More buyers now ask ChatGPT, Claude, Perplexity, or Gemini for a tool before they ever open Google. So we started measuring what those models actually say when someone asks for software in a category. This week we ran a simple probe across three developer categories, and the result was less about who won and more about the shape of each race. Some categories are frozen at the top. Some are wide open. And the shape tells a founder what is even worth attempting.
We took three categories, asked one plain buyer question in each, and ran that question 5 times on each of the four models. That is 20 answers per category. We counted how many of the 20 named each company.
Here are the three categories, the top named tools, and how many of the 20 answers named each one.
20 answers per category (5 runs x 4 models: ChatGPT, Claude, Perplexity, Gemini)
Product analytics Feature flags Error monitoring
(frozen at the top) (wide open) (owned, but deep tail)
----------------------- ----------------------- -----------------------
Mixpanel 20 / 20 LaunchDarkly 20 / 20 Sentry 20 / 20
Amplitude 20 / 20 Flagsmith 17 / 20 Bugsnag 19 / 20
Heap 19 / 20 Optimizely 17 / 20 Rollbar 15 / 20
PostHog 15 / 20 Split 15 / 20 Raygun 14 / 20
--- cliff --- GrowthBook 13 / 20 LogRocket 11 / 20
Segment 7 / 20 Unleash 13 / 20 Datadog 6 / 20
Hotjar 6 / 20 Statsig 11 / 20
FullStory 5 / 20 PostHog 11 / 20
Read the three columns as three different shapes, because that is the whole point.
Mixpanel and Amplitude were named on every single answer, all four models, all five runs. Heap held at 19 of 20. PostHog came in at 15, and then there is a cliff: Segment 7, Hotjar 6, FullStory 5. The top of this category is concrete. If you are a new analytics tool, the generic "best product analytics tool" question is not a fight you win right now. The models have already settled on the top two, and asking the plain question harder will not change that.
Now look at the middle column. LaunchDarkly led at 20 of 20, the way you would expect from the incumbent. But under it the field stays alive: Flagsmith 17, Optimizely 17, Split 15, GrowthBook 13, Unleash 13, Statsig 11, PostHog 11. That is eight tools with real, repeated share in the same 20 answers. There is no cliff here. A challenger in this category genuinely gets named, because the models have not collapsed the answer down to two names. If this is your category, the move is not clever. It is to describe what you do clearly enough that a model can place you, because the slot is actually available.
Sentry owned this one at 20 of 20. Bugsnag was right behind at 19, then Rollbar 15, Raygun 14, LogRocket 11. So the top is locked, but unlike product analytics there is no cliff after the leader. Five and six tools deep still get named.
The most useful number in the whole probe is Datadog at 6 of 20 here. Datadog is everywhere in monitoring. Ask the generic question and it shows up constantly. It came in low in this probe because we asked the question the way a small team asks it, "for a small dev team", and that sub-intent quietly pushed the heavyweight down and let the leaner tools surface. The lock is real on the generic question. It breaks on the sub-intent.
The shape of your category tells you what to even attempt.
If you are in a frozen top-two category like product analytics, do not spend your energy trying to get named next to the incumbents on the generic question. You will not, and the probe shows why: the top two are named on every answer and there is a cliff behind them. The opening is in the sub-intents. The questions that carry a qualifier, "for small teams", "cheapest", "for [specific use case]", are where the lock cracks, the same way Datadog drops out the moment the question says "small dev team". A frozen category is not unwinnable. It is winnable one specific buyer question at a time, not on the category label.
If you are in an open category like feature flags, the game is different and easier. The models are still deciding, so you get named by being legible. Describe what you actually do, in the words a buyer uses, and a model can place you next to the eight names already in the mix. You do not need a trick. You need to be clear while the category is still open.
Either way, the first thing to know is which shape you are in, because it decides whether you chase the generic question or go hunting for the sub-intent where the incumbents thin out.
Pick your category. Ask one plain buyer question, run it a handful of times on ChatGPT, Claude, Perplexity, and Gemini, and count who gets named. If the same two names show up on every answer with a cliff behind them, you are frozen, and your opening is in the sub-intents. If a real field of tools keeps appearing, you are open, and your job is to describe yourself clearly enough to join it.
If you want the fast version, run your own domain through the free scan and see who AI recommends in your category and how deep the field goes: run the free scan.
In your category, is the top frozen at two names, or is the field still open?