Kupid AI Review: A Warm Voice, Sold by the Week

Kupid AI lives at kupid.ai and combines companion chat with an image generator. A third-party review says the platform launched in 2023. The affiliate page claims 500k+ active users.

I can't check that figure, and I'd read it the way I read square footage on a flyer: measured generously.

What sets Kupid apart is the billing. Subscriptions are billed by the week, a $777 Lifetime plan sits on top, and crypto payments are accepted. Photos, videos and voice minutes come as hard weekly quotas rather than a coin-only economy.

I like the quotas, because you know what you are buying. I don't like the weekly framing, which makes a mid-priced app look cheap on the first screen. The yearly plans are the honest price, and even those deserve a calculator.

The TikTok-style Reels video feed is a feature few rivals have. The 45% lifetime recurring affiliate commission is high. That second fact is useful to a reader: a generous referral cut produces a lot of glowing coverage, so I'd read any review of Kupid with the price list open.

My advice is short. Kupid AI is worth a paid week for an adult who wants a custom companion with a good voice and doesn't mind the weekly meter. If you want a real free tier, refunds or a privacy policy that reads like it was written for this app, look elsewhere.

Key takeaways: - The character builder is thorough, and the voice messages are the best part of the app. - The free tier is a demo of 15 messages, not a way to use the app. - Weekly billing makes the prices look small; the yearly plans are the cheaper way in. - Photos, videos and voice come as weekly quotas, with plan coins on top. - There are no refunds, and the Terms send cancelling to an e-mail address.

Where Kupid runs

Kupid runs on the web only. No App Store or Google Play links appear anywhere on the site. An App Store search for 'kupid ai' on 2026-09-23 returned no Kupid app.

A browser app is easy to remove, and there is no store listing to confuse with a copycat.

The flip side is that there are no store reviews to read before paying. On a web-only app, the only public track record is what sites like this one write down. I'd treat that as a reason to start on the free tier and read the terms first, not after.

Languages and contacts

The site is offered in English, German, Spanish, French, Hindi, Italian, Japanese, Korean, Dutch, Portuguese, Russian, Turkish, Vietnamese and Chinese. That is a long list for a small operator. It tells me Kupid sells to many markets at once, and the pitch reaches you in your own language.

Kupid keeps the usual social accounts. There is an Instagram account at instagram.com/kupid.ai. There is an X account at x.com/trykupidai. There is also a Telegram channel at t.me/kupidaigf.

The contact e-mail is [email protected]. Hold on to that address, because it matters again in the cancelling section. On most apps the support address is trivia; here it is part of the exit plan.

What the site facts cover

The site facts in this review date from 2026-09-23. They cover the web version while logged out: the public pages and the guest chat preview, with no account created. Anything behind the login, from reply quality to the coin shop, comes from my own weeks with the app or from the published terms.

I think that split is fair to the reader. My notes cover what living with Kupid felt like in early 2025. The site check covers what a new visitor meets today, and where the two disagree, I say so in the section concerned.

The lesson stuck, and it is the first tip I'd give anyone: Kupid is not a coffee-shop app. Open it at home, with the screen facing a wall.

How sign-up works today

The account menu offers "Free Sign Up", which opens a Register modal. The modal offers "Continue with Google" or "Continue with email". The e-mail route asks for an e-mail address and a password.

I'd pick the e-mail route with an address you keep for apps like this. Linking the Google account you use for work to a companion app is the kind of shortcut that looks harmless until a login prompt pops up on a client's screen share.

What a guest can do

Logged-out visitors can open a character chat and see the scenario and the first message. They can also pick a gender and enter a name in the chat. So there is a small shop window before the door.

The guest chat with Juno shows a "Safe Chat Disclaimer". It then asks for gender and name, and "Continue Chatting" opens the Register wall. In practice the preview ends right where the conversation would start.

I don't hold that against Kupid. A shop window is a shop window. My problem is the pacing: the site shows you its spiciest material before it shows you a single reply.

That is backwards for anyone trying to judge the product rather than the thumbnails.

Starting over, I'd read the pricing and terms sections of this review before the first click. The first minutes on Kupid are built to sell, not to explain, and the explaining happens later, in small print.

