How AI Recommends Apps: What ChatGPT, Gemini and the App Stores Read, and What You Can Fix

Neither the App Store nor Google Play names a single AI crawler in its robots.txt. Both let every bot read app listing pages, and both put the description, rating and reviews in plain HTML. So when ChatGPT or Gemini gets your app wrong, a blocked crawler is rarely the reason.

This guide covers what each surface reads, what the platforms document, and the four things an app publisher can check this week. Every claim below was read off the platform's own page on 24 October 2026. Where nobody has published a measurement, we say so.

The stores are already open to AI crawlers

Apple's robots.txt has one group, User-agent: *. It disallows /WebObjects/, /api/, /includes/, /v1/ and search result URLs, and nothing else, so app listing pages are open to GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended alike.

Google Play's robots.txt also has only a wildcard group. Listing pages under /store/apps/details are allowed. The deeper review feed at /store/getreviews is disallowed, so a crawler sees the reviews printed on the listing page, not the full archive.

We also requested one large app's listing on each store as seven user agents, from a browser to GPTBot, ClaudeBot and PerplexityBot. Every request got a full page. These were spoofed user agents from one network, not the vendors' real crawler IPs, so treat it as a spot check, not proof.

If your app is described wrongly in AI answers, the store page is almost never the part that is blocked.

What each AI surface reads

Surface	What the platform documents as input	Your control
ChatGPT, Claude, Perplexity, Gemini answers	Nothing app-specific. They read the web, store pages included	Your own site's crawler access; facts that agree everywhere
App Store tags	Your App Store Connect metadata, AI, and human curation	Review and deselect tags (US only)
App Store review summary	User reviews, filtered for spam, profanity and fraud	Report a concern
Google Play review summary	English user reviews, by machine learning	None documented
Ask Play, Play in Gemini	Not documented	Unknown
Siri and Apple Intelligence	Nothing on recommending App Store apps	Unknown

Nobody has measured what AI cites for "best app for X"

The one number we found is from Ahrefs: App Store pages made up 6.6% of ChatGPT's 1,000 most-cited pages in September 2025. That is one month, the top 1,000 pages only, with no split by query type, and Ahrefs had an AI model sort the pages, which it says "may not be 100% fool-proof".

We found no study, with a stated method, of what assistants cite when someone asks for an app recommendation: the store listing, a review site, or the app's own website. Anyone quoting a figure for that is quoting a guess. The honest move is to ask the questions your users ask and read the sources yourself.

The app stores now have their own AI

Apple: tags and review summaries

Apple's App Store tags "are based on the App Store metadata you provide in App Store Connect, artificial intelligence, and human curation." They are shown only in the United States, in US English. You can deselect tags you don't want, though Apple warns that "deselecting all tags may affect your discoverability."

On iOS 18.4 and later, Apple shows a reviews summary "generated using large language models." Apple's research write-up says it first filters out spam, profanity and fraud, then writes a paragraph of 100 to 300 characters, refreshed periodically. You cannot edit it, but you can report a concern from the device or from App Store Connect.

Google Play: review summaries, Ask Play and Gemini

Google Play builds review summaries with machine learning from English reviews. An app "needs to have enough similar reviews around more than one theme" to get one. Google documents no way for a developer to correct a summary.

In October 2026 Google said Ask Play "can generate a comparison between apps, highlighting key differences and user sentiment," and that Play in Gemini will bring Play's apps into Gemini conversations. At I/O 2026 it announced app discovery inside the Gemini app. Google has not said what feeds any of these, or what a developer can change.

Siri

Apple's September 2026 Siri announcement describes acting inside apps people already use. It says nothing about Siri recommending new apps from the App Store, so there is nothing yet to optimize or check.

Four things an app publisher can check this week

1. Check that AI crawlers can reach your own app website. The store page is open, but your landing page, help centre and pricing page are yours. If your firewall refuses OAI-SearchBot or your pages render only in the browser, the assistant gets your facts from the store and from third parties instead. Our robots.txt tester guide shows how to check per bot.
2. Use one name, or list every name you use. Say a made-up app is called "Roomly" on its website and "Roomly – AI Interior Design" in the App Store, with a web version on a different domain. A reader can join those up, but you have no evidence an assistant will. Pick one primary name, say the other names out loud on your own site, and track every variant when you check AI answers.
3. Review your App Store tags. If you sell in the US, open App Store Connect and deselect any tag that describes your app wrongly. Tags come from your metadata, so a wrong tag usually points at a wrong line in your description. Whether those tags change what ChatGPT or Gemini says about your app is unmeasured.
4. Read both AI review summaries for wrong facts. A summary that says your app has no free tier, or crashes on launch, is a claim shoppers read before they tap Get. Report it on Apple's side; on Google's side the only route is newer reviews on the themes that matter. Then check whether ChatGPT and Gemini repeat the same claim: see our guide to fixing wrong AI answers about your brand.

What does not help

Firebase App Indexing is no longer recommended for Google Search; Android App Links are its replacement for sending search users into an app. Google's guide to AI features and your website never mentions apps, App Links or Universal Links. So the claim that deep links lift you in AI Overviews or AI Mode has no evidence behind it.

ChatGPT has its own plugin directory, which only matters if your app has a backend worth exposing as a tool. By default people find a plugin by a direct link or by searching its name. OpenAI says featured placement and proactive suggestions are earned and that "developers cannot request enhanced distribution."

Every "ASO for AI" tactic we came across while researching this came from a vendor selling it. None came with a measurement.

Measuring the result

Knowing what AI says is half the job; knowing whether it sends installs is the other half, and it is harder for apps than for websites. Our companion guide, tracking app installs from ChatGPT, covers what App Store Connect, Google Play and your own deep links can and cannot show.

FAQ

Can ChatGPT read my App Store listing?

Yes. Apple's robots.txt allows every crawler on listing pages, and the description, rating and reviews are in the page's HTML. Whether a given answer used your listing is something only its cited sources can tell you.

Can I edit the AI review summary on my app?

No. Apple lets you report a concern about a summary; Google Play documents no developer control. Both are built from user reviews, so the summary changes only as reviews do.

Do App Store keywords affect what AI assistants say about my app?

Unmeasured. Apple says your metadata feeds its own App Store tags. No platform documents ChatGPT, Gemini, Claude or Perplexity reading App Store Connect keywords, and no study has tested it.

See what AI says about your app before your users do. Run a free AI visibility audit on the questions people ask about apps like yours: no credit card.