Creator: Enrico Tartarotti · Published: 2026-07-08 · Length: 13:15 · Watch on YouTube

Notifications shape your behaviour more than any algorithm, more than endless scrolling, and almost none of them are an accident. (from the video description)

1) Detailed outline

[0:00] Cold open: the Duolingo "giving up on you" notification

• Notifications are "the most secretly influential piece of tech you use every day." They affect behavior more than endless scrolling or any algorithm, and they're literally built to change your behavior.
• Example on screen: Duolingo, "These reminders don't seem to be working, so we'll stop sending them for now." It looks innocent, like the app is giving up on you, but it's hyper-engineered:
  • Sent 23.5 hours after you last opened the app, but before 10:00 p.m. so you can still act on it.
  • It plays on a specific psychological principle (revealed later as loss aversion).
  • Result shown on screen: cancellations down 40% and daily active users up 4.5x.
• Claim: notification design ended up mattering more to the product than any change to the user interface. Designers obsess over buttons and flows, but the products that really understand notifications don't use them to get a click right now. They use them to build a habit.
• The video is organized into three layers.

Layer 1: The engineering of the notification itself

[1:00] Rebuilding the Google Photos memories notification

• Starting point: a generic daily Google Photos "memories" notification ("Enrico, view some memories"). Enrico calls it bad and sets out to turn it into one of the most effective, beloved notifications people get.
• Myth: the secret is personalization (putting your name in it). That doesn't work, as you know from thousands of emails with your name in them.
• Real principle: specificity, not personalization. Add a photo preview and make it about something that happened on this day some years ago (his trip with friends four years earlier).
• How to detect a meaningful memory:
  1. Photo count per day: trips, parties and great days tend to have the most photos, so rank years by that.
  2. Location: Google knows where you live, and EXIF data stored with the photos shows when pictures were taken somewhere else, which suggests a trip.
  3. Visual analysis tags: Google Photos already tags images (for search), so it can filter out days of boring document photos and pick a theme like skiing or mountains.
  4. Run the tag through a basic large language model to make the wording playful.
• "The more specific the detail, the harder it is to look past, but also the more complex it is to build." Doing this for millions of people daily requires historical photo analysis, visual recognition, personal data and an AI layer.

[3:00] The curiosity gap (Loom and Hinge)

• Loom says someone viewed your video but deliberately doesn't say who. That breaks the specificity rule on purpose to create a curiosity gap: "Did my boss see this? I need to know."
• Hinge: "Nadia liked you." Without the name, the question is "Who liked me?" With it, the question becomes "Is Nadia hot?" (Running joke: she was, he told her about notifications, and she blocked him. The description's sign-off asks fans to type "Nadia, unblock me.")

[3:45] Loss aversion

• Curiosity isn't the same as action. Duolingo doesn't say "come learn Spanish." It says "don't lose your streak."
• Loss aversion: we're more afraid of losing what we already have than of gaining something new. Many notifications aren't offering anything; they threaten to take something away, such as the daily window to post (a BeReal-style prompt).

[4:05] Timing is really context

• Duolingo doesn't remind you at a fixed time. It waits 23.5 hours after your last session, because that's when you were free yesterday, and caps it at 10 p.m. so it doesn't arrive when you're asleep.
• Getting the timing right, and nothing else, lifted engagement by around 60% (the captions read "lift tabs," most likely "taps").
• Even messaging notifications are complex: Enrico shows Slack's flowchart for deciding whether to notify you (Do Not Disturb, channel and global preferences, @channel/@here mentions, threads, mute, highlight words, presence, desktop vs. mobile).

[4:50] Why this matters: the only push lever

• Notifications are the most powerful tool products have to build habits. BeReal only exists because of its notification, and Duolingo probably too.
• Everything else requires you to act (open an app, visit a site). Notifications are the only push lever a product has. Do them badly and users uninstall; learn the principles and you can shape behavior at scale.

Layer 2: Why notifications became so powerful

[5:15] From the BlackBerry light to iOS 3

• The BlackBerry's blinking notification LED was so effective at making people check their phones that it gave rise to the term "CrackBerry."
• Back then a notification meant one thing: do something right now, within the next minute.
• iOS 3 (iPhone OS 3, the first version with push notifications) used a full-screen, blocking pop-up with no Notification Center. That made sense because notifications were for things like incoming messages.

[6:00] Act now vs. deal with it later

• A study of over 200 million real notifications found many are still the "one-minute" kind (SMS, WhatsApp), tapped about 15 seconds after being seen on average.
• At the other end is email: technically similar (it arrives instantly), but the inbox means it sits and piles up. Spam filters, tags, labels and folders all try to manage that pile. Email is meant to be dealt with later, usually within a day, if it doesn't sit there forever.
• For years there were only those two kinds: act now, or deal with it later.

[6:55] The "next hour" zone and repetition

• The same study found only 26% of notifications at the top of the screen got people to stop and tap. There's a middle zone: things you won't do now, or tomorrow, but maybe in the next hour.
• Enrico calls this one of the most powerful ways tech gets you to do what it wants, because it's about repetition, not a single notification.
• Duolingo data: after a 7-day streak, you're 2.4x more likely to come back tomorrow on your own. Eventually you open it not because it pinged you but because not doing it feels wrong.
• This is the premise of the book Hooked: How to Build Habit-Forming Products (Nir Eyal).

[7:45] What happens if you turn them all off?