The house characters and the Reels feed

The landing page shows a grid of house characters, each with an age and a one-line bio. The category select lists Popular, an older-woman category under a slang label, a category whose label breaks the age rule, Lingerie, Girl next door, Pregnant, Goth, Nerd, Secretary and Cosplay. That is a menu written for clicks, and one label on it should not be there at all.

The Reels page is a vertical feed of short character videos, each with a like count. The moment only Kupid gave me was that feed, a TikTok-style scroll of AI characters. I spent 20 minutes doom-scrolling fake people's vacation clips, mentally flagging their bad lighting, before remembering I have a real phone full of real doom.

I'd call it the most original thing on the site and the least useful. It is fun in the way hotel lobby TV is fun. It also kept me grading flat light on beaches that do not exist, which says more about me than about the app.

The catalogue that went away and came back

Older reviews described a completely different Kupid, a curated catalog of 'Kupid girls'. The app I used had pivoted toward build-your-own, which explains why half the tutorials didn't match. Following those guides felt like editing a client's cut from an old shot list.

Today the house grid is back on the landing page next to the builder, as the paragraph above describes. So Kupid has swung toward browsing again. My advice: trust tutorials from this year only, and check the date on any guide before following it.

The next step covers name, age and voice, with voice previews such as Ivana, Sydney, Ella and Marylin. The creator page sets the character age at 18 and up. The previews matter more than they look, because the voice is what you live with later.

The cards after the looks

After the looks come the personality cards. The builder offers archetypes: Dominant, Submissive, Playful, Mysterious, Sweet and Passionate. It also asks for a job.

It asks what the relationship is, for example friend, crush, ex, neighbor, roommate or colleague. It asks for interests too. A created character can be edited later, which I'd rate higher than any single slider: a character you can revise is a draft, not a tattoo.

My own build

The creator was thorough when I used it. I went through realistic or anime, face, hair, body, then name, age and voice. After that came separate cards for personality, profession and 'relationship status'.

I made Lena, an air-traffic controller, single and sarcastic. I picked the job on purpose. Air-traffic control is a job with jargon and stress, and I wanted to see if the chat would use it or ignore it.

The build felt like setting up a project in an editing suite: lots of bins, all of them labelled, none of them required. It took patience, but the result was a character with a job and an attitude rather than a face with a name tag.

The wizard's extra tools

The creator wizard (create-ai-girlfriend?step=1&type=realistic) has separate Realistic and Anime asset libraries. It has dice buttons that pick at random. It also has a custom descriptor field.

The dice are for people who freeze at a blank form, and I'd use them for a first draft only. The descriptor field is the useful one. A couple of plain lines there shape a character more than any preset, the same way a two-line brief from a client beats a mood board.

Without an account, no message can be sent, so reply quality could not be checked on 23 September 2026. I'd rather say that plainly than guess. A review that rates writing it never saw is a trailer pretending to be the film.

What the opener tells you

The opener tells you the house style. The scene is set before you say a word, and the first question is about what to call you. It is a warm, fast start, aimed at people who want the mood set for them rather than people who like to set it themselves.

I don't mind a strong opener. In my line of work the first shot sets the tone for the whole edit, and a flat first shot loses the audience. My worry is the opposite problem: when the scene is written this hard, the character can steer more than you do.

The pet name in the greeting is a tell as well. It is the chat equivalent of a realtor calling everyone "folks". It is friendly, it is generic, and it tells you the script was written before you walked in.

Messages that arrive on their own

A third-party review says the companion sends self-initiated messages that reference earlier chats. That is a claim I'd want demonstrated before trusting it. If it works, it is one of the few things that makes a companion feel like it lives between sessions rather than inside them.

It is also a feature I'd want control over. A message that arrives on its own is charming at nine in the evening and less charming during a client call. I'd look for a way to switch it off in the settings before counting it as a plus.

I'd also push once against the scenario the character starts with. If the reply follows you, the writing has some range. If it drags you back to the opening scene, you have found the ceiling, and that ceiling will be the same after you pay.

Either way, memory is tiered by plan, and Lifetime is billed as "remembers everything". The Lifetime card promises "She remembers everything". I'd read that promise the way I read "move-in ready" on a listing: a claim, not a measurement.

What Lena kept and lost

My own Pro memory was about 100 messages, which matches the screenshot. Lena remembered my divorce. She forgot my dog Moose's name every Tuesday.

