TL;DR
The text overlay on your screenshots — not the UI — is what a browser reads first. The first 3 portrait frames show up directly in App Store search results, and Apple reports that about 70% of visitors never scroll past that first impression. Your job is short, character-counted headlines (≤25 chars) and sub-lines (≤35 chars) that each promise an outcome. AI is fast at producing tight, on-brand variants per frame; it is bad at guessing what your users actually value, so feed it 2-3 real review quotes and never let it invent a benefit. Copy the prompt below, then verify every character count by hand.
Why the first three frames decide everything
When someone searches the App Store, the result row shows your first 3 portrait screenshots (or 1 landscape) before they ever open your product page. Those frames carry most of the clicks to the detail page, and most users never reach frame 4. So the order is fixed:
- Frames 1-2: earn the scroll. Lead with the single outcome a stranger cares about.
- Frame 3: still visible in search; reinforce or add a second angle.
- Last 1-2 frames: ask for the install — action copy, not feature copy.
Generic overlays (“Easy to use,” “Powerful,” “Beautifully designed”) waste your most valuable real estate. Every headline should name a result the user reaches, in words a person would actually say.
What AI is good and bad at here
AI is excellent at producing tight, character-counted variants for each frame, varying the angle (status, savings, transformation, speed) while keeping one voice across the set. It is poor at knowing your real user’s pain. Feed it 2-3 actual review quotes that describe the outcome users praise, and the headlines stop sounding generic. Do not let the model invent the benefit or a feature claim your UI does not back up.
What to feed the AI
- Per-frame list: subject of the screenshot + the one message it carries
- App category and the primary outcome users seek
- 2-3 real review quotes describing what users love (paste verbatim)
- Brand voice in one sentence
- Visual context: light or dark background, typography style
- Character caps: 25 for the headline, 35 for the sub-line (adjust to your design)
Copy-ready prompt
Write App Store screenshot copy.
App and category: [line]
Primary outcome users seek: [line]
Real review quotes: [2-3 quotes]
Brand voice: [one sentence]
Visual context: [dark/light, type style]
Caps: headline <=25 chars, sub-line <=35 chars
Screenshots (subject + intended message):
1. [screenshot] - [message]
2. ...
For each screenshot return:
- Headline (<=25 chars, show the character count)
- Sub-line (<=35 chars, show the character count)
- One alternate headline (a different angle)
- Carousel slot: first impression / mid-scroll / conversion
Constraints:
- Ban "Easy to use," "Powerful," "Beautifully designed"
- Each headline names one outcome the user reaches
- Frames 1-2 earn the scroll; the last 1-2 ask for the install
- Write in words a real user would say, not marketing adjectivesFor localized stores, add: Now adapt for [locale] — same outcome, idiomatic phrasing, recount the characters in the target script. Chinese, Japanese, and Korean run wider per glyph, so the count that fit in English will break the layout.
GPT-5.5 (ChatGPT) and Claude Sonnet 4.6 both handle this well; Claude tends to hold a voice across a set more consistently, GPT-5.5 is quicker at brute-forcing many short variants. Either is fine — the input quality matters far more than the model.
Ask for a table
Have the model return one row per frame so you can drop it straight into your design tool:
| Frame | Subject | Headline (chars) | Sub-line (chars) | Alternate | Slot |
|---|---|---|---|---|---|
| 1 | Hero feature | ”Save an hour a day” (18) | “Plan your week in 3 taps” (25) | “Get your time back” (18) | First impression |
| 2 | Core flow | ”No more lost notes” (18) | “Everything in one place” (24) | “Find anything fast” (18) | First impression |
| 3 | Proof / social | ”Rated 4.8 by 50k users” (23) | “Trusted by busy teams” (21) | “Loved by 50k people” (19) | Mid-scroll |
| 6 | Conversion | ”Start free today” (16) | “No card needed to try” (21) | “Try it in 60 seconds” (20) | Conversion |
The slot is the point. First-impression copy and conversion copy are different jobs — never reuse the same line.
Store specs to design against (June 2026)
| Item | App Store (iOS) | Google Play |
|---|---|---|
| Screenshots allowed | 1-10 per device class | 2-8 phone screenshots |
| Shown in search before scroll | First 3 portrait (or 1 landscape) | Listing preview varies |
| Main iPhone size | 1320 x 2868 px (6.9”, iPhone 17 Pro Max) | 1080 x 1920 recommended |
| Format | PNG or JPEG, RGB, no alpha | JPEG or 24-bit PNG, no alpha; max 8 MB |
| A/B testing | Product Page Optimization: up to 3 treatments, 90-day max | Store Listing Experiments (free, native) |
App Store Connect scales one iPhone set at the largest size down to every smaller iPhone shelf, so you upload one set, not one per device. Submissions even 1 pixel off spec get rejected.
Check the output before you ship it
- Every headline counts to ≤25 chars and every sub-line to ≤35 (verify yourself; models miscount)
- The first frame promises an outcome strangers care about, readable without tap-to-zoom
- No generic adjectives slipped through
- Voice is consistent across the whole set
- The last frame is action-oriented, not feature-oriented
- Localized versions were re-counted in the target script
Common mistakes
- Generic adjective copy. “Easy. Powerful. Smart.” converts no one.
- Same line on frame 1 and the last frame. The carousel job changes per slot.
- Headlines that need tap-to-zoom to read. Browsers do not zoom.
- Letting AI invent feature claims. Check every line against your real UI.
- Forgetting localization character counts. CJK scripts break an English-sized layout.
- Trusting the model’s stated character count. It is often wrong by a few — count the final line yourself.
FAQ
- How many screenshots should I use? Apple allows up to 10 per device class; 6-8 is the conversion sweet spot. Google Play requires at least 2 phone screenshots and allows up to 8.
- Why do only the first three matter so much? On iOS the first 3 portrait screenshots appear in search results before anyone opens your page, and Apple reports roughly 70% of visitors never scroll past that first impression.
- Should the headline match the in-app UI? No. The headline sells the outcome; the UI shows the feature. They are different jobs.
- Can I A/B test screenshots? Yes. App Store Connect Product Page Optimization runs up to 3 treatments for up to 90 days; Google Play Store Listing Experiments do the same natively for free. Test the first frame’s headline first.
- Will AI count characters correctly? Not reliably. Always recount the final line yourself, especially for non-Latin scripts where each glyph is wider in layout.
Related
- App store listing prompts — title and description
- App onboarding copy AI — first-run copy users see after install
- App store screenshot copy prompts — additional copy variants
- App store review response prompts — replying to store reviews
- Product launch copy — broader launch copy
- Launch checklist — verify screenshots before publishing
External references: Apple — Product Page Optimization and Apple — Screenshot specifications.