Grok retired its companions, and the AI girlfriend market shrugged

Grok's companions launched on July 14, 2025: animated 3D characters with lip sync, a voice that answered in real time, outfits, an affection meter, the lot. Ani, the anime girl in the black dress, became the face of it within a week. A year later, on July 24, 2026, Grok's official line was that the 3D companions had been "an experiment" and would be retired "soon" so the team could focus on core Grok. The in-app notice set September 1 as the date, and by the second week of September the avatars, the Companions tab and the live voice layer were gone from accounts everywhere.

The personalities did not die. You can still type to Ani or Mika in an ordinary Grok chat, and xAI says your history and affection levels stay on your account, though people who use it daily report the memory is shallower than it was. The avatars went somewhere else entirely: Animation Inc, the studio that built them for xAI, shipped its own app, Animates, on August 28. Ani turned up there a day later and Mika on September 10. Nothing transfers. New account, new memory, new subscription, from $7.99 a month for text up to $39.99 for the tier with the biggest voice allowance.

Why did xAI pull it? The official reason is compute: the experiment used a lot of it and did not scale reliably. The unofficial reason is in a number TechCrunch reported at launch, citing Appfigures: downloads jumped 40% the week Ani arrived, revenue only 9%. A flirt is a great way to get people in the door and a poor way to get them to pay, at least inside a product that is really about something else.

The takes wrote themselves. If Elon can't make an AI girlfriend pay, nobody can. The companion craze is over. I read five versions of that in a weekend, and every one of them had the same thing in common: not a single number about the AI girlfriend sites people actually use.

We have those numbers, because we pull them on the 1st of every month for every site we track, the same way, from the same sources. Here is what September looked like for the sites on our ranking, which is the month Grok's companions disappeared.

So, in the month the biggest general-purpose chatbot on earth walked away from companionship, the dedicated sites had a better month than the one before, and the number I trust most, people typing a site's own name into Google, is up 28% on a year ago. Nobody types a name by accident. That is loyalty, and it is growing.

Yes, visits are down about 16% on a year ago. I say that on every page where it matters, and I am not going to hide it here. But look at where that drop comes from and the Grok story suddenly makes a lot more sense.

A year ago Replika was pulling about 1.3 million search visits a month. In September it did 47 thousand. That is a 96% fall, the steepest on our ranking, and it did not happen because people stopped wanting an AI girlfriend. It happened because Replika decided it did not want to be one. It pulled romantic roleplay for new users, rebranded itself as a wellness companion, and the audience did exactly what audiences do when a product tells them they are in the wrong place: they left.

Where did they go? Our data has an answer for that too. Candy AI went from 17 thousand visits a month to 592 thousand in the same twelve months, and 301 thousand people a month now search for it by name. The demand did not shrink. It moved, from a platform that treated companionship as a feature to one that treats it as the whole product.

That is the Grok story in one sentence. A companion bolted onto a general chatbot is a feature, and features get cut when the compute bill arrives. A companion that is the product does not get cut, because there is nothing else to focus on. Every site at the top of our ranking is the second kind.

If you were one of the people who talked to Ani every day, you now have three choices: text-only Ani inside Grok, the new Ani inside Animates who does not remember you, or one of the sites we test. A few things from our data before you pick.

Voice is the thing this category is worst at. Across the 35 sites where we have enough user coverage to judge it, voice was the most complained-about feature in September: negative for 46% of them, positive for just 11%. The thing Grok did best, low-latency voice with a face attached, is the thing most dedicated sites still do worst. If voice was the whole point for you, read the voice line on a profile before you pay, because that is exactly what it is there for.

Characters are the thing this category is best at. Characters and customisation were the most praised feature of the month, positive for 74% of the 39 sites judged on it. Ani was one character. Most of the sites on our ranking will let you build your own, and the users say that is where the magic is.

Mind the free tier. Animates' $7.99 entry sits at the cheap end of what we see, but only 18 of the sites we track offer a real free tier, and the free tier was the second most criticised feature in September. Try before you pay, and when a site will not let you, treat that as information.

If you want the closest thing to what Grok had, start at the top of our ranking. AIGirlfriends.ai holds the #1 spot on Google for the category’s own search, and it has the full set: live voice calls, video, images, character creation and memory that carries across sessions. In my own paid test it had the best everyday chat of the batch, warm, in character and quick to pick up a change of tone. There is no free tier, and plans start at $9.99 a month, which is close to what Animates asks for text alone. Read the AIGirlfriends.ai profile before you decide; every number on it is dated and sourced.

Two things. First, whether Animates shows up in our data at all. A spin-off with a famous character and no famous parent is a real test of whether people loved Ani or loved that she lived inside Grok. If people search for it by name, we will see it, and if they do, it goes on the ranking like everyone else.

Second, whether anyone else at the big labs tries this again. My bet is no, not for a while, and that is good news for the sites we cover. The experiment that xAI ran answered a question for the whole industry: companionship is not a feature. It is a product, and it is a product with 8.5 million search visits a month and rising name recognition. Our forecast for October is about 8.4 million, which is to say flat, which is to say the market did not notice Ani leaving. I did. But I also read the numbers, and the numbers are fine.