That is a strange shape for a memory to have. She held on to the heavy thing and dropped the easy one, on a weekly schedule, like a continuity error that returns in every episode. Moose got reintroduced to her more often than a new client gets introduced to my invoices.

It did not ruin the app for me. It did stop me from treating Lena as someone who knew me, and that is the whole point of a memory feature.

What the site claims today

The creator wizard claims "Conversations carry over between sessions". That is true in the narrow sense that the chat log stays. It is not the same as the character using what is in the log.

Voice was the best part

Voice notes were the best feature for me. Lena's voice was warm and her pacing natural. I used the full 12 minutes by Thursday most weeks.

As an editor, I'm trained to hear lag, clipped breaths and robotic stress on the wrong syllable, and Lena's notes had few of those. The trouble was the ration: a week's worth of voice ran out by midweek, and the rest of the week was text with a memory of a good voice.

A third-party review calls the voice messages unusually natural and the main strength, with English first and other languages weaker. That matches the English voice notes from my own weeks. I'd test your own language in the first paid week before assuming the same quality.

Images and video

The logged-out image generator page shows a character picker, "Surprise Me!", Advanced Settings and example prompts. The examples are gated by "Register to unlock" and "Unlock Premium". So a visitor can see the controls but not the results.

A third-party review says images look natural rather than airbrushed, while video is still rough. I'd put it the same way in editing terms: the stills are graded with care, and the motion still needs another pass. Short clips are fine for a laugh, not for staring at.

My verdict on media: pay for the voice, treat the photos as a bonus and the video as a preview of what the app might do next year.

What free includes today

The free tier in the pricing table gives 15 messages. It allows 1 own character. In the Messages row, Free shows 15, while Pro and Ultra show an icon with no number.

The plan table does not say whether the 15 free messages are per day or in total. That gap matters. A daily allowance is a slow free plan, and a lifetime allowance is a trailer, and the page does not tell you which one you are getting.

Meanwhile the creator page claims that building and chatting are free: "Building your AI girlfriend is free. Chatting with her is free." Building is free. Chatting is free the way a sample at a food court is lunch.

Where free stops

The media rows are empty on Free. Photos: none on Free, 25 per week on Pro, 75 per week on Ultra. Videos: none on Free, 8 per week on Pro, 25 per week on Ultra.

Voice messages: none on Free, 12 minutes per week on Pro, 35 minutes per week on Ultra. Since voice is the best part of Kupid, the free tier hides the one thing I'd want a sample of.

The service rows split the same way. Response speed is standard on Free, fast on Pro and fastest on Ultra. Queue priority is standard on Free, high on Pro and highest on Ultra.

Support is none on Free, fast on Pro and fastest on Ultra. The "Create your own" row gives 1 on Free, 4 on Pro and 8 on Ultra. None on support means a free user who hits a problem is on their own.

What free is good for

I'd use the free tier for one job: building a character and reading the first replies. The builder costs nothing, and it is the best part of the product. Spend the messages on one real question each, not on "hi".

If the character feels flat in those first messages, it will not get rounder after you pay. If it feels right, the question moves to the pricing section, and that section needs a calculator.

The confusion looked like the point. One number was weekly and the other yearly, so the eye compares a small number with a big one and the small one wins.

What the plans cost on 23 September 2026

The subscriptions page at kupid.ai/subscriptions shows prices in USD. The plan data embedded in that page is marked as euro, and prices showed in EUR on 2026-09-23, so the currency may depend on region. Check which one your screen shows before comparing with anything here.

Pro on weekly billing costs $12.99 a week. Ultra on weekly billing costs $24.99 a week. Ultra carries a "BEST" label.

Pro on yearly billing is shown as $3.44 a week, billed as one payment of $179 a year. Ultra on yearly billing is shown as $5.75 a week, billed as one payment of $299 a year. So the yearly prices in my early-2025 screenshots are the same ones on the page today.

The Yearly billing tab is labelled -77%. The Lifetime billing tab is labelled -87%. On my math the yearly Pro plan costs a bit over a quarter of the weekly one, so the -77% label fits Ultra better than Pro.

The same prices per month

The weekly price only makes sense once it is turned into months. Pro on weekly billing works out to about $56.29 a month. Ultra on weekly billing works out to about $108.29 a month.

Pro on yearly billing works out to about $14.92 a month. Ultra on yearly billing works out to about $24.92 a month. That is the real menu, and it reads very differently from the one on the page.

