AI in pregnancy apps, ranked: what the AI actually sees, and where it runs

What the AI actually sees, and where it runs

Most pregnancy-app AI is a cloud chatbot that sees your typed question, and sometimes your tracked data, with the model behind it unnamed. Flo and Glow market in-app AI but disclose no base model. Ovia, What to Expect, and BabyCenter ship no user-facing AI at all. myCocoon runs its AI on your phone and sends a structured briefing to the cloud only when you ask.

The roundups that rank pregnancy apps treat AI as a checkbox feature. None of them ask the two questions that actually decide your privacy: what does the AI see, and where does it run? This post is the rubric the listicles skip. For the narrative version of how myCocoon arrived at its own answer, see AI in pregnancy apps: what is actually shipping. This one is the per-app ledger.

One app in this ranking is built the other way around. myCocoon reads several Apple Health signals · your sleep, your resting heart rate, your steps · and watches each one against your own rolling baseline, raising a quiet nudge when any single signal drifts: a run of short nights, a resting heart rate running high for you, an unusually active day. A separate rule notices when your logged mood trends down for days at a stretch. It is all plain rules computed on the phone, no model involved; reading those signals jointly, as one correlated picture, is on the roadmap, not shipped. Its everyday on-device work · journal cleanup, photo captions, story narration · runs on Apple’s on-device model, so the core works with no cloud at all. And it carries no ads, no data sale, and no advertising trackers. The honest trade sits right next to that: it is iPhone-only on iOS 26, with no Android version, no peer community, and a smaller library than Flo or What to Expect. If breadth or a community feed is your deciding factor, one of the incumbents is the rational pick, and this ranking says so.

Last reviewed: 23 June 2026.

How we ranked it: four questions, not one

We looked at the AI claims of the nine biggest pregnancy apps, Flo, Ovia, What to Expect, BabyCenter, Glow, Pregnancy+, The Bump, Peanut, and myCocoon, plus the wave of standalone AI chatbot apps. We scored each on four axes, not on a vibe. “Smartest answer” is the wrong contest. “Where does my sentence go” is the right one.

• What the AI does. Is there a real user-facing AI feature, or is “AI” a marketing line on the App Store listing?
• Is the model vendor disclosed? Can you find out which model is answering you, by name, before you type a health question into it?
• What goes to the cloud. Your raw question? Your tracked symptoms and vitals? Nothing? Is the payload itemized anywhere you can read it?
• On-device or cloud. Does the model run on your phone’s own chip, where the text never reaches a server, or on a remote server you cannot see?

The fourth axis is the one nobody else ranks on, and it is the one that determines the first three. A model that runs on your phone cannot leak a payload it never sends.

AI in pregnancy apps, side by side

The table below scores the apps that make an AI claim, plus the standalone chatbots, on the four axes. Apps that disclose no user-facing AI are grouped in one row.

Pregnancy+, The Bump, and Peanut round out the nine and disclose no user-facing AI either, so they sit with the “said no” group below.

Flo: a named assistant, an unnamed model

Flo ships the most prominent assistant in the category. Ask Flo is a generative health assistant gated behind the Premium tier, and a June 2025 Databricks press release confirms Flo fine-tunes its own large language models on Databricks (Databricks, 2025). What that press release does not do, and what Flo does not do anywhere a reader would look, is name the base model underneath, or itemize what your questions and your tracked data send to the cloud. You are typing health questions into an assistant whose model you cannot identify.

Flo also has the longest public data record in the category. It settled with the FTC in 2021 over sharing health data, including pregnancy status, with Facebook and Google analytics; the order required affirmative consent, user notification, and independent review (FTC, 2021). More recently, an August 2025 California jury found Meta liable under the state’s Invasion of Privacy Act for collecting Flo users’ data; damages have not yet been set (The Record, 2025). The full history belongs in the privacy comparison; we cover it in pregnancy app privacy: where your data actually goes. For AI specifically, the relevant point is narrower. The company that has been least careful with what leaves your phone is also the one running the biggest cloud assistant, on a model it will not name.

Glow and the wave of AI chatbots

Glow markets AI more loudly than it documents it. Its Nurture app advertises that it “uses AI to provide personalized insights and advice,” and the App Store listing has carried the alternate title “Glow Nurture AI Pregnancy App” (Glow Nurture, App Store). What Glow does not disclose is which model produces those insights, or how your data is handled on the way there. Glow also paid a $250,000 California settlement over failing to safeguard users’ health information, after Consumer Reports first flagged the app’s privacy flaws (Consumer Reports, 2016). That record is worth carrying into any “AI insights” claim from the same company.

Below the named apps sits a flood of standalone AI chatbots. Apps like Femia, ProMom, and PregnancyAI.app lead with a conversational “AI midwife” as the whole product. These are cloud chatbots by construction: your question goes to a server, an answer comes back, and the disclosure of which model answered, what it retained, and whether it trains on your prompts is usually thin to absent. The chat reads warm. The plumbing behind it is the part you cannot see, and it is the part that matters at 11 p.m. with a real symptom in the input box.

