For ten years, the playbook for getting an app discovered was simple to draw, even if it was hard to execute: pick your keywords, win store search, defend your rank, pour paid UA on top.
However, discovery now forms somewhere else: someone sees a clip on TikTok, asks a friend, reads a Reddit thread, types “best app to fix my sleep” into ChatGPT, then searches your name in the store. By the time they reach your product page, the decision is mostly made. Your App Store page is the bottom of the funnel.
This is the new reality of app discovery: ASO (App Store Optimization), SEO, and GEO (Generative Engine Optimization) are three surfaces of one job.
Below, we’re looking into them, and see where app discovery now happens and what changed on each:
- The store, where search went semantic and the App Store quietly moved onto the open web.
- The web, where the cost of publishing hit zero and volume stopped signaling anything.
- AI answers, a new shelf that’s mostly empty.
Three surfaces, one shift
1. Inside the store, search went semantic, while the store left the store
Two things happened at once, and most teams only noticed the first.
First: Apple’s search algorithm shifted from exact-keyword matching toward semantic, intent-based matching. Stuffing your subtitle and keyword field still works, but it’s the most competitive it’s ever been. Apple states that search results may include app tags, “generated using large language models (LLMs) based on the metadata you’ve provided in App Store Connect”. There is now a model sitting between your metadata and what the store decides your app is.
Second, and bigger: the App Store is on the web now. Product pages, editorial stories, in-app events, charts… all live, structured, crawlable URLs. Which means your product page is no longer only a store asset. It’s a web page competing in Google, and a document an AI assistant can read and cite when someone asks which app to use. You are ranking on the open web whether you meant to or not.
That quietly rewrites the metadata rulebook. Your long description, for a decade the field ASO people wrote for conversion and nobody optimized for ranking, is now the richest block of text a crawler or a model has to understand what your app actually does.
Same fields for two audiences, with two sets of rules, running simultaneously, and almost nobody is writing for both.
We pulled apart the web App Store in detail here few months ago. Check it out!
2. On the web, the bar is rising
The cost of producing a competent-looking blog post has gone to roughly zero. Anyone can ship and everyone is, which means the old “More posts, more keywords, more rankings” is over. It worked when publishing was expensive and volume was itself a costly signal of effort. But it isn’t anymore, so it signals nothing.
When supply of adequate content is infinite, the only currencies left are the ones a model can’t manufacture: original data, a named human with credentials behind the claim, a genuine point of view, something that earns a citation instead of assembling one.
The machinery underneath rewards exactly that. We know this less from Google than from a courtroom: the DOJ antitrust trial (2023–2024) surfaced internal systems like QBST (Query-Based Salient Terms). It’s a “memorization system” that learns which terms a relevant page should contain for a given query. Search “best running shoes” and it expects “cushioning” “stability” “mileage”. Stuffing the exact phrase ten times over doesn’t move it anymore. This algorithm matches that vocabulary against your page; its sibling Navboost weighs the same against real click data.
The headline for app teams: a blog is now an entity-authority engine, and it has rules most apps ignore. They start below the content: even a post that earns a citation never gets one if the page it sits on is unreadable to a crawler. Most of those rules are about engineering and technical SEO, which is exactly why product teams skip them. The SEO basics live in the code (and you can run free audits with the demo of Screaming Frog): a unique, intent-led title tag and a single clear H1 per page; a real heading hierarchy (H1 > H2 > H3) that maps the structure of the content; images and video that are compressed, lazy-loaded, and carry actual alt text and transcripts; and a Search Console property that’s connected, so you measure what you rank for. The classic own-goal is keyword cannibalization: 3 thin pages all chasing “calorie tracker” split the signal so none of them ranks, whereas one strong page would win (we suggest you use Semrush Cannibalization Report tool to discover them). Unglamorous, mostly invisible in a demo, but those trace the line between a page a model can read and one it skips.
3. In AI answers, a whole new shelf exists and it’s mostly empty
When someone asks ChatGPT or Perplexity for “the best app for X”, your app is either in that answer or it isn’t. The signals that earn you a mention are your presence on the sources AI actually cites:
- Community platforms like Reddit and Quora are essentials here.
- Freshness also matters, and recently updated content gets cited at multiples of stale content.
- You have to be a recognized entity, consistently described across the web, before a model will confidently recommend you.
The base layer is a Wikidata entry: the open, structured knowledge base that both trains the models and feeds Google’s Knowledge Graph. So a clean, consistent entry is you describing yourself to the machines in their own format. On top of that, a Knowledge Graph entity Google actually resolves your brand to. Leave them blank and every model is guessing who you are from scattered third-party mentions.
To clarify, every AI answer engine reaches your site through a named crawler: GPTBot, ClaudeBot, PerplexityBot, Applebot… If your robots.txt disallows them, you are absent from them, by your own instruction. Most of the robots.txt files we find on app sites are copy-pasted defensive templates from an industry with the opposite business model, like publisher sites.
The stack, by surface
The store (ASO): Your existing stack still holds. What changes is what you’re writing for: the keyword field feeds a semantic layer now, and your long description feeds the open web.
Read more about The Best ASO Tools for Indie Developers here.
The web (SEO): Search Console, Screaming Frog for crawl and indexation, Semrush
The crawl layer: Before anything else, check if crawlers can reach you with LLM Pulse’s robots.txt checker and tells you which of GPTBot, ClaudeBot, PerplexityBot and Google-Extended your own file is blocking.
AI answers (GEO):
LLM Pulse is domain-first, tracks share of voice and citations across ChatGPT, Perplexity, Gemini and AI Overviews, and it now also tracks App Store and Play links specifically, which is the signal that matters most.
AppTweak’s AI Visibility for Apps is app-first and starts from the intents in your app category and measures whether AI assistants recommend your app, by name.
The surface marketing can’t fix
Discovery no longer stops at the store: Siri, Spotlight, and Apple Intelligence can surface an app’s actions. “Log my run”, “Start a 20-minute focus session”. The mechanism is App Intents, the framework that declares to the system what your app can do, in a form Apple’s models can read.
We’re not covering App Intents here; it’s real engineering scope and deserves its own piece. We’re flagging it because of who owns it as it lives in the codebase.
Discovery has quietly become a product problem too.
The connective insight: all three surfaces now reward one clear story about who you’re for and what you solve, expressed in language that machines interpreting intent can understand.
FAQ: ASO, SEO & GEO for app teams
What is GEO for apps?
Generative Engine Optimization for apps is the practice of getting your app named and recommended inside AI-generated answers such as ChatGPT, Perplexity, Google AI Overviews, rather than only ranking in App Store search. Unlike ASO, which optimizes store metadata, GEO optimizes the web content and entity signals a model reads before it decides which apps to recommend.
Does the App Store affect SEO?
Yes, and it now works both ways. App Store product pages, editorial stories and in-app events are live, crawlable URLs on the open web. Your product page can rank in Google and be cited by an AI assistant, which means your long description, historically written only for conversion, is now the richest text a crawler has to understand your app.
How do I get my app recommended by ChatGPT?
3 requirements:
- Be crawlable (check your robots.txt for GPTBot, ClaudeBot, PerplexityBot, Applebot).
- Be a recognizable entity (Wikidata entry, consistent description across the web).
- Be present on the sources AI actually cites, which in app categories means Reddit, comparison content, and recent, credible reviews.
Is ASO still worth it in 2026?
Yes. Store search still converts intent better than any other surface. But now, the decision is often made before a user ever opens the App Store.