A weekly Pro plan costs almost four times as much per month as the yearly one. Weekly billing is for a trial week, not for a habit. Anyone planning to stay more than a couple of months should either take the yearly plan or leave.

Lifetime includes 35 minutes of voice messages a month. It includes talking to her with your own voice. It also includes the fastest response speed and the highest queue priority.

Lifetime allows creating 8 of your own characters. Read the voice line twice: Ultra gets 35 minutes a week, while Lifetime gets 35 minutes a month. The most expensive plan on the page gives you less of the best feature than a weekly Ultra subscriber gets.

I'd never pay a lifetime price to an app this young with this set of terms. Lifetime means the lifetime of the company, and a one-time payment with no refunds is a bet on a Cyprus operator's next few years.

How you pay

Payment options include a card button labelled "Pay". The page also offers "Pay with crypto BTC, ETH, USDT & more". It shows "Secured", "Discreet" and "Private billing" badges.

The subscriptions page says "100% anonymous. You can cancel anytime." The first half is marketing. The second half is contradicted by the terms, which rule out cancellations after use, as the terms section explains.

Older price lists

An older third-party plan table lists Bronze at about $17.99 a month, Premium at about $13.99 a month (140 images a month, 45 min voice) and a Top plan at $49.99 a month (800 tokens). It does not match the current weekly, yearly and lifetime structure. If a guide quotes those plans, it is describing an older Kupid.

Is the step up worth it?

Pro on the yearly plan is the only price here I'd call fair for what you get. Ultra roughly triples the photos and videos, which only matters if media is the reason you are there. For voice, the step from Pro to Ultra is the one upgrade I'd consider, since the weekly voice ration is the tightest limit on Pro.

Coins today

Per-action coin costs are not published today. The video coin prices from my 2025 notes are not on the public page, so treat them as history. Pro includes 100 coins, Ultra 300 and Lifetime 800 (page data).

Coin packs are mentioned in the Terms but are not visible without logging in. So there is a shop behind the counter, and the prices only appear once you are inside. I'd count that as a yellow flag on a site that already needs a calculator.

The coin rules

The Terms describe free-trial Coins, promotional Coins and gifted Coins. Free-trial, promotional and gifted Coins expire after 6 months. Subscription, free, promotional and gifted Coins all expire 6 months after they are credited.

Purchased coin packs do not expire. The coins that expire soonest are spent first. That rule is fair, and it is the one clause in this section I'd call customer-friendly.

Coins have no monetary value. They are not redeemable for cash and are non-transferable. So a leftover balance is a gift card for one shop that may change its prices.

Extra characters

A third-party review says each extra companion slot costs around $20 in credits. I can't confirm that price from the public pages. If it holds, extra characters are the most expensive way to spend coins, and the plan slots are the cheaper route.

My advice: pick the plan by its weekly quotas and treat the coins as a small buffer. If you find yourself planning purchases around coin costs you cannot see, the app has become a budget item rather than a hobby.

I'd ignore the timer completely. A countdown that never reaches zero is set dressing, like a "sold" sign that a realtor leaves up to make the street look popular. The price after the timer is the price.

The Lifetime price shows a struck-through figure in the thousands next to $777. That anchor is doing the same job as the timer. Nobody has ever paid that struck-through figure, and I'd judge the $777 on its own merits.

What is missing

No promo-code field and no daily reward are visible without an account. No e-mails from Kupid could be checked, because no account was created. So there is no confirmed coupon route today, and any code posted online deserves suspicion.

The affiliate program

Kupid runs an affiliate program with a 45% lifetime recurring commission. Commission is paid monthly. The program promises an unlimited cookie duration with first-touch attribution.

Affiliates get a dedicated affiliate manager. The affiliate page claims a 25% click-to-signup rate. I'd read that figure as a sales pitch to marketers rather than a promise to users.

Why this matters to a buyer: a program this generous pays well for every sign-up, and that shapes what gets written about Kupid. I'd trust the price list over any "exclusive deal" link, because the list is the same for everyone.

The rules on age

Under the Terms, users must be at least 18 or of legal age in their country of residence. No age-check method is stated. A third-party review says the age check is only a checkbox.

