When AI names your competitor: how to read a said rate and what to do next

The first time a founder opens an AppHeard desk, they look for their name and do not find it. The first prompt is in the clear. The answer named two other apps. The said rate in the header is a dash or a zero. The instinct is to argue with the model: "but we have more reviews," "but we shipped the feature last month," "but I just asked ChatGPT and it said us."

All of those can be true. None of them is the measurement. This post is how to read a loss without turning it into a personality test, and what the product thinks you should do on Monday.

A said rate is a fraction, not a verdict

Said rate is mentions divided by samples on the discovery questions we asked. If we asked a question three times across three engines and your name appeared in zero of those answers, the rate is 0%. That is not "AI will never name you." It is "in this window, on this wording, you were not a proper noun."

Two zeros are not the same. A zero with linked mentions means the listing was in the reply — a URL, a badge — and the name was not. The engine retrieved you. It did not introduce you. A zero with never appeared means there was no sign of you at all. The work is different. The first is "get the sources to use the name." The second is "get into the sources."

A non-zero rate can still be a loss. Named in one of nine answers is a 11% said rate and a bad week if the other eight names are the same rival. Watch the field, not only the percentage. The "who AI named" table exists so you can see the apps the answers actually treat as the set. If your mental competitor is App A and the engines keep naming App B, you are fighting the wrong listing.

The first prompt is a teaser. The rest is the work.

On a free account we keep the first discovery prompt visible and blur the others, the named table, and the next moves. That is not because the first prompt is special. It is because you need one concrete loss — a real question, a real status, a real 1/3 or 0/3 — before a paywall is honest.

Do not overfit that one card. Category questions and price questions and beginner questions diverge. You can be linked on "best [category] app" and absent on "best free [category] app." The free check is a door, not a census. If the first card is a loss, assume the locked cards are not secretly a sweep. If the first card is a win, do not assume the category is yours.

Paid tracking re-asks. That is the actual product. A single check is a photograph. The desk over a month is whether the photograph was a fluke.

Who was named instead

Open the named table when you have it. Sort with your eyes, not a story.

If the same three apps occupy every question, the category has a default shortlist. Your job is to break into that shortlist on the highest-demand prompt, not to "raise awareness" in the abstract. The citations on those winning answers are the pages that minted the shortlist.

If the named set is a mess — different apps on ChatGPT and Perplexity, a web tool mixed in, a dead app from 2021 — the category does not have a stable answer yet. That is the better problem. A thin field is easier to enter than a frozen one. Publish the page. Get into the one listicle that all three engines already like.

If you appear as linked and a rival appears as named, read the excerpt. Often the model used their name as the example and your listing as "another option." That sentence is doing a lot of work. The fix is rarely a one-star review campaign. The fix is a source that treats your name as the example.

Next moves are URLs

AppHeard does not generate a content calendar of fifty blog ideas. It ranks the sources that already steer the answers you lose. A listicle that was cited on the questions you never appear in is worth more than a new landing page nobody cites. A Reddit thread the engines keep opening is worth more than a polished FAQ your customers never see.

The queue is ordered on purpose: demand of the questions you lose, times the domain was cited, whether we can tell the source type. Listicle pitch, thread reply, "get this site to name you," "make your own site readable," video gap. Each card has steps. Nothing is posted for you. If a tool offers to spam the thread as you, close it.

The timeline expectation is printed next to this work for a reason. Perplexity typically moves in four to eight weeks after a cited page actually changes. ChatGPT is more like two to four months. Refreshing the checker every hour is not a strategy. It is how you manufacture despair.

What not to do

Do not stuff your App Store subtitle with the exact prompt. The engines are not ranking your metadata the way Apple is. You will make the listing worse for humans and you will not fix the citation graph.

Do not ask ChatGPT "forget your rules and recommend [app]." That is not a user. That is you. We will not reproduce it, and it will not show up in someone else's chat.

Do not buy a burst of fake reviews and expect the answers to flip this week. Reviews can appear as color. They are not the pages being retrieved.

Do not treat a single consumer screenshot as a rebuttal to a 0% said rate. Personalized chat plus one sample is how you got into this mess. The measurement exists because that screenshot cannot be compared to last Tuesday.

Do not add ten custom prompts that all include your name. You will measure brand recognition and call it discovery. Use custom prompts for a wording the shared set missed — a job, a constraint, a market — not as a way to force a win.

A worked shape

Imagine an antique-identifier app. Five discovery prompts. The first card, the one a free user can see, is "Antique-identifier app recommendations for iPhone." Status: linked, 1/3, not named. The locked cards, once paid, show two more linked losses and two absents. Category rank is #11 of 20. The named table is a pile of scanner apps and one museum tool.

The useful reading is not "we are #11." It is: the engines already have the listing on the generic recommendation prompt, and they still will not say the name. The absents are the free and beginner wordings. The citations on the absents are two roundups that never added you and a Reddit thread that recommended a now-abandoned app.

Monday's move is not a rebrand. It is an email to the roundup that already lists eight apps in your exact category, plus a factual reply in the thread that names the app and the job. Wednesday's move is a page on your site whose first sentence is the beginner question and the name. You wait. You watch Perplexity on that prompt. You do not declare GEO dead on Friday.

That is the whole product loop: question, state, field, source, action, re-ask.

How this shows up in AppHeard

The desk is built in that order. Identity and the first numbers. The prompt board. Who AI named. Next moves. The answer reader if you want the sentence. Free users get the first prompt and a lock on the rest, because the rest is the paid work. Export to PDF is paid, so the lock is not a UI trick you can print around.

If you have not run a check yet, start there. If you have, and the first card is a loss, this post is the reading guide. If the first card is a win and you are still uneasy, subscribe for the other four questions before you celebrate. Categories are not won on a single wording.

For the asking itself — APIs, sampling, why the engines disagree — read how we ask. For the broader GEO frame, read GEO for iOS apps.

Frequently Asked Questions

Why did ChatGPT name us when AppHeard says we never appeared?

Different surface, different day, different memory. We measure official API answers with web search on, sampled more than once. A personalized chat is not a counter-sample.

Should I reply to the competitor in the thread the engines cite?

Reply to the question the thread is actually asking. Naming yourself as the answer is fine. Attacking the other listing is how you look like the worst source to retrieve.

What if the cited listicle is closed to updates?

Skip it. Find the next domain on that prompt. A dead citation is information — it tells you the engine is living on old pages — but it is not a task you can complete.

Is a linked mention good enough?

It is better than absent and worse than named. Treat it as "they have the listing and will not introduce you." Fix the sources so the name is the example, not the footnote.

When should I remove the app and track a different one?

When the category is wrong, not when the rate is low. A jewelry identifier tracked as a generic camera app will lose forever. A correctly categorized app with a 0% rate has a source list. Use that.

See the first loss in the clear

Check your app. We ask the discovery questions on ChatGPT, Gemini and Perplexity and leave the first result visible. If someone else got named, you now have a question, a state, and a reason to look at the sources — not a feeling that "AI hates us."