Ovia, What to Expect, and BabyCenter: the apps that said no

Three of the biggest pregnancy apps ship no user-facing AI, and that is a defensible answer, not a gap. Ovia, owned by Labcorp since 2021 (Labcorp, 2021), discloses no AI in its consumer app and describes none in its privacy policy; its coaching is staffed by human registered nurses. One caution: Labcorp markets a separate generative-AI tool for clinicians called Test Finder. That is a clinician product, not part of the Ovia pregnancy app, and the two should not be conflated. Ovia’s own data record is a different conversation, including employer data-sharing reporting (The Washington Post, 2019) and a Mozilla warning (Mozilla, Privacy Not Included), but on the AI axis it sends nothing to a model because it runs none.

What to Expect, owned by Everyday Health, discloses no user-facing AI. BabyCenter, owned by the same group, has no verified user-facing AI assistant; the only AI it has publicly adopted is a back-office loyalty-marketing system for its rewards programme, which is a marketing tool, not a health assistant. Both apps have their own ad-supported data-sharing concerns, flagged by Mozilla’s Privacy Not Included audit of pregnancy trackers (Mozilla Foundation, 2022). On AI, though, they are honest by omission: no chatbot, no model, no payload. There is nothing to disclose because there is nothing running.

myCocoon: on-device first, cloud only when you ask

In one line: myCocoon keeps the Apple Health signals it reads in an on-device store that never syncs to iCloud or any server, runs its everyday AI on Apple’s on-device model so the core needs no cloud, and sends a structured briefing to Google’s Gemini only if you opt in. No ads, no data sale. It is iPhone-only on iOS 26, newer and smaller than Flo or What to Expect, with no Android version and no community feed.

myCocoon ships AI, and it would be dishonest to pretend otherwise in a post that holds other apps to that standard. The difference is architectural. Everyday work, cleaning up a voice-dictated journal entry, drafting a gentle prompt, narrating the chapters of your pregnancy story, runs on Apple’s on-device model inside iOS 26. That text never reaches a server. The most distinctive thing myCocoon does is not a chatbot trick at all. It reads your sleep, your resting heart rate, and your steps from Apple Health and watches each one against your own rolling baseline, raising a quiet nudge when any single signal drifts from your normal · a run of short nights, a resting heart rate running high for you, an unusually active day. A separate rule watches your logged mood for a sustained downward run. The day’s readings are gathered into one on-device digest; reading them jointly, as a single correlated picture, is on the roadmap, not shipped. No cloud assistant in this table does per-signal, baseline-aware nudges from passive Health data at all, because none of them hold those signals to begin with. For the longer treatment, see how a pregnancy app reads what is already on your phone.

Cloud AI exists, behind a switch, off by default. Turn it on and your question goes to Google’s Gemini with a structured briefing attached: your week and trimester, your first name and your baby’s name if you have set them, profile basics like age or blood type when they apply, your recent symptom names and how often you logged them, your mood and sleep trends, and a daily read of your sleep hours, your resting heart rate, and your steps computed on the phone. Those are the fields the in-app privacy dashboard lists in plain words before you ever switch it on. What is never sent, in either mode, is your journal text, your photos, your voice recordings as audio, and your full Health-app history behind that one-day read. The honest line is not “nothing leaves,” and it is not “only the numbers leave.” It is “a defined, readable briefing leaves, your name and baby’s name included, and only when you ask.” That is still a sharper answer than any incumbent gives.

Two architectural facts sit underneath all of this, and no competitor in the table discloses either. First, the data the iPhone Health app shares with myCocoon lives in a separate on-device database configured never to sync, not to your iCloud and not to any server (in the code, a second store set to cloudKitDatabase: .none). Second, myCocoon watches each of those signals · sleep, resting heart rate, steps · against your own rolling baseline and raises a quiet nudge when any one of them drifts, with your logged mood tracked separately for a sustained downward run. These are independent, per-signal rules, computed on the phone against your own data, with nothing sent. Reading the signals jointly, as one correlated picture, is in design, not shipped. The cloud chatbots do none of this, because they never hold the signals in the first place.

That kind of disclosure is what the incumbents skip. This design is self-asserted and documented inside the app, not independently audited. myCocoon is also new, iOS 26 only, and far smaller than Flo or BabyCenter, with no community feed and no Android app. It is available on the App Store today. If breadth and a peer community are what you want, it is not the rational pick, and our comparison with Flo says so plainly.

On-device or cloud: the line that actually matters

Here is the contrast no other pregnancy-app content draws. The interesting question is not which AI gives the warmest answer. It is where the model runs, because that single fact decides what a server ever sees. A model on your phone’s chip cannot leak a payload it never sends. A model on a remote server sees whatever the app chooses to transmit, and most apps do not tell you what that is.

Sort the category on that axis and it splits cleanly. On one side, the cloud chatbots: Flo, Glow, and the standalone “AI midwife” apps, where your question, and sometimes your tracked data, leaves your phone for a model you cannot name. On the other side, the apps with no AI at all: Ovia, What to Expect, BabyCenter, which send nothing because they run nothing. myCocoon sits in a third position that the category treats as impossible: AI that runs on-device by default, with a cloud path that is opt-in and carries a structured briefing rather than your raw Health history.