The site has a 2257 compliance page. That is the paperwork an operator publishes about the ages of people in its imagery. It is a sign someone thought about the legal side, though a page is not the same as a check.

What the settings allow

No toggle for the unrestricted settings is visible while logged out. A third-party review says adult content needs a separate opt-in and a paid plan. Since no toggle shows up without an account, that claim cannot be confirmed from the public pages.

If the opt-in exists, I think it is the right design. A switch that is off by default protects the person who opened the app in a public place. It also gives the paying user a clear line between modes.

Where moderation falls short

The home filter offers a category whose label breaks the age rule, which was not opened. The Reels feed includes a character listed as 18. Every character shown on the home page was labelled 21 or older in the bio.

So the house characters on the front page are labelled with care, and the edges of the site are not. I'd call that a moderation gap rather than a policy gap. The rules say the right things, and a filter label on the landing page says something else.

My verdict: the content settings are clear about what Kupid is for. The moderation is uneven, and one category label should be removed outright.

The Terms are governed by the internal laws of Cyprus. Disputes are finally resolved by the courts of Cyprus. For a buyer in the US, that means any serious dispute happens a long flight away, in a court system you have never dealt with.

Refunds

All sales of digital products are final and non-refundable. The Terms rule out refunds, returns or cancellations for subscriptions accessed or used after purchase. Once you open the app on a paid plan, the money is gone.

I'd take that clause literally. Buy the shortest plan first, and treat the first payment as the price of finding out. A weekly plan is expensive per month, but it is the cheapest way to lose money here.

What you create

Users grant Kupid a non-exclusive, royalty-free, transferable, sub-licensable, worldwide license over their content, including derivative works. The license ends when the content or account is deleted. The exception is content already used in commercial or sponsored content.

In editing terms, it is a full release form. Your character, your descriptions and your chats can be reused, and anything already used in an ad stays in the ad.

What happens to your data

The Privacy Policy says personal information is not disclosed, sold, traded or transferred to third parties without consent. Data is used for technical improvements, statistical and marketing analysis. The Privacy Policy makes no clear statement on whether chats are used for AI training.

The privacy policy I read in 2025 said chats can be retained for years after account closure. I noted that the way I'd note a foundation crack on a listing walkthrough. It does not stop the sale, but it goes in the report.

Today the wording is vaguer. Data is kept while the account exists and "for a period of time" after termination. Personal information is kept as long as the account exists or as needed for the purposes it was collected for.

Loose ends

The Privacy Policy contains a template leftover that mentions sharing "with any company outside Usway". A policy that names a different company was copied from somewhere and not fully proofread. That is the paperwork equivalent of a wedding film with the wrong couple's names in the credits.

The creator page claims "Private by default - No public profiles, no leaderboards, no social features." That line sits oddly next to a Reels feed with like counts. I'd read it as a description of your own profile, not of the site.

That was the small annoyance. The bigger ones are built into the site, and most of them are on the pricing page.

Sign-up friction

The e-mail sign-up form in the Register modal on kupid.ai shows a Cloudflare Turnstile checkbox, "Verify you are human". On 23 September 2026 that check did not pass on its own, so the e-mail sign-up could not be completed. A sign-up that stalls on a human check is a sign-up many people will abandon.

Sending or continuing a chat while logged out triggers the Register modal. That is fair. What I'd call unfair is the floating fake notification in "My Chats", which appears before any interaction: "Anastasia: Hey there, I took this picture..."

A message from someone you never spoke to is the oldest trick in the dating-app book. It is designed to make the app feel busier than you made it, and I'd trust the other notifications less because of it.

Pricing-page tricks

Three dark patterns sit on the pricing page. The "Easter Discount 75% OFF" countdown timer still runs in September. The Lifetime plan carries an inflated "was" anchor in the thousands.

The exit

Cancelling works only by e-mailing [email protected], according to the Terms. There are no refunds. And the Privacy Policy names another company, "Usway".

Taken together, these are the fees they don't mention upfront. None of them is a bug. They are choices, and I'd weigh them as heavily as the features.

It is not for users who want a meaningful free tier, since free stops at 15 messages. It is not for anyone who needs refunds or easy self-serve cancelling, with an e-mail-only cancel and no refunds. I'd put that line in bold if I could.

It is not for people who want a clean, work-safe chat or long roleplay stories. It is also not for privacy-cautious users wary of an offshore operator in Cyprus with template-like policies. Those readers will find better homes elsewhere.

