Mobile apps get recommended inside AI answers with no app-specific playbook behind it. The pages that carry those recommendations — a few review sites, the app’s own site, and sometimes a store listing — are ordinary web pages, and they are auditable the same way any web asset is. This post is the shape of that surface, and a baseline test you can run this week with nothing but a chat window.
Why apps are a different case
The short version: a SaaS or consumer site earns AI visibility on a domain it owns — get crawled, rank, get cited. A mobile app does not. Its primary listing lives on someone else’s domain, the App Store or Google Play, and the page an assistant actually cites when it recommends the app is frequently a page the app team never wrote: a review site, an explainer, a “best X apps” list.
So the first question for an app is not “where do we rank?” It is “which pages represent us at all, and do those pages say the same true thing?” That is a curation question, not a ranking question. If you have not done any AI visibility work yet, start with the AEO/Geo basics; this post is the app-specific mechanics.
How assistants surface apps: two channels that get conflated
The short version: no major assistant publishes a documented mechanism for choosing which apps to recommend. In practice there are two separate channels — the assistant’s web answers citing pages that mention apps, and apps that run inside the assistant itself. Most “get recommended” confusion comes from mixing those up.
Channel one: cited pages. When an assistant’s answer names an app, the name usually arrives via a citation or, more often, without one. We covered the mechanics of the cited case in Why AI answers triangulate: the answer is assembled from a small committee of pages, roughly six unique sources per cited conversation in Profound’s study of ~700,000 ChatGPT conversations (Profound, February 3, 2026: https://tryprofound.com/blog/chatgpt-citation-sources/). No assistant documents how it decides which apps to include in that committee, so the “best sources” list for app categories is observational, not documented.
One scoping note: no published AI citation dataset covers app categories. Profound’s work measures ChatGPT web citations, and Ahrefs’ monthly most-cited-domains study (September 2026: https://ahrefs.com/blog/most-cited-domains-in-chatgpt/) tracks web domains — reddit.com at a 16.8% mention share. Neither measures how often assistants cite store listings, review sites, or app domains per category, so we do not pretend to have that number.
Channel two: apps that run inside the assistant. OpenAI’s Apps in ChatGPT is a documented, separate thing: apps built on the Model Context Protocol that run inside ChatGPT conversations, with a review process for developer submissions. Being available there says nothing about whether your App Store listing gets cited in a “what should I use for X?” answer, and the inverse is also true. If someone pitches you on “getting your app into ChatGPT,” ask which channel they mean — the assets, and the work, are completely different.
Your store listing is a fetchable page — here is what we found
The short version: both stores publish their listing pages as ordinary HTML to search engines, and as of our check on October 28, 2026, neither one blocks AI crawlers. Both also serve structured data that describes the app. You cannot control either store’s index, but you can verify that your listing is in principle reachable and well-formed, and it takes twenty minutes.
What we checked, and what we found (our own tests, October 28, 2026 — treat as a single observation from one network, not a statement about how either store indexes anything):
- Crawlability. https://apps.apple.com/robots.txt allows the app listing paths and disallows internal paths only; https://play.google.com/robots.txt allows the store sections and disallows data-safety, XHR, and admin endpoints. Neither file contains rules for AI crawlers — GPTBot, Google-Extended, ClaudeBot, PerplexityBot. Neither domain serves an
llms.txt(both 404, same date). - Fetchability under an AI user agent. We sent plain HTTP requests — no browser — to a real App Store listing (Duolingo) and a real Play listing (YouTube), once with the user agent
GPTBot/1.0 (+https://openai.com/gptbot)and once with a standard browser user agent. All four requests returned HTTP 200. An initial URL we had for Duolingo (id568504177) returned 404 under every user agent, including a real browser; the live listing sits at a different URL. That is the caution: verify your own canonical listing URL the way a crawler would — stale app URLs 404 quietly, and a dead canonical URL is invisible to you until you ask. - Structured data. The Apple listing page served three JSON-LD blocks, including a schema.org
SoftwareApplicationwithOfferandAggregateRating(plus organization and breadcrumb markup). The Play listing served one JSON-LD block withSoftwareApplication,Offer, andAggregateRating. So whatever the stores do about AI, the machine-readable description of your app — name, developer, price, ratings — is already on the page an assistant would fetch.
You still cannot control whether GPTBot’s index includes your listing, or how often it re-crawls. What you can do is make sure the page it would fetch is complete, current, and consistent.
The pages an assistant leans on for an app
The assistant-side mechanism is not documented, so treat this as the surface we have observed, not a weighted list:
- A small set of review sites, per category. The same few domains carry a surprising share of the answer for a given category. Identify them per app — “best running apps” and “best budgeting apps” have different committees.
- The app’s own site, when it exists and is well-formed. Most app teams have one; the ones that win the citation often have the clearer, more structured one — a proper “what it is / who it is for / how much it costs” page, not a marketing splash.
- Editorial “best X” lists and comparison pages. These are the triangulation layer, and they are where a factual mismatch with your store listing does the most damage.
- Wikipedia, when the app is big enough for it. For smaller apps this is not a lever; do not build a Wikipedia-first strategy for an app.