That third position is the one the roundups miss. The assumption running through every pregnancy-app comparison is that you trade privacy for intelligence: either a smart cloud chatbot that sees everything, or a dumb private tracker that sees nothing. The architecture myCocoon uses, an on-device model plus an opt-in cloud path with a defined, readable payload, is a third option, and it is the one the listicles have not noticed exists. The catch, stated plainly, is that you have to take the on-device claim partly on trust until a third party audits it. The incumbents ask for the same trust and disclose less.

Common questions

What is the best AI pregnancy app? It depends on what you want the AI to see. If you want the largest in-app assistant available on iOS and Android today, Flo ships one, behind Premium, on a model it does not name. If you want AI whose core runs on your phone, that keeps your raw Health history in a local-only store, and that sends a structured briefing to the cloud only when you opt in, never your journal text, photos, or voice audio, myCocoon is built that way, though it is iOS-only and new. There is no single winner. The honest axis is where the AI runs.

Which pregnancy apps actually use AI? Of the nine biggest pregnancy apps, two market a user-facing generative assistant: Flo’s Ask Flo and Glow’s AI insights. A wave of standalone apps, Femia, ProMom, and PregnancyAI.app among them, are cloud chatbots first. Ovia, What to Expect, BabyCenter, Pregnancy+, The Bump, and Peanut disclose no user-facing AI. myCocoon runs AI on-device and uses an opt-in cloud model for questions.

Does Flo use AI, and what model is it? Yes. Flo ships an in-app assistant called Ask Flo behind its Premium tier. A June 2025 Databricks press release confirms Flo fine-tunes its own large language models on Databricks, but the base model underneath is not publicly named, and Flo does not itemize what your questions and tracked data send to the cloud. You are talking to an unnamed model.

Is it safe to ask a pregnancy chatbot health questions? It depends on where your words go. A cloud chatbot sees your typed question, and sometimes your tracked data, and routes it to a server. The consumer versions of general assistants are not covered by health-data contracts. Before you type, the useful questions are which model is answering, what is sent, how long it is kept, and whether it trains future models. Most pregnancy apps do not tell you.

Does Ovia have an AI chatbot? No user-facing AI is disclosed in the Ovia consumer app, and its privacy policy describes no AI or machine learning. Ovia’s coaching is staffed by human registered nurses. Its owner Labcorp markets a separate generative-AI tool for clinicians called Test Finder, which is not part of the Ovia pregnancy app. Do not conflate the two.

What does on-device AI mean, and is myCocoon’s claim verified? On-device means the model runs on your phone’s own chip, so the text never reaches a server. myCocoon uses Apple’s on-device model for everyday work and an opt-in cloud model for questions. When you opt in, your question goes to Google’s Gemini with a structured briefing: your week and trimester, your first name and baby’s name if set, recent symptom names and counts, mood and sleep trends, and a daily read of your sleep hours, resting heart rate, and steps. It never sends your journal text, photos, or voice audio. That design is self-asserted and documented in the app, not independently audited. myCocoon is also new, iOS 26 only, and far smaller than the incumbents.

Is there a private pregnancy app that keeps health data on your phone? Yes. myCocoon keeps the data it reads from Apple Health in a separate on-device store set never to sync, not to iCloud and not to any company server. Its everyday on-device work · journal cleanup, photo captions, and story narration · runs on Apple’s on-device model, so the core works with no cloud at all. Cloud AI is off until you switch it on. Turn it on and it sends a structured briefing to Google’s Gemini: your week and trimester, your first name and baby’s name if set, recent symptom names, mood and sleep trends, and a daily read of your sleep hours, resting heart rate, and steps. It never sends your journal text, photos, voice audio, or full Health history. The honest trade: it is iPhone-only on iOS 26, with no Android version, no peer community, and a smaller library than Flo or What to Expect.

Bottom line

Ask where the AI runs before you ask how smart it sounds. Most pregnancy-app AI is a cloud chatbot on a model you cannot name, sending a payload you cannot read. Flo and Glow market AI and disclose no model. Ovia, What to Expect, and BabyCenter ship none, which is a clean answer. myCocoon runs AI on-device and sends a structured briefing to the cloud only when you opt in, though that claim is young and self-asserted rather than audited.

If you are choosing on AI, do not chase the warmest chat bubble. Choose the app whose answer to “where does my sentence go” you can actually read. For the wider field, see the best pregnancy apps of 2026 and where each app’s data actually goes.

If your answer to “where does my sentence go” is “I want to read it before I send it,” myCocoon is built for exactly that, and it is on the App Store today, for iPhone on iOS 26. Stay clear-eyed about the trade: it is new, iOS-only, with no Android app and no community feed, so if breadth or a peer community is what you need, Flo is the rational pick and our comparison says so. If a documented, on-device-first payload is what you need, download myCocoon and open the privacy dashboard before you ever turn Cloud AI on.

§