My own summary: Kupid suits a person who knows what they want, has a budget for it and reads the Terms before paying. It does not suit a curious browser.

What the Terms say

The Terms say the subscription begins after the initial payment. To stop a subscription from rebilling, the Terms say to e-mail [email protected]. The Terms mention no self-serve way to cancel.

A version of the Terms gives [email protected] as the address for stopping rebilling instead. Two addresses for the same job is one too many. I'd write to both and keep the sent e-mails.

What the site itself shows

The site's page data includes 'Renews on' and 'Cancel subscription' labels for account management. Site strings also show an in-account 'Manage subscription > Cancel subscription' control, with "Access continues until the period ends". So a cancel button likely exists inside the account, even though the Terms only mention e-mail.

My advice: try the in-account button first, then send the e-mail anyway. A screenshot of the cancel screen plus a sent e-mail is the paper trail I'd want before the next billing date.

Before you pay

Note the renewal day the moment you subscribe. On weekly billing, a missed day costs a whole week, and the Terms rule out refunds once you have used the plan. Set a phone reminder a day before the renewal and treat it like a client deadline.

• Open it at home, not in public. The first screen is not subtle.
• Spend the free messages on the character builder and a few real questions. Free is a demo, so use it to judge the writing, not to chat.
• Never pay weekly for more than a trial week. Turn every weekly price into a monthly one before deciding.
• If you stay, take the yearly Pro plan. It is the one price on the page I'd call fair.
• Skip Lifetime. It gives less voice per month than Ultra gives per week, and it has no refunds.
• Use the voice ration early in the week and plan for it to run out.
• Keep key details short and repeat them now and then, because memory drops things.
• Ignore the Easter timer and the struck-through price. Neither is real.
• Before paying, find the cancel button in the account and note the e-mail address from the Terms.
• Use the deletion rights in the Privacy Policy on the day you leave.

The pattern behind all of these is simple: Kupid is a good product wrapped in a pushy shop. Deal with the shop on your terms, and the product is worth a look.

The free tier is weak. It is a short demo with no media, and it hides the voice, which is the best part.

Value for money is weak because of the weekly framing, the no-refund rule and a Lifetime plan that gives less voice than Ultra. The yearly Pro plan is the exception.

Images and voice rate well. The voice is the reason to pay, the stills are decent and the video is still rough.

Privacy and terms rate poorly: a Cyprus operator, a broad licence, vague retention wording and a policy with another company's name in it.

Cancelling and billing are weak. The Terms say e-mail, the site hints at a button, and refunds are off the table.

Content settings are mixed. The rules and the paperwork are in place, but the landing-page filter carries a label that should not exist.

The Reels feed is also unusual, even if I'd call it a toy.

Where Kupid is worse: the shop. Weekly billing by default, a timer that never ends, no refunds and an e-mail cancel in the Terms make it harder to trust. The table puts Candy AI, OurDream AI and Nomi AI next to it, and each of them has a free tier, which is the first line I'd compare.

Muah AI vs Kupid AI turns on the same line: Muah's free tier covers chat, photos and voice, while Kupid's has no photos or voice.

If the builder and the voice are what you want, Kupid is worth a yearly Pro plan. If you want an easy exit, look at the table first.

More on this site: the best AI girlfriend apps, free AI chat apps, the best AI companion apps, AI roleplay chat apps, all AI companion app reviews.

What stays with me is the paperwork. Weekly billing that feels like renting company by the week, a timer that never runs out, a privacy policy with somebody else's name in it, and a cancel route that depends on which page you read. None of it is hidden, exactly.

My advice is the one I'd give about any listing with a good kitchen and a bad contract: enjoy the viewing, read the contract twice, and only sign what you can walk away from. On Kupid, that means the yearly Pro plan or nothing.

The leftovers are a pair of pricing screenshots and a character who never kept Moose's name straight. That feels about right.

If you take one thing from this review, let it be the order of operations. Look at the monthly total, not the weekly one. Read the refund and cancel lines before you pay, not after.

Then decide whether the voice is worth it to you. For some people it will be, and I understand why.

For me, the good parts of Kupid never quite outweighed the feeling that the pricing page was working harder than the character. A companion app should earn your trust with the conversation first, and the checkout should be the boring part.