👋 I’m Sarah: 10 years measuring mobile marketing, now measuring what ChatGPT says about your brand
I’m Sarah, the founder of Operato AI, the self-serve AI Visibility platform.
In 2014 I joined AppsFlyer as an early employee and spent 10 years there, building the French market and helping marketing teams measure where their customers actually came from.
One lesson stuck with me: every time a new acquisition channel appears, the companies that measure it first are the ones that win it. Mobile apps were that channel in 2014.
This year I noticed the next one. Buyers are asking ChatGPT, Claude, Gemini or Perplexity “what’s the best tool for X?” instead of Googling. The assistant answers with 3 or 4 names, and if you’re not one of them, you never even get the visit. Yet most companies have no idea what those answers say about them.
So I built Operato AI. Every week it asks the assistants the questions your buyers ask, then shows how often you’re named, who gets recommended instead, which websites the assistants trust, and what to fix first. In English and French 🇬🇧🇫🇷
We’re launching here on Nov 17, and I’d love to learn from this community before then:
Have you ever checked what ChatGPT says about your own product?
Did anything surprise you?
Happy to swap notes with anyone working on AI search, GEO or marketing measurement. 🙌
Replies
Welcome Sarah. Your old employer is in my app, so one note from the other side of the SDK. The thing I got wrong first was the order of operations: the SDK runs with manual start, the session is what carries the customer user id, so the id has to be set and then the session started, ahead of anything awaited, or a slow database write on launch costs that launch its attribution. And a launch with nobody signed in still starts anonymously, because losing an install over an abandoned sign up form defeats the point of measuring installs at all. Both are the kind of thing that only shows up as a hole in the numbers weeks later. On your new channel, a question rather than an answer, since I have no data of my own there: for a consumer app, when someone asks an assistant for a song for their mother's birthday, does it name apps at all, or does it just write the lyrics itself and the visit never happens for anyone?
@gbasovv Thanks George, that’s a really useful note. Setting the user id before the session starts, and still counting anonymous launches, are exactly the kind of bugs nobody notices until the numbers look off weeks later.
Your question is the right one, and the honest answer is: mostly the second. If someone asks for lyrics for their mom’s birthday, the assistant just writes them. The task is done inside the chat, so no app gets the visit. That’s the zero-click part of this channel and it’s real.
Apps get named when the assistant can’t finish the job by itself:
The user wants something it can’t produce well. For example “turn this into a real song with vocals,” “make a video for her,” or “print it on a card.” That’s where tools get recommended, often as a follow-up after the lyrics.
The user asks for a tool directly. For example “best app to make a personalized birthday song.” Those prompts return named apps every time, and the list is pretty consistent between engines.
The need repeats. One birthday stays in the chat, but “I do this for every family event” pushes toward a tool.
So for a consumer app, the question isn’t whether the assistant names apps in general. It’s which of your users’ prompts fall into those three buckets, and whether you’re the app it names there.
@sarah_rolland That is a genuinely useful answer, thank you. Reading your three buckets against my own app: the first one is the product itself, "turn this into a real song with vocals" is exactly the step the chat cannot finish, and the lyrics the assistant already wrote are a fine input for it. The third one is a tab I built, birthdays and anniversaries from the phone's calendar with a song suggested for each, so the repeat case is covered inside the app but I had never thought of it as a thing an assistant would route to. The second bucket is the one I can act on this week: ask the direct question, "best app to make a personalized birthday song", in a few assistants and see whether the consistent list has me on it. I will come back with what I find rather than guess now.
@gbasovv Love that you’re testing it instead of guessing. A few things that make the manual check much more reliable:
Ask several phrasings, not one. “best app to make a personalized birthday song”, “app that turns lyrics into a song with vocals”, “how do I make a custom song for my mom”. Each one can return a different list.
Ask each one 2 or 3 times, in a new chat. Answers vary between runs, so being named once isn’t the same as being named consistently.
Note who gets named instead of you, and which sites the assistant cites. That’s the useful part: those sources are usually what you need to show up on (review sites, “best of” lists, Reddit threads).
Try the main assistants: ChatGPT, Gemini, Perplexity and Claude. They don’t agree as often as you’d expect.
Your calendar tab is a nice angle, by the way. “App that reminds me to make a song for every birthday” is exactly the kind of prompt where a tool wins over a one-off chat answer.
If doing this by hand gets tedious, that’s what Operato AI automates: the same questions, every week, across assistants, with who’s named instead of you. Either way, I’d love to hear what you find.
@sarah_rolland Thank you, that turns a vague plan into a checklist. The part I would have missed is the sources. I was only going to look at whether my app is on the list, but who sits there instead and which pages the assistant cites is the actual to do list, since those are pages a small team can do something about. Running each phrasing a few times in a fresh chat makes sense too, I had assumed the answer would be stable. I will do it across the four assistants with your three phrasings and report back here with what came out, even if the answer is that nobody names me yet.