How to get your app recommended by AI assistants

Why this matters for app makers

People used to find apps by searching the App Store. Increasingly they describe a problem to an assistant: "is there an app that shows file copy progress in the menu bar?" The assistant names two or three apps. If yours is not one of them, the customer never reaches the store search where your keywords were waiting.

I know this from the receiving end. A customer found my Mac app, Cartage, by asking ChatGPT exactly that kind of question, and bought it. I then tested all 11 of my product sites to see how typical that was. It was not typical at all: eight of the eleven were never mentioned. The case study has the numbers.

Where assistants find apps

The stores

In August 2026 the app-store-optimisation company AppTweak published an analysis of more than 125,000 ChatGPT responses to 9,489 app-related prompts in the US, collected in May 2026. Its findings on what ChatGPT cited:

• Apple App Store pages: 37.6%
• Google Play pages: 9.5%
• The open web, mostly apps' own sites and aggregators: 27%
• No source cited: 22.6%

AppTweak sells tools in this area, so read it as a vendor's study. It is still the largest I know of, and its method is stated.

The open web

My own test was much smaller: 39 questions, one run each, through OpenAI's search-enabled model, across 11 products. Where my apps lost, the cited sources were mostly small competitors' own websites, far more often than App Store pages.

The two results are not in conflict. Store pages are cited a lot in total, because every app has one. But for a specific question, the page that wins is often an app's own site, because it can say things a store listing has no room for: who the app is for, what it replaces, how it compares.

One of my test questions was answered with programming code and no sources at all. The assistant did not know a suitable app existed, so it explained how to build one. That is what invisibility looks like.

Fix the store listing first

Your listing is public web content. The stores publish it at an ordinary address that crawlers can read.

• Write the first sentence for someone with a problem. "Shows Finder copy progress in your menu bar" answers a question. "The ultimate productivity companion" does not.
• Use the words people use. If they say "menu bar", say "menu bar", not "status item".
• Say who it is for and what it replaces.
• Be exact about price. Free, paid once, or subscription, with the amount.
• Keep the name consistent. The same name on the store, your website, your social profiles and in your press kit. My customer called Cartage "Cartedge" and the assistant still found it. I would not rely on that.

A rename is a particular risk. For weeks the old name lives on in the store, in reviews and on other sites, and assistants will repeat whichever they found.

Give your app a real website

An app with only a store page has one source. An app with a good website has two, and the second is the one you control. From what I found on my own sites, here is what the website needs.

A download link that works

It sounds too obvious to write down. On one of my sites, the "Download on the App Store" button had never worked: a coding slip sent it nowhere. Two other sites had no store link at all. An assistant that found them had nowhere to send anybody. Click your own button today.

A plain description at the top

What it is, what it runs on, who it is for and what it costs, in the first paragraph, as text.

Structured data

Use the SoftwareApplication type (or MobileApplication) with the operating system, the category, the price and the store link. How schema markup works.

Questions in your users' words

A FAQ answering what people actually ask: does it work offline, does it need an account, is there an Android version, how is it different from the built-in feature. If the answer is "no, it does not do that", say so. An assistant that can rule your app out correctly trusts it more when it rules it in.

An honest alternatives page

This was the most useful thing I learned. When Cartage was recommended in my tests, its "alternatives" page was one of the pages cited. For another app, Reroute, I wrote a comparison with the best-known competitor that includes when the competitor is the better choice and what my app cannot do.

People ask assistants "what are the alternatives to X?" all the time. If the only pages answering are your competitors', guess who gets named.

Guides for the problem, not the product

Write about the job your app does. Someone asking "how do I see copy progress on a Mac?" is your customer before they know apps like yours exist. Cartage has seven such guides, and it was the best-performing site in my test.

The technical basics

Crawlers allowed in robots.txt, a sitemap, a canonical address, pages indexed in Bing and Google. A 20-minute check.

Be mentioned where assistants look

Assistants weigh what others say about your app.

• Listing sites. Get onto the directories people use for your kind of app, such as AlternativeTo and Product Hunt, with an accurate description.
• Reviews and round-ups. A fair write-up on a relevant blog is worth more than ten press releases.
• Communities. Genuine answers in forums where your users gather. Not spam: moderators and machines both notice.
• Store reviews. They are part of your store page, and the store page is the most-cited source there is.

What will not help

• Keyword stuffing in the listing. It reads badly to people and adds nothing a machine cannot already see.
• A website that is one screenshot and a badge. There is nothing in it to quote.
• Hiding the price. If the assistant cannot tell what it costs, it may say so, or pick an app whose price it knows.
• Claiming features you do not have. It will be repeated, then contradicted by your reviews.

How to test it

Write three or four questions a user would ask before knowing your app exists. Ask ChatGPT, Claude, Gemini and Perplexity. Note which apps are named and which pages are cited. Then open the cited pages and compare them with yours. A fuller method.

Frequently asked questions

How do AI assistants decide which apps to recommend?

They combine what the model already knows with a live search. For ChatGPT, a 2026 AppTweak study found App Store and Google Play pages made up 47.5% of cited sources, with apps' own websites and aggregators making up most of the rest. A clear listing and a real website both count.

Does App Store Optimisation help with AI recommendations?

Partly. A clear title, subtitle and description help, because store pages are public and widely cited. Keyword tricks aimed at store search do not carry over. Assistants respond to plain descriptions of the problem the app solves.

Does my app need its own website?

It helps a great deal. A website is a second source you control completely, and it can hold comparisons, guides and answers that a store listing cannot. In my own tests, apps were often beaten by small competitors with better websites, not bigger ones.

Why does ChatGPT recommend competitors with fewer downloads?

Because download numbers are not what it reads. It finds pages that answer the question. A small app with a clear website, a comparison page and a few honest reviews can be recommended ahead of a larger one with a vague listing.

How do I track whether AI is sending people to my app?

On your website, look for visits with utm_source=chatgpt.com and referrers such as perplexity.ai. For store pages, tracking is limited, so ask new customers how they found you. One question in a support reply is how I found out about ChatGPT.

Sources

• AppTweak, AI search and LLMs bring a new era of app discoverability, 12 August 2026. A vendor study: US market, May 2026 data.
• OpenAI, Publishers and developers FAQ. Checked 10 November 2026.
• My own baseline of 39 questions across 11 product sites, 10 November 2026: the case study. One run per question, so a sample, not a measurement.