GEO for iOS apps: getting named when someone asks AI

App Store Optimization taught a generation of iOS teams to think in keywords, screenshots, and conversion. That work is not obsolete. The store is still where the download happens. What changed is the shelf before the store. A growing share of "what app should I use?" no longer starts in Search. It starts in a chat box that returns three names, two links, and a paragraph that sounds like a friend.

Generative engine optimization — GEO — is the clumsy name for that shelf. For an iPhone app the job is specific: when someone asks ChatGPT, Gemini or Perplexity for an app like yours, does the answer say your name, link your listing, or skip you? AppHeard is built for that question. This post is the use case, not the pipeline.

ASO and GEO are different funnels

ASO is a ranking inside Apple's index. You pick keywords, you watch impressions, you A/B the icon. The unit is a search term in the store. The competitor set is whoever else bid for that metadata.

GEO is a ranking inside someone else's answer. You do not pick the prompt. The buyer does — or the model rewrites it. The competitor set is whoever the engine is willing to name, including apps you have never considered peers and websites that are not apps at all. The unit is a discovery question: "best free antique-identifier app for iPhone," "app to identify jewelry from a photo," "fasting tracker with custom windows."

Winning ASO and losing GEO is now a common shape. You own the store keyword. The chat names a rival whose blog post, Reddit thread, or 2023 listicle is what the engine retrieved. The buyer never opens the App Store search box. Your impression share looks healthy. Your AI said rate is a zero.

The reverse also happens. An app with a mediocre store listing can get named because a YouTube script or a niche roundup keeps getting cited. GEO is not a participation trophy for good ASO. It is a second game with a different ball.

The questions that matter are not branded

Founders instinctively check "what is [our app]?" That is a vanity prompt. If the buyer already knows the name, you have already won the hard part. The prompts that decide installs are unbranded. They name a job, a price, a skill level, a platform. They sound like a person who has not picked yet.

AppHeard writes those from the listing: category, job-to-be-done, price, context. We ask them as they would be asked, including the "for iPhone" constraint, because a lot of answers otherwise drift to Android clones or web tools. We do not let the customer type a prompt that only their app can win. That would be a demo, not a measurement.

Demand is not equal across that set. "Best [category] app" is usually louder than "app for beginners in [category]." We estimate demand from App Store and search signals — the engines do not publish prompt volumes — and we put the high-demand losses at the top of the desk. If you only have time to fix one question, fix the one people actually ask.

Named, linked, absent

The answer has three states that lead to different work.

Named. The reply used your name. This is the said in "said rate." You can still lose the install if you are fourth in a list of five, but you are in the conversation.

Linked. The reply pointed at the listing — an App Store URL, a badge, a "get it on iOS" chip — and never said the word. Founders read this as a win. It is a half-win. The engine already retrieved you. It declined to treat you as a proper noun. That is often a citation problem: the pages it read use the listing and a generic label ("this scanner app") instead of the name.

Absent. No name, no link. The reply is about the category and you are not in it. This is the one that feels like death and is usually the most actionable. Someone else is in the sources. Those sources are the next moves.

A dashboard that collapses all three into a single "visibility" grade will send you to write homepage copy when the actual job is to get into a listicle, or to get a listicle to use your name. The desk keeps the states separate on purpose. The free teaser shows you the first prompt in the clear so you can see which state you are in before you subscribe.

Citations are the ranking

ASO people look at Apple's algorithm and guess. GEO people can look at the footnotes. Perplexity especially, and the others when search is on, will show the domains they used. Those domains are the ranking factors you can actually touch.

In practice the citation graph for a small iOS niche is short. A handful of listicles, a subreddit thread, a YouTube "I tried five apps" video, maybe the developer's own site, maybe a docs page that does not mention you. If three of those never added your app, you will not be named, no matter how clean your App Store subtitle is.

This is why AppHeard stores cited sources with the answer and turns the ones on questions you lose into a queue. Pitch the listicle. Reply in the thread. Publish a page that names the app in the first sentence. None of that posts on your behalf. The product stops at the draft and the URL. You still send the email.

It is also why stuffing "best jewelry identifier" into your App Store description is a weak GEO move. The engines are not grepping your metadata. They are reading the web they already trust. Change that web.

Category rank is a field, not a trophy

Once you measure a whole category — every app the answers name, not only yours — you get a leaderboard. That number is seductive. It is also easy to misuse. Being #11 of 20 is not an App Store chart. It is "of the apps the engines bothered to name this window, you were the eleventh most said." An app can sit at #3 with a 20% said rate. That is a thin field, not a crown.

Use category rank to see who you actually lose to. Use said rate to see whether you exist. Use the per-question board to see which wording you win and which you never appear in. The founder who only watches the hash number will "improve" by adding more apps to the denominator.

What a week of GEO work looks like

Monday: read the questions you lost. Not the score. The questions. Pick the highest-demand absent or linked-only prompt.

Tuesday: open the citations on that prompt. Write down the three domains that appear most. Ignore the ones you cannot influence this quarter (a huge publisher with a closed updates page).

Wednesday: ship one artifact aimed at those domains. A page on your site that answers the question in the first two sentences and names the app. A reply in the Reddit thread the engines keep citing. An email to the listicle editor with a factual gap they already mentioned.

Thursday: do not refresh ChatGPT fifty times. You will see noise and call it a trend. If you are on a paid AppHeard plan, the daily pulse will tell you if Perplexity moved. ChatGPT is slower. Our public expectation is still Perplexity in weeks, ChatGPT in a couple of months.

Friday: check whether you were linked but not named. If yes, the next artifact should use the name as a proper noun in a place those domains might copy. If you were absent, you are still in the "get into the source" phase, not the "tweak the wording" phase.

That is GEO for an iOS app. It is editorial and relational. It is not a keyword tool with a new acronym.

Where AppHeard sits

We ask the same questions on all three engines, store who was named and what was cited, and show you the first prompt even on the free plan. The rest of the desk — every prompt, who AI named, the next moves — is what the subscription is for, because that is the work after you decide the game is real.

If you want a sample of a finished category, the sample report is a live one. If you already know you lose a question and want the source list, that is the paid view.

Frequently Asked Questions

Is GEO just ASO for ChatGPT?

No. ASO ranks you inside the App Store. GEO is whether an answer names you before the buyer ever opens the store. The levers are citations and pages, not your subtitle.

Should I optimize for one engine?

No. They do not share an index. A Perplexity win that never appears on ChatGPT is still a win, and the reverse is common. Measure all three.

Do more App Store reviews help GEO?

They help ASO, and they sometimes appear in answers as social proof. They are not a substitute for being in the pages the engines retrieve.

How long until a listicle pitch shows up in an answer?

Perplexity often moves in four to eight weeks if the page actually changes. ChatGPT is slower. We print that expectation next to the tasks because a weekend is not a test.

Can I GEO my way around a bad product?

You can get named once. You cannot stay named if the answer's next sentence has to describe a product that does not exist. Accuracy is a scoring component for a reason.

Run the questions on your listing

The free check writes discovery prompts from your App Store page and asks ChatGPT, Gemini and Perplexity. You will see whether you are named, linked, or absent on the first question. That is the GEO baseline. Everything after it is work on the sources, not work on the chat box.