ASO and AI: A Framework for the AEO Era

When users stop typing keywords and start describing problems

Search volume on large language models has grown 5.5 times faster than Google’s over the past year, and roughly 20% of informational queries now happen inside chatbots instead of a search bar. For Gen Z specifically, that number climbs to 34%. None of this is a future trend to prepare for, and it is already reshaping how people find apps.

That was the starting point for our first Phield Notes session, a webinar series launched after the company’s Growth Phestival in Berlin brought together nearly 400 mobile growth practitioners across ASO, CRM, UA, creative, and AI. Our senior growth consultant Richard Velez opened the session with a reframe of app store optimization: traditional ASO has always been built around one question: what does the user type? That question assumes discovery starts and ends inside the App Store and Google Play, with an algorithm deciding what surfaces for a given keyword.

The problem is that keywords were never a great proxy for intent. Someone searching “spending tracker” might really be asking how to stop overspending when they go out, and a keyword match can miss that entirely. Users don’t always know what app they need. Sometimes they don’t know they need an app at all, just a solution to a problem. Keywords capture a narrow slice of that problem-solving intent, which is exactly the gap generative AI tools are stepping into.

This matters for anyone running ASO or mobile growth today, not because keyword-based optimization is dead, but because it is no longer the whole picture. The question worth asking now is not what the user types, but what the user actually needs, and the answer increasingly comes from a conversation with ChatGPT, Claude, Gemini, or Perplexity before anyone opens an app store at all.

The shift from keyword targeting to intent targeting

Google Play and the App Store are already adapting to this shift, which is a useful signal that it’s not a hypothetical. Google Play has started prioritizing intent-based metrics alongside install counts, and Ask Play, its conversational search feature, surfaces results based on what a user is trying to accomplish rather than a fixed set of keywords. Apple added natural language search as a core part of its own search experience, and app intents now help surface apps based on what they let a user do, not just what they’re called.

Outside the stores, discovery depends on a different mix of signals entirely: what the LLM infers about a user’s underlying need, public sentiment toward an app across reviews and blogs, whether it shows up in editorial content, and, somewhat surprisingly, how a prompt is phrased. We tested this last point directly. A query like “what are the best finance apps to track my spending” surfaced citations from blogs, forums, and community pages. A near-identical query, “what are the top finance apps to track my spending,” pulled in the same apps but added App Store and Google Play listings and user reviews to the mix. Small changes in wording changed which sources got cited.

To see how consistent this pattern was, the team ran more than 150 queries across 20 categories through ChatGPT, Claude, Gemini, and Perplexity. The pattern held everywhere: context and phrasing shape what gets referenced, and each model’s own data retrieval approach determines what gets recommended. That second part matters more than it sounds, because it changes how often you need to update your optimization work and how long it takes for a change to actually show up.

 

Why each AI model cites your app differently

Not every model decides what to recommend the same way, which means a single optimization strategy won’t perform evenly across all of them. Phiture’s research grouped the four major tools by how they retrieve and cite information.

Model	Primary retrieval mode	What that means for citations
ChatGPT	Memory and training	Rarely cites an explicit source unless web search is on; subscription tier affects citation behavior
Claude	Memory and training	Same pattern as ChatGPT: no explicit source unless web search is requested
Gemini	Mixed: memory and live web	Frequently references blogs and store copy; cites Google Play listings often
Perplexity	Live web	Quotes App Store and Google Play copy verbatim, along with editorial sources

The practical implication is about timing as much as content. A model that leans on memory and training only picks up your updated metadata once a new training window includes it, or once a user happens to trigger a live web search. If an edit to your description or your custom store listing misses that window, you’re waiting for the next one, or hoping the user’s search settings compensate for the gap. That’s a reason to treat metadata updates as an ongoing cadence rather than a one-time cleanup, something we’ll come back to in the measurement section.

 

Build intent clusters, not feature lists

This is where the framework gets actionable, starting with keyword strategy itself. The traditional approach targets individual terms, builds one custom store listing per feature set, and measures keyword rank. The AI-aware version keeps all of that but reorganizes it around intent clusters: groups of keywords representing distinct user personas, each pointing to its own custom store listing (CSL).

