How to Get Your App Recommended by ChatGPT and Claude

Short answer

We found no form or setting that makes ChatGPT or Claude recommend your iOS app in an ordinary answer. With web search on, they build answers from pages they find, so the work is publishing clear public pages that answer your users’ questions, then measuring whether you appear.

Our evidence comes from 340 API runs on one day about ASO tools. It suggests what may work for other categories and proves nothing. OpenASO itself barely appeared, and our blog launched the same day, so we cannot yet show this advice worked.

Is this the same as submitting an app to ChatGPT?

No. In December 2025 OpenAI opened submissions for apps that run inside ChatGPT. They are built with its Apps SDK, connect through an MCP server, and appear in a directory. OpenAI’s current submission docs call them plugins. None of this is a way to list a native iOS app.

AI answers blur the two. We asked OpenAI’s API “How do I get my iOS app recommended by ChatGPT?” three times. All three answers offered building a ChatGPT app as a route, and two called it the main or practical path. Claude’s single answer said there is no way to sign up or pay.

What did we measure?

We ran prompts through both APIs with web search on, then counted each distinct URL once per run. Source figures below exclude the four branded prompts and 20 template URLs, leaving 66 prompts and 2,285 citations.

Which sources do AI answers cite for ASO questions?

The top domains were developer.apple.com (27.9%), github.com (8.1%), appfollow.io (3.5%), apptweak.com (3.4%), and ads.apple.com (2.7%). “Vendors” comes from a rough domain list we sorted by hand, so treat that share as approximate.

OpenAI leaned on Apple, Claude on vendors. The gap holds on the first run of each prompt alone, at 59.9% and 17.1% for Apple. Claude also gave more links per run, 11.4 against 4.8, though it wrote them into the answer text while OpenAI mostly returned structured citations.

The mix depends on the question:

Two results deserve care. Listing pages were 22 distinct URLs across 19 runs, and 26 of their 41 citations came from prompts asking for a tool. AppTweak reports that listings were 47.5% of ChatGPT citations across 125,000+ recommendation responses. That is AppTweak’s own claim, and the article does not give its prompts, period, or classification rules. Its gap with ours may simply reflect different questions.

Reddit’s nine citations all came from one prompt that names Reddit, and Claude retrieved no Reddit pages in 118 runs. AppTweak argues that communities matter a lot for recommendation questions. Our panel does not show that for ASO, and it cannot say for your category.

What are the limits of this data?

• API, not the consumer apps. Search was required on OpenAI, and every prompt asked for current sources and links. Real answers may look different.
• Modelled prompts. We wrote them. Nobody outside the AI companies can see real prompts.
• One day. All runs took place on November 1, 2026.
• Two models. Other models may behave differently.
• Uneven repeats. Claude ran once on 40 prompts, so those rates are less stable.
• One topic. The prompts cover ASO, not consumer categories such as fitness or budgeting.

The panel measures what these engines cite for ASO questions. That only suggests what works elsewhere.

How did OpenASO do?

Badly, so far. Across 156 non-branded runs in our first phase, our app was named once, on OpenAI. Four other answers matched the name but pointed to openaso.ai, a separate site that uses the same name. Across the 160 runs in our second phase, it was named zero times. Claude never named it in either phase.

Our blog went live the day we ran this, so the result says little about the content. One lesson stands: read the answers, because a match on your name may not be you.

What can you do that the evidence supports?

1. Answer real questions on pages you own. Vendors took the largest share, which suggests published answers get picked up, at least in ASO.
2. Let the search crawlers in. OpenAI says sites opted out of OAI-SearchBot will not appear in ChatGPT search answers, apart from navigational links. Anthropic says blocking Claude-SearchBot or Claude-User may reduce visibility.
3. Make your listing plain and specific. Apple says your description is used for web engine search results. AppTweak calls store descriptions a supporting signal and says their impact is not yet proven. See our description guide.
4. Find the pages engines already cite. For alternatives questions, vendor pages dominated. Your category’s cited sources will differ, and your own log will show them.
5. Keep a clear repo if you serve developers. GitHub was the top source on AI workflow prompts. This will not help most consumer apps.

How can you repeat this for free?

You need a spreadsheet and access to the assistants you care about. No paid tool is required.

1. Write 10 to 20 prompts in your users’ own words. Take the language from reviews and support email.
2. Run each several times in a fresh chat, on each assistant you care about.
3. Log the date, assistant, prompt, whether you were named, whether you were linked, which rivals appeared, and which sources were cited.
4. Compare rates across repeats, not single answers.
5. Repeat monthly with the same prompts. Change the list only when you accept losing the comparison.

Start with ten prompts today, and tidy your listing with the metadata checker. We plan to rerun our panel once our pages have been indexed, to test this advice on ourselves.