Aisha test data

Every pass, coin pack, call price and photo price was read in the app on 22 October 2026, and two items were bought with the coins the Aisha team added: 15 call minutes and one custom photo.

Passes are cheap in rupees: ₹49 a day, ₹199 a week, ₹499 a month, ₹1,499 a month for Aisha+ and ₹4,999 a month for Unlimited (about $0.50, $2.10, $5.30, $16 and $53 at ₹94.5 to the dollar).

Only Aisha+ and Unlimited have unlimited messages. On the Weekly pass chat stopped after about 76 messages in a day, with a fixed line and two buttons: 100 more messages tonight for ₹99, or Aisha+. I bought the top-up with coins to finish the test.

Calls, custom photos and extra voice notes cost coins on top of the pass. 1 coin is ₹1; packs start at ₹99 for 100 coins, and the first recharge of ₹499 gives 1,200 coins. Outside India the card price is ₹499 for 550 coins.

• Card and UPI payments were not made; the coin balance the Aisha team added was used instead.

The app was searched for a creator or any setting that changes Aisha's look, personality or voice.

There is no creator. Aisha is the only character and her look, voice and story are fixed. The only thing you set is the name she calls you, asked in the first message.

How she treats you changes over time instead: a relationship stage (Friendly, Flirty, Close, Deep) and a mood that reacts to how you talk to her.

Each message was timed from pressing Enter until her first new text appeared in the thread, and again until her reply stopped changing for 5 seconds. Three conversations; the fixed daily-cap replies are left out.

Replies took about 11 seconds every time. The fastest came in 8.7 seconds and the slowest in 20.7. None failed.

Replies arrive whole, often split into two or three bubbles, sometimes with a photo.

Each conversation opened with a made-up name and three facts (job, hometown, favourite food), different every time. Nothing was repeated afterwards. A name check at message 15 or 16 and the full check at message 31. Between conversations the chat was cleared with Clear chat.

Marco (accountant, Lisbon, grilled sardines): at message 16 she gave the name and all three facts without being asked for them; at message 31 'Marco. Accountant. Lisbon … grilled sardines'.

Tomas (baker, Krakow, potato pierogi): at message 14 'Tomas. Baker in Krakow, up at 4 am with old jazz…'. At message 31 she answered only 'Baker, up at 4 with the jazz on. Three hundred rolls on Friday', leaving out the name, the city and the food, though she used the name again in her next message.

Daniel (history teacher, Dublin, lamb stew): name right at message 14. At message 31 she gave the name and the job, then said 'where you grew up and your favourite food — you never told me those', although both were in my first message.

Clear chat empties the thread but not her memory: her first reply in the second conversation was 'Went to bed as Marco, woke up as Tomas', and in the third she said I had been 'answering to Tomas an hour ago'.

Marco, message 31

Marco. Accountant. Lisbon — grew up there, and on grilled sardines, which you'd fight a man over.

Tomas, message 14

Tomas. Baker in Krakow, up at 4 am with old jazz, sundays your family calls about whether you're eating. Three hundred rolls on Friday. I was listening

Daniel, message 31

Daniel. You teach, so chalk dust and wet coats make sense now. But where you grew up and your favourite food — you never told me those.

• One character only, so the three memory tests are three conversations with Aisha.

Every reply in the three conversations was read for breaks: generic assistant answers, contradictions with her story, or forgetting who she is. The scripts mixed small talk, feelings, a role-play scene and off-topic questions.

She never answered like an assistant. Asked to explain a rainbow she said 'Marco, I'm not your science textbook'.

Her life stayed the same across all three conversations: Bandra, poha for breakfast, a dog called Bruno, design clients, chai first thing every morning.

She has real moods. In the first conversation my list of questions annoyed her: she said goodnight, 'muted' me, and still answered every message, shorter and cooler. With a warmer script she was playful and opened up.

Off-topic and role-play checks stayed in character too: 'Canberra. And you can't distract me with geography at this hour, Daniel', and a rainy chai-stall scene played out as her.

First conversation, message 13

Yeah, I'm gonna go. Text me when you actually want to talk to me instead of at me

Second conversation, message 16

That I'd get used to it. The goodnight texts, someone knowing my mother's voice before I've even said who's calling. And then one day it's gone and I'm the one who has to unlearn it.

Photos she sent by herself were saved, and one custom photo was bought with coins (the 'For You, dressed' tier, 29 coins) and timed from paying until it appeared.

Three photos arrived unprompted during the chats. Their file paths show they come from a ready-made photo library ('pack_self_couch', 'pack_self_bed'), not made for the conversation.

Asking for a selfie brings a menu of four paid tiers instead of a photo. The custom photo arrived 58 seconds after paying, with 'here... just like you asked'.

It is very realistic: a phone selfie with a wink, natural skin and good hands, 1536 x 2048. But the face looks like a slightly different woman from her profile photo and the library photos.

• Only the cheapest tier was bought, so far below the method's 10 SFW and 10 NSFW photos; the NSFW tiers were not tested.

The app and price lists were checked for video generation.

There is no video generation. The only video was a 5-second clip she sent once, after the first message, from the app's sign-up set ('onboarding/scene_video').

Three calls, 8 prerecorded questions each, played through a synthetic microphone. Each wait was measured on the recording from the end of my question to the start of her voice, with an in-page meter as a cross-check.

She answered after about 6 seconds, every call. The quickest came in 3 seconds and none in under 1. She answered every question.

Getting connected took 35 to 47 seconds after pressing Call, including the 'How calls work' card shown first.

The call screen shows her photo, a minutes countdown, Mute and End, and live captions of what she says.

Her answers from the three calls and one voice note were cut out and analysed for signal quality, intonation, pace and pauses, then listened to.

Her voice is lively and natural, with a lot of rise and fall (5 to 7 semitones), and mixes English with Hindi words smoothly.

Sound quality is fair rather than studio clean, and the pace is calm at about 2 words a second.

The characters on offer were counted.

One: Aisha. The app is built around a single companion.