The Algorithm Now Reads Behavior, Not Just Keywords
The App Store and Google Play ranking systems have fundamentally shifted how they evaluate app quality. Where exact keyword placement in title fields once dominated, both platforms now incorporate wiki:retention-rate and session frequency as primary signals. Apps with strong Day 1 and Day 7 retention climb in visibility even when metadata remains unchanged. Conversely, a spike in downloads followed by rapid uninstalls triggers algorithmic penalties that no amount of keyword optimization can overcome.
This shift reflects a move from input-based ranking (what developers write) to outcome-based ranking (what users actually do after install). The stores are measuring whether the promise made on the product page matches the experience delivered inside the app. When they diverge, the algorithm notices through user behavior before the team does through analytics.
Custom Product Pages and Screenshot Text Are Now Indexed
Two technical changes in 2025 expanded the keyword surface area available to developers. First, Apple began indexing wiki:custom-product-pages organically in July 2025, raising the limit from 35 to 70 pages per app. Each CPP can target distinct keyword clusters and appear in organic search, allowing a single app to present different value propositions to different audience segments. A fitness app can now have one page optimized for "weight loss workouts" and another for "yoga for flexibility" without diluting either message.
Second, Apple started extracting visible text from screenshot captions via OCR or metadata parsing around mid-2025. Practitioners observed apps ranking for terms that appeared only in screenshot overlays, not in title, subtitle, or keyword fields. The prominent caption "Track Your Sleep Patterns" began surfacing apps in searches for "track sleep" and "sleep patterns" even when those phrases were absent from traditional metadata. This effectively adds hundreds of characters of indexable content beyond the 160-character limit of title, subtitle, and keyword field combined.
The screenshot text that gets indexed: large, readable captions positioned outside the device mockup. Small in-app UI text, decorative fonts, and fine print do not register. One keyword theme per screenshot works best — diluting a caption across multiple unrelated terms reduces both conversion and ranking signal.
Ratings Operate on a Binary Scale, Not a Gradient
Developers report that Apple's editorial selection and ranking algorithm treat the five-star rating system as effectively binary. A rating of 4.0 or below reduces visibility; 4.5 and above signals quality. The problem: users perceive the scale as a true gradient where three stars means "met expectations" and four stars means "good." In practice, any rating below five actively harms the app's standing. A 4.1-star app loses ground when users leave four-star reviews intending them as positive.
This creates a conflict. Users resist interruptions for rating prompts, but developers have no choice — editorial featuring and algorithmic boost depend on achieving a critical mass of five-star reviews. The alternative is obscurity. Some practitioners argue Apple should switch to a thumbs-up/thumbs-down system to align user intent with algorithmic interpretation. Until then, the star system functions as a de facto like/dislike vote, and anything less than five is a vote against.
Google Play Prioritizes Short Description Over Title
Machine learning analysis of 512 Google Play metadata iterations found that adding or moving a keyword into the Short Description field correlated with ranking improvements in 84.2% of cases — 46.5 percentage points above baseline. By contrast, placing a keyword only in the Title improved positions in just 15.8% of iterations, 21.9 points below baseline.
This inverts conventional ASO wisdom. The Short Description is visible, user-facing, and limited to 80 characters. The Full Description, while indexed, showed weaker impact: 40.5% improvement rate when keywords appeared there alone. Notably, having duplicate mentions of a keyword in the Full Description before metadata changes increased success rates to 54.5%, suggesting that prior semantic relevance helps new keyword placements gain traction.
The data also showed that results from metadata updates appear faster than widely assumed. On iOS, median rank shifts occurred the day after an update. On Google Play, shifts appeared by day three. Waiting two weeks to assess iteration outcomes delays learning cycles unnecessarily in most cases.
Partial and Soft Keyword Matches Outperform Exact Matches
In the iOS sample, iterations using partial or soft keyword coverage (e.g., "strategy" to match "strategy game," or "tactical game" as a semantic proxy) achieved approximately 60% improvement rates with a median gain of six positions. Exact matches did not consistently outperform partial coverage across rank buckets. Apps ranking between positions 11–20 saw partial matches work better than exact matches, likely due to how the algorithm interprets competitive context in that segment.
This reflects the algorithm's improved semantic understanding. Lemmatization — reducing words to root forms — has always been standard in search systems, but Apple and Google now parse intent and related terms rather than requiring literal string matches. "Yoga for relaxation" can surface an app optimized for "meditation and calm" if the surrounding metadata and user behavior signals align.
The practical takeaway: covering a keyword partially across multiple fields (Title + Subtitle, or Short Description + Full Description) often works as well or better than forcing an exact phrase into a single field. Splitting "meditation app" into "meditation" in one field and "mindfulness app" in another can capture both terms and related searches.
Field Combinations That Correlate With Ranking Gains
On iOS, the strongest pattern for new keywords was simultaneous addition across Title, Subtitle, and Keywords field: 76.3% of such iterations improved position, with a median lift of 30 ranks. When a keyword moved from Title-only to Title + Subtitle (splitting it across both), 80% of cases saw improvement.
Negative patterns also emerged. Moving a keyword from Subtitle + Keywords into Title + Keywords dropped the success rate to 33.3%, well below baseline. The exact mechanics behind these differences remain unclear without a larger dataset segmented by category and query type, but the signal is consistent: expanding keyword presence across multiple fields tends to work better than concentrating it in one, and certain field combinations perform better than others.
One hypothesis: the algorithm interprets field distribution as a signal of thematic coherence. A keyword present in title, subtitle, and hidden field suggests the app genuinely centers on that concept. A keyword jammed into the title alone may read as opportunistic keyword insertion.
Engagement Metrics Are Now Ranking Factors
Both platforms explicitly or implicitly factor post-install behavior into wiki:search-visibility. Session length, session frequency, and uninstall rate now influence where an app appears in search results. An app with high download velocity but poor Day 7 retention will see its rankings erode. An app with steady, moderate installs and strong engagement will climb over time.
This is the most consequential shift in ASO since app stores launched. Metadata optimization can no longer compensate for a product that fails to retain users. The algorithm effectively outsources quality assessment to user behavior, treating retention as a proxy for product-market fit. Developers who focus only on conversion rate optimization cro without addressing onboarding, core loop, and retention will see diminishing returns from ASO work.
Update frequency also matters, but quality trumps cadence. Apps in the top 1,000 update at least once per month, signaling active development. Stagnant apps lose ground as the algorithm interprets lack of updates as abandonment. However, shipping a meaningful update every four to six weeks outperforms weekly cosmetic changes.
Practical Implications
Metadata no longer operates in isolation. The page must align with the product experience, or behavioral signals will override keyword optimization. Teams should:
- Use all 70 Custom Product Pages to target segmented keyword clusters and audience intents that cannot coexist on a single default page
- Write screenshot captions as keyword-rich benefit statements (3–8 words, high contrast, readable at thumbnail size) to gain indexable text beyond the 160-character metadata limit
- On Google Play, prioritize Short Description for primary keywords; Full Description for context and semantic reinforcement
- On iOS, distribute important keywords across Title, Subtitle, and Keywords field rather than concentrating them in one
- Track engagement and retention metrics alongside keyword rankings; a ranking gain with poor retention will reverse itself
- Understand that four-star reviews function as negative signals in Apple's system, not positive ones
- Update metadata and visual assets in 4–6 week cycles rather than waiting for quarterly reviews; ranking impacts appear within days, not weeks