Take a fitness app as an example. Instead of one generic listing, you’d build separate clusters for people over 65, for bodybuilders, and for busy people squeezing in workouts around a full schedule. Each cluster gets keywords relevant to that persona and a CSL that highlights the features that persona actually cares about: cardio and health monitoring for older users, strength programming for bodybuilders, quick routines for the time-strapped group. That’s not a new idea for ASO consultants who already segment by persona. What’s new is confirmation, directly from a Google representative we spoke with, that the custom store listing shown in Ask Play results varies based on the prompt. Google Play is already surfacing these clusters in AI-driven search, not just in standard store browsing.

That turns CSL strategy into a dual-purpose exercise. You’re still building listings to lift conversion rate for a specific audience, but you’re now also building the raw material Google Play’s conversational search pulls from. A cluster built well does both at once.

 

Write descriptions that answer questions and hit keywords

Descriptions need the same dual treatment. The traditional rule still applies: keep App Store descriptions readable and scannable, and strategically place keywords in Google Play descriptions to support ranking. But descriptions are also being actively scanned, cited, and quoted by LLMs, including App Store descriptions, even though Apple doesn’t index that text for its own keyword search. Run a single query on ChatGPT or Claude using only their memory, and nothing may surface. Run the same query at scale, across dozens of variations, and descriptions start showing up as citations.

The practical fix is a hybrid format: write sentences that directly answer the question a user might be asking an LLM, and fold in priority keywords within that same sentence. For a hypothetical budget tracker app, that might read: “Budget Tracker is a budgeting app that helps you stop overspending and stay within a monthly budget, with automatic expense tracking and spending alerts.” The first half answers the implicit question (what does this app do, what problem does it solve). The keyword terms sit naturally inside that same sentence rather than being bolted on as a separate keyword-stuffed line.

In a live poll during the webinar, 80% of attendees correctly identified this blended format as the strongest approach, ahead of pure keyword-style writing or pure conversational Q&A formatting on their own. Neither extreme does the whole job. Traditional ASO drives keyword indexation, but only a small percentage of users read a full description anyway, and LLMs don’t index for keywords the way search engines do. AI-style answer formatting serves specific use cases well but can crowd out the broader keyword coverage you still need for traditional search. The blended approach is the one that doesn’t force a trade-off between the two.

 

Ratings, reviews, and showing up where LLMs already look

Ratings and reviews aren’t just social proof for human shoppers anymore. LLMs surface them directly, sometimes verbatim and sometimes paraphrased, which raises the stakes on keeping ratings high and responding to reviews promptly. Apps that respond quickly to both negative and positive reviews see measurably better odds of a positive review afterward, which compounds into better LLM citations over time. One practical workflow: use an AI browser extension to pull reviews out of App Store Connect or Google Play Console, ask it to surface sentiment patterns month over month, and sense-check whatever it returns before acting on it. That handles the pattern-spotting at scale; writing the actual responses still deserves a human, genuine touch rather than an AI-generated reply pasted in wholesale.

The bigger lever, though, is what we call source footprint: showing up in listicles, review media, blogs, and community forums. An app absent from these sources is simply unlikely to get cited, no matter how clean its metadata is. This is also the part of the framework that levels the playing field. In our query testing, some of the most-cited apps had unpolished descriptions and few reviews, but they had a strong presence on high-authority sites in their category. A new app with limited brand recognition can win citations the same way an established one does, by getting mentioned on the blogs, forums, and listicles that LLMs already treat as credible sources, rather than waiting to out-rank bigger competitors inside the store alone.

For an established app, that means keeping a steady stream of editorial coverage flowing around major updates and features, so there’s always fresh content for a model’s next training cycle to pick up. For a newer app, it means prioritizing exactly the same channels, just with more urgency: get into relevant Reddit threads, pitch your own listicles, and build relationships with the publications already covering your category.

 

Measuring AI discovery: still a black box, but not untrackable

Measurement is the question we get asked most, and the honest answer is that there’s no single source of truth yet, nothing equivalent to the 15-plus years of maturity behind Google Search Console. That doesn’t mean it’s unmeasurable, just that it takes combining a few imperfect signals rather than trusting one dashboard.

• ASO measurement platforms. Several have added their own AI tracking features, but the link between a store visit and a specific AI citation is still unclear, so treat this as directional rather than definitive.
• SEO and AEO platforms. These monitor brand visibility across AI answers, tracking share of voice, estimated citation volume, and which prompts and sources are driving mentions. The caveat is that most of this data comes from running simulations at scale rather than pulling directly from each AI platform’s own logs.
• Self-attribution. Simply asking users how they heard about you is, somewhat counterintuitively, the most reliable way to confirm an LLM-driven acquisition. Not every user will answer, but the ones who do give you a direct, unambiguous signal.
• App visibility and referral tools. Some platforms now flag ChatGPT, Claude, and other AI apps as referral sources. A portion of users will still fall through the cracks here too, especially anyone who gets a recommendation in a chat and then searches for the app directly, which shows up as a plain direct search rather than an AI referral.