- The store listings themselves — present as pages an assistant can fetch, as above, but they are rarely the primary cited source in an “app recommendation” answer. That is an observational finding from our tests, not a documented rule.
Store listing hygiene: the same checklist both stores publish
The short version: there is no separate “AI version” of store hygiene. Both stores publish guidance for humans, and because the page an assistant reads is the page a human sees, the human guidance is the AI guidance. The item that comes up in the most app results is cross-store consistency — name, category, and pricing should be the same in both places or the difference is intentional and documented.
Google Play. Play Console’s Get discovered on Google Play search (accessed October 28, 2026) is concrete: make the title unique and focused, and avoid common terms, subtle misspellings, and excessive emoji or ASCII; write the description around what the user gets, with the most important text visible in the top half of the web page where no scroll is needed; give each screenshot unique alt text.
Apple App Store. Per Apple’s App Store Connect documentation (accessed October 28, 2026), the product page is the app name, subtitle, keywords, and description — available in up to 40 languages — plus up to 10 screenshots and three app previews per supported language. Apple publishes A/B testing for the page itself (Product Page Optimization), so “the page works” is something you can measure on Apple’s side, not just assume.
What we check first in an app audit (the app slice of our methodology), in order:
- Same name (or a deliberately different name, documented) across both stores.
- Description’s first screen matches the name’s promise on both stores.
- Pricing and free-tier terms identical across both stores and the app’s own site. This is the single most common fact-check error in app-category answers we have seen: the listing says free, the site says a trial, the review site says a subscription, and the assistant is left reconciling it from memory.
How to run the baseline test
The baseline test is 10–20 buyer questions, run monthly in the assistants your buyers actually use — for most apps, ChatGPT, Gemini, Perplexity and Google AI Overviews — with every answer logged for which apps get named and which pages get cited. It is the app version of the first-20-minutes AEO audit, which walks the web side. No tools and no accounts are needed; it takes about twenty minutes.
- Write down 10–20 questions your buyers would actually ask an assistant. “Best app for X” is the obvious one; include narrower ones (“X without a subscription”, “X for [use case]”).
- Ask each in the assistants your buyers actually use. The answer differs across surfaces, and there is no cross-surface score.
- For each answer, record: which apps were named, in what position, and which pages (if any) were cited.
- Open each cited page. Note factual claims about your app — pricing, availability, features — and check them against both store listings and your site.
Two discipline notes. First: run each question more than once, in different order, and re-run the set monthly — answers are not deterministic, and a single run is an anecdote. Second: this tells you nothing about volume. It tells you where you stand on the pages that carry the recommendation — which is the only part you can act on anyway.
What the results typically point to
In the app category, the audit usually surfaces the same four gaps: fact-check failures in the answer itself, absence from the third-party pages that carry the category, a weak or stale own site, and inconsistencies between your store listings. The ranking-versus-citation gap applies to apps too — a good store ranking does not guarantee that assistants name you.
- Fact-check failures. The assistant’s answer contains a claim about your app — a price, a platform, a feature — that no current page supports, or that contradicts one of your own pages. The assistant built the claim from the committee; if the committee disagrees, the answer is a coin flip.
- Missing from the committee. You are not on any of the review or comparison pages that carry the category answer. This is a content-distribution problem, not an SEO problem, and it is usually the biggest gap.
- A weak or stale own site. Many apps have a domain that is a splash page, a dead link, or last updated years ago. The fix is a real “what it is / who it is for / how much it costs” page.
- Store listing inconsistencies. Name, category, or pricing differ between App Store and Play, or between the store and the website. The cheapest fix on this list, and the most common.
The point is not the fixes themselves — all four are ordinary work. The point is that an app’s AI visibility is a property of a set of pages, some of which you do not control, and the audit is how you find out which set before you spend time on the wrong one.
What this does not tell you
Be clear about the edges, because this is where app teams over-buy:
- No assistant documents how it recommends apps. The “pages it leans on” list is observational, from our tests and from reading answers. It is a useful map, not a spec.
- No published dataset covers app-category AI citations. The public citation studies (Profound, Ahrefs) cover web domains, not app categories. Anyone who quotes you a precise “app citation rate” is inventing it — including anyone who quotes it to you.
- A good baseline is not a score. A monthly pass over 10–20 questions is a starting map. It is not a rank, a trend line, or a number you can put in a board deck.
- The store-side checks are a single observation from one network, on one date. Stores change their robots, their HTML, and their structured data without notice. Re-run the checks when it matters.
The honest summary: app AI visibility is real, it is mostly about curation rather than ranking, and the parts you can influence are findable in an afternoon. The parts you cannot influence — how the committee is chosen, what share of assistant traffic an app category gets — do not have honest numbers yet, and we would rather say that than give you a fake one.
If you want this run on your app. The AI Visibility Audit is a one-off $99 engagement, no subscription: you send us the brand, both store URLs, and the main competitor, and within 1 business day we reply with 30–50 draft buyer questions to review. Payment starts the clock; the PDF — with fact-check errors and a ranked fix list — lands in 3 business days.