• In a study (on screen: "Productive, Anxious, Lonely: 24 Hours Without Push Notifications" by Martin Pielot and Luz Rello), people disabled all notifications for a day. They got more done, but felt more anxious, oddly lonely, worried about missing out and about letting others down. Enrico says every participant turned them back on, and switching them off didn't stop people checking their phones.
• Pull vs. push: scrolling, however addictive, needs you to get bored and open the app yourself (pull). A notification is push: it reaches into your pocket and creates the moment on the product's schedule, not yours.
• Not inherently good or evil: it can build good habits (language learning, workouts, sports), but the same habit loop can be used to monetize you.
• His favorite: the Google Photos memory notification, which lets him relive moments and send memories to friends, even though Google Photos is built to push you into a storage subscription and lock you into the ecosystem.

[9:00] Sponsor segment: Figma's design agent

• Demo in a file of every notification screen, state and variant for an app: the agent renames the app to "Flask" on every screen, switches everything to dark mode with a red icon, and builds a new empty-state screen from the file's existing components and spacing.
• Points made: you can tag any asset or component for the agent to use; output is a real design inside your Figma file, not a generic wireframe in a separate environment, so you can keep editing by hand. Open beta at the time of the video.

Layer 3: AI between the product and you

[10:20] Summaries undo all of Layer 1

• Until now the flow was simple: product fires notification, device shows it, you read it.
• Apple's notification summaries (Apple Intelligence) take the carefully engineered notification and squash it back into a boring, unclickable one, "exactly the type that doesn't work."
• Apple and Google have both shipped versions (on Android it depends on the manufacturer). Every notification can now go through an on-device model that can reword, reorder and shorten it, and possibly decide you don't need to see it at all. Google holds a patent on a system that decides whether and when to deliver a notification.

[11:15] Real damage: the BBC summary

• Apple's AI summarized a BBC News notification into a false claim; on screen: "Luigi Mangione shoots himself." The BBC never reported that; the AI invented it and showed it under the BBC logo on millions of home screens.
• Apple scrambled and switched summaries off for news apps, which shows how much power this layer already has.

[11:40] AI on both ends and llms.txt

• In email it's already here: marketing, sales and coworkers use AI to write emails and docs, and recipients use AI to summarize and filter them. Whole products exist just to sit between you and your messages.
• Enrico shows the landing page for Flask (his product), made for humans, and its llms.txt file, meant to be read by AI agents. Websites are filling up with pages and files written for agents rather than people.
• His prediction: notifications may need invisible metadata that guides your "agent secretary" on whether and where to show a notification, in what context, and whether it may change the wording.
• Endgame: Layer 1 becomes a relic. It won't be about getting humans to build habits but about convincing their agents to show them something. We'll move from building products for people to building products for people's AI agents.

[12:55] Outro

• Teases another video about a hidden set of decisions that are neither design nor engineering but are "the least talked about and most important part of the tech you use every day."

Key takeaways

• Specificity beats personalization. A name isn't enough; the specific, relevant detail is what's hard to ignore, and it takes real data and engineering to produce.
• Psychology does the heavy lifting: curiosity gaps (Loom, Hinge) and loss aversion (streaks).
• Timing is context, not a clock: 23.5 hours after your last session, before 10 p.m.
• Notifications are the only push lever, which is why they build habits better than any UI change.
• AI now sits in the middle, rewriting and filtering notifications, so products may soon design for your agent instead of you.

2) Things mentioned

Products and companies

Research cited

• A study of over 200 million real notifications (not named in the captions; it matches the 2014 CHI paper Large-Scale Assessment of Mobile Notifications by Sahami Shirazi, Henze, Dingler, Pielot, Weber and Schmidt, which analyzed about 200 million notifications from over 40,000 users. That attribution is context, not stated in the video).
• "Productive, Anxious, Lonely: 24 Hours Without Push Notifications" by Martin Pielot and Luz Rello (arXiv 2016, revised 2017; MobileHCI 2017), the Do Not Disturb Challenge with 30 volunteers, shown on screen.

Design and psychology concepts

• Specificity vs. personalization in notification copy.
• Curiosity gap: deliberately leaving out a detail so you want to find out.
• Loss aversion: streaks and "don't lose it" framing.
• Context-based timing: sending relative to your past behavior, with a quiet-hours cap.
• Push vs. pull engagement, and notifications as the only push lever.
• Habit formation and the habit loop (the Hooked model).
• Urgency tiers: act now (messages), next hour, deal with it later (email inbox).
• AI notification summarization, reordering and filtering, plus the risk of hallucinated summaries.
• Agent-facing metadata (llms.txt, and a proposed hidden notification metadata layer), and designing for people's AI agents.

3) Biographies

Enrico Tartarotti (host, writer)

Italian-born tech creator and designer behind the YouTube channel Enrico Tartarotti, where he goes "behind the scenes of the tech products you use every day" through designed video essays on UX and product design. He builds Flask (flask.do), a video collaboration and review tool, and takes sponsorships through Rakugo Media. Credits for this video: written by Enrico Tartarotti, produced by Bartek Malinowski, edited by Seequence. He's on X as @EnriTarta and on Instagram as @enritarta.

People referenced

• Nir Eyal: author of Hooked: How to Build Habit-Forming Products (2014), which describes the trigger, action, variable reward and investment loop that habit-forming products use. He later wrote Indistractable (2019) about managing distraction.
• Martin Pielot: HCI researcher known for mobile notification and attention studies (formerly at Telefónica Research, later at Google), co-author of the 24-hours-without-notifications study shown in the video.
• Luz Rello: researcher known for work on dyslexia and accessibility (founder of Change Dyslexia), co-author of the same study.
• Luigi Mangione: the man charged in the December 2024 killing of UnitedHealthcare CEO Brian Thompson. Apple's AI summary falsely told BBC News app users he had shot himself.
• Nadia: Enrico's Hinge match in the example notification (a joke; not a public figure).
• Bartek Malinowski: producer. Seequence: editing studio.