How to get ChatGPT to recommend your app

In August 2026 we asked a search-grounded assistant to recommend apps in 78 categories, three times each, and put every answer next to what App Store search returned for the same words that day. All 78 rows are at metarun.dev/research/ai-app-discovery.

The assistant reaches for smaller apps. Streaks was named first for "habit tracker" over an app with 144,966 ratings, and Mela beat the store's top recipe app. Store success guarantees nothing: 14 category leaders never came up in any run. The answers were stable too, with the same #1 coming back 93% of the time, so absence isn't noise you can wait out.

For a question like "best budgeting app", an assistant with web access searches, reads the results and summarizes the apps those pages recommend. It cites web pages, not only your App Store listing: apple.com, home of the public App Store listings, was cited in 90% of our categories, but alongside YouTube, Reddit and a long tail of review sites and blogs, 376 domains in all.

Each assistant searches a different index, so ranking well in Google doesn't guarantee a ChatGPT mention. Two things hold everywhere: a page has to be indexed to be retrieved, and it has to say something quotable about your app. Ahrefs' study of 75,000 brands found branded web mentions correlated with visibility in Google's AI Overviews at 0.664, against 0.218 for backlinks. That's correlation only, but it points the same way as our data.

Start with roundups. In Ahrefs' study of 750 ChatGPT prompts, "best X" lists were 43.8% of the page types cited, and brands placed higher on those lists were mentioned more. Take the cited pages from your manual check and write to each author with something checkable: what your app does differently, current pricing, a promo code so they can try it.

Then go where people talk. Reddit was cited in 59% of our categories and YouTube in 74%. Answer questions in your niche's subreddits as yourself, say you made the app, and bring it up only where it fits; Reddit's content policy asks for authentic participation and bans content manipulation. For YouTube, send the app to reviewers who already cover your category. Your own articles count too: habi.app, habitbox.app and getfinny.app were each cited in three to five categories.

Pick one category phrase people ask with ("habit tracker", not "behavior design platform") and use it in your homepage title, first heading and first paragraph, next to plain facts: what the app does, platforms, price. Microsoft's guidance for AI answers says to keep key answers in visible text rather than tabs or accordions, because assistants may not render hidden content. A short plain-text FAQ gives them sentences to quote.

Your App Store listing should say the same thing. The name (up to 30 characters), subtitle (up to 30) and the opening of the description are what any reader of the listing sees first, so state the category there, and set the Marketing URL to your site.

They overlap in your listing text and split everywhere else. ASO works on Apple's ranking: keywords, conversion, ratings. Getting named by an assistant, often called answer-engine optimization, works on what the web says about you. Our data says a strong store position doesn't transfer by itself: 14 of 78 category leaders weren't named at all.

llms.txt: Google says you don't need AI text files or special markup to appear in its AI features. Ahrefs checked 137,210 domains in May 2026: 28% published an llms.txt, and 97% of those files got zero requests that month. It's harmless, but don't expect it to change what ChatGPT says.

Keyword stuffing: there's no evidence that packing pages with competitor names gets you recommended, and on the App Store it breaks guideline 2.3.7. Assistants quote sentences, and a keyword list gives them nothing to quote.

Fake reviews and planted threads: manipulating App Store reviews or search breaks guideline 5.6.3 (Discovery Fraud), and Apple says it removes the apps and expels the developer. Reddit bans content manipulation too, and a thread exposed as planted is the opposite of the mention you wanted. MetaRun takes the same line: it measures where you stand and never posts, comments or writes reviews for anyone.