The approach we recommend combining all four with is a simulation framework run on a schedule: write prompts that mirror real user conversations, run them across the major LLMs, and have an AI tool analyze the resulting conversations for total citations, citation type, and citation rate. Cross-reference those findings against your ASO performance and content publishing calendar to spot overlaps, and repeat the whole exercise monthly rather than weekly or biweekly, since shorter cycles don’t give enough volume to produce a reliable read. The core metrics worth tracking every month are estimated citation volume, citation rate, citation type by volume, and your standard ASO KPIs alongside them.

 

Where to start, and how to make the case internally

If you’re trying to convince leadership to invest time or budget here, the measurement gap is usually the objection you’ll hit first. Richard Velez’s answer during the webinar’s Q&A is worth repeating directly: the cost of not acting on what AI discovery already represents is much higher than the cost of testing one intent-driven CSL or adjusting a description. The risk of falling behind competitors who are already building their source footprint outweighs the modest cost of the first few experiments.

Assuming you have buy-in, or are building the case for it, two moves give you the fastest signal with the least friction. First, build out intent-based CSLs using your existing keyword list. This is low-friction because you can validate impact through conversion rate lift alone, independent of anything happening in AI search. Second, start building source footprint: get your app covered in blogs, listicles, and the communities already being cited by LLMs in your category. Between metadata work and external presence, external presence is consistently the highest-cited source type across the queries we tested, which makes it the second priority once your CSL foundation is in place.

Neither traditional ASO nor AI-driven discovery works in isolation anymore, and this isn’t a reason to abandon one for the other. It’s a reason to run both in parallel, using the same keyword and intent research to feed both.

Keep the conversation going: Phield Notes, Phiture’s new webinar series, continues these conversations beyond the annual Growth Phestival. Follow us on LinkedIn for the next session, or reach out directly to discuss how an intent-driven ASO and AI strategy would apply to your app.

 

FAQ

What is AEO for apps?

AEO, or answer engine optimization, is the practice of optimizing your app’s metadata, reviews, and external presence so that AI tools like ChatGPT, Claude, Gemini, and Perplexity recommend it when users describe a problem instead of typing a keyword. For apps, it sits alongside ASO rather than replacing it, since discovery now happens both inside app stores and inside AI conversations before a store is even opened.

Does traditional ASO still matter if AI discovery is growing?

Yes. Keyword targeting, rankings, and store conversion still drive the bulk of installs today, and nothing in this framework suggests abandoning them. The shift is additive: pair intent-based keyword clusters and custom store listings with the same metadata work you’re already doing, rather than treating ASO and AI discovery as competing priorities.

What’s a custom store listing, and why does it matter here?

A custom store listing (CSL) is a variant of your store page tailored to a specific audience segment or keyword cluster. Google has confirmed that the CSL shown in Ask Play results varies based on the user’s prompt, which means CSLs are no longer just a conversion-rate lever. They’re also raw material that Google Play’s conversational search pulls from directly.

How do you measure whether AI tools are actually driving installs?

There’s no single source of truth yet, so the reliable approach combines four signals: ASO platforms with built-in AI tracking, dedicated SEO/AEO visibility platforms, direct self-attribution by asking users how they heard about you, and a monthly simulation framework that runs representative prompts across the major LLMs and analyzes citation volume and rate. Repeating this monthly, rather than weekly, gives enough volume for a reliable read.

Can smaller apps compete with bigger, more established ones in AI search?

In Phiture’s research, some of the apps cited most often by LLMs had unpolished descriptions and few reviews, but strong presence on high-authority blogs, forums, and listicles. Source footprint levels the playing field in a way store ranking alone doesn’t, which makes it one of the highest-leverage moves for a newer or smaller app specifically.

What’s the first thing to optimize for AI app discovery?

Start with intent-based custom store listings built from your existing keyword list, since you can validate impact through conversion rate lift even before any AI citation data comes in. Then prioritize source footprint: getting covered on the blogs, listicles, and forums that LLMs already treat as citable sources in your category.