first thousand downloads exist to teach you three numbers: what people search for when they find you, how many of them stay past day seven, and how many pay. Every one of those is cheaper to learn organically. That was true in 2023 and it is truer now, because the stores got better at rewarding apps that people actually keep.

What changed is who does the work. In 2023 you did the keyword research, wrote the listing, cut the videos and answered the reviews yourself, on evenings. In 2026 you brief an agent, review what it proposes, approve, and read one report on Monday. This article is written that way: each step ends with the brief you give the agent, whether that is AppTug or something you run yourself. The steps are the same. The hours are not.

Three store changes make organic more valuable than it was. Apple now generates App Store Tags from your metadata and uses them for browse placement, so a clear listing gets found in places a keyword never reached. Apple indexes the captions on your screenshots since mid 2025, so the first three screenshots are search text as well as conversion assets. And both stores now weigh retention and rating recency in ranking, which means an app that keeps users climbs without buying anything.

The old approach started from a keyword tool and a global volume number. The 2026 approach starts from three apps that rank where you want to rank, in each of your top five countries, and from the words your own beta users used. Search volume has no global value: flags quiz is 48,700 monthly searches in the US and Flaggenquiz is 19,200 in Germany, and neither is a translation of the other in difficulty or intent.

Google Play adds a wrinkle since September 2025. Guided Search splits broad queries into intents, so a search for fighting games shows sub categories before results. Broad head terms send less direct traffic than they did, and mid tail terms with a clear intent send more. The keyword plan should therefore lean toward specific phrases people would say out loud, which is also what Ask Play and Apple's natural language search reward.

The agent does the reading. Your job is to name the competitors, the countries, and the one sentence that describes what the app does for whom. Everything else is derived.

Every term goes in exactly once. Title first, because it weighs most. Subtitle next. Keyword field for the rest on the App Store, long description on Google Play. Count the characters: 30, 30 and 100 on the App Store. Then the part that did not exist two years ago: the captions on the first three screenshots are indexed, so they carry keywords too, in six words or fewer each, and the first three screenshots now appear directly in search results, so they have to make the promise before anyone taps.

App Store Tags are generated by Apple from your metadata. You cannot write them, you can only remove wrong ones. The practical consequence is that the description must state plainly what the app is and does, in the words a category browser would use, because that is what the tag model reads. Vague copy gets vague tags. Google's Ask Play reads the same description plus your website to answer questions like which app is best for learning capitals, so the description and the website must agree with each other and lead with the strongest claim.

Write each language from its own keyword list. A German listing built from German search terms outperforms a translated one every time, and since Google added CSV upload for localized metadata, shipping twelve locales is an afternoon for an agent, not a week.

Short form video is still the only free channel with real reach, and it has quietly become a search engine. TikTok search volume grew around 40 percent year over year, and people type best app for sleep sounds into TikTok before they type it into the App Store. That changes how a hook is written: the first line should contain the words people search, not only the words that stop a scroll.

Do not write from templates. Take the sentences your beta users already said and turn each into a hook, a script with timed beats, and a caption that carries the search phrase. One a day, on TikTok first, cross posted to Reels and Shorts. Google Play is opening its own short form surface, Play Shorts, in the US, so the same videos will soon work inside the store as well. Three of ten hooks will do nothing, six will do a little, one will bring more installs than the rest combined, and you will not know which in advance.

Reviews are search text on Google Play, conversion text on both stores, and since 2025 the raw material for Google's AI review summary that sits under your rating as a paragraph called users are saying. What people write about you now gets condensed into one paragraph that every visitor reads. Answering fast changes what gets written next, and answering the complaint that repeats fixes the summary at its source.

Apple weighs rating recency more than it did. A wave of good ratings this month counts more than a great average from last year, which makes the reply that turns a three star into a four star worth real ranking. Comments under your posts are the other half: they are the next hooks, and a reply within hours is what keeps a thread alive on TikTok.

After 14 days you have impressions and conversion per country. Change one thing, usually the title, and wait 14 more days. That is the only way to learn what worked. One title change on MapZap brought 2.1 times the impressions in 14 days with nothing else touched, and that number is only knowable because nothing else was touched.

App Store Connect added over a hundred metrics in March 2026, including cohort views and peer benchmarks, so the agent can now tell you whether your conversion is above or below the category median, not only whether it moved. Use that. A change that lifts you from below median to above it is worth more than one that moves an already good number.

Five numbers: impressions, downloads, trials, day 7 retention, rating trend. One line per number: up, down, why. One proposed change. One thing the agent needs from you. If the Monday report takes longer than two minutes to read, the brief was wrong, not the report. The first thousand downloads usually arrive between week six and week twelve on this loop, and they arrive with the three numbers you needed to decide whether paid is worth starting at all.