Onboarding

I try to get teams to treat a trailer less like advertising and more like a claims document. Whatever it promises (power fantasy, fairness, friends, mastery), the first hour gets audited against it, line by line, by someone who doesn’t even know they’re auditing. And when the promise and the product disagree, the player doesn’t blame the trailer. They blame the game, and they say so in the one venue your next cohort reads first.

The subtler failure is recruitment. An oversold promise doesn’t just disappoint, it fills your servers with the wrong players: people who came for a game you didn’t make. They churn, and they tend to churn loudly. You can ship a flawless tutorial into that cohort and watch it accomplish nothing, because the breakage happened weeks earlier in a YouTube pre-roll.

The cleanest version of this rule I’ve watched up close is Apex Legends, which launched in February 2019 with essentially no promise window at all. It was announced and playable the same day. First Seeing was the real game, running live, on streams, and because the promise and the product were the same artifact, the promise was unbreakable. I’d been at Respawn about five months, and I’m not going to pretend a shadow drop is a repeatable strategy (it isn’t). But the principle underneath it holds: the closer First Seeing sits to the actual game, the less debt the next two hits inherit.

Player counts mostly decay from launch. The largest audience your game will ever have shows up in week one and meets the least battle-hardened version of everything: servers, balance, FTUE, all of it. The funnel math is brutal and public: GameAnalytics’ benchmarks put roughly 20% of installs gone within two minutes of first launch, median D1 retention around 26–28%, and D7 near 8%. The cliff lives at the top of the funnel, in the hours you control least. Celia Hodent put the priority order plainly: “If your game fails to captivate your audience’s attention quickly, there won’t even be any retention challenges to care about.”

I’ve lived this failure mode myself. New Champion Select shipped in January 2016 and collapsed under live load on day one. Riot’s public statement was straight about it: “we used a simulated environment to load test the entire system… However, some subsystems of the dynamic queues didn’t scale up correctly when running on a larger live environment.” We rolled back to the 2015 systems and re-rolled out region by region. The feature recovered, but the lesson stuck with me: features get a second rollout, first impressions don’t.

And the cohorts chain. Your earliest adopters are the loudest people in your community. They write the day-one reviews, cut the clips, and set the tone, so a meltdown they experience becomes the First Seeing for everyone who arrives after them. Fumble hit two and you’ve retroactively fumbled hit one for the next wave. That’s why I treat launch stability as a design deliverable, not an engineering afterthought: the live game on day one is the most-played level you will ever ship.

I call it the First Head Nod, and to me it’s the actual deliverable of onboarding, not tutorial completion or D1. It’s the moment Koster’s line stops being theory: fun is learning, and the head nod is the sound of a pattern snapping into place. It has to be inferred, which is why the scripted “great job!” beat never produces one. The player knows they were led there, and so do you.

The craft here is navigation, not decoration. Find your game’s head nod, then measure the distance to it in minutes. GameAnalytics’ research says the viable first session runs 10–20 minutes, so the nod has to land inside that window or it lands for nobody. Then clear the road to it. Just the road, though, not the whole landscape.

That last distinction is where I think live service went wrong, and Elden Ring is the receipt. By 2022 we had over-learned frictionlessness into a kind of cargo-cult smoothness, copying the polish of successful FTUEs without asking what the smoothness was for. Then a game shipped that explained almost nothing, violated most of the usability playbook, got publicly prosecuted by AAA UX designers on Twitter in March 2022, and sold over 25 million copies.

My read is that the lesson transfers but the budget differs. A free-to-play live-service game can’t charge Elden Ring’s toll at the door and survive its own D1 chart. Nobody has proven otherwise, and I wouldn’t volunteer to be the test case. But the direction of the error matters. When you strip every obstacle from the first hours, you also strip discovery, identity, and pride, and those are the raw materials of the head nod. In my experience the enemy of onboarding was never friction as such, only friction the player hasn’t been set up to overcome. A head nod can’t be given, only earned, so just be sure you leave the player something to earn.

The Hextech trade from chapter 05 (agency given up for a reduction in cost) turns out to be the tutorial designer’s bargain too. Every piece of scripted teaching confiscates some of the player’s agency in exchange for lowering their cognitive cost. It’s a real trade, and sometimes worth making, but it’s also a debt instrument: rails teach the rails. The player who clears your scripted course has learned your course, not your game. Then in their first unscripted match they relearn everything under fire, except now it costs you trust instead of tuition.

I agree with every word of that sentence. I also note what the same 2018 post admits about the redesigned tutorial: “It doesn’t tell you about last hitting. It doesn’t even teach you about buying items.” That isn’t hiding the learning, that’s triage: a complexity budget blown years before the tutorial team ever got to spend it. It’s easy to agree with the philosophy, but the budget is the actual policy. This is why I push teams off the word “tutorial” entirely. A live-service game has no World 1-1. Its teaching surface is champion select, the menus, the battle pass screen, the first ranked match, and those systems teach whether or not you staffed them as teachers. Only about 40% of players finish a typical tutorial anyway (devtodev’s published optimization target is 70–80%), so the real game is the only classroom with full attendance.

When you must use words, ration them. George Fan’s rule from the Plants vs. Zombies talk is still the most concrete in the canon: “There should be a maximum of eight words on the screen at any given moment.” Go count the words on your first-hour screens. I’ll wait.

Bots earn their place in the first hours. They give a new player safe reps on the controls, room to breathe, and the nerve to queue again. But be precise about what they teach, because Riot published the answer in its 2023 bots post: “regardless of how many co-op vs. AI games a new player played prior to entering a PvP queue, these games would not improve the new player’s chances of winning.” Bots don’t really teach the game; they teach the interface and lend confidence. That’s still worth shipping (confidence is real). But if the skill doesn’t transfer, then the only durable thing an undisclosed bot win delivers is a false belief about yourself. You handed the player pride they didn’t earn, and the real game will repossess it with interest.

The industry’s disclosure spectrum is public record. Fortnite added bots in September 2019 that “will behave similarly to normal players and will help provide a better path for players to grow in skill” and then didn’t tell players which matches contained them. Marvel Snap built its early ladder on undisclosed bots up to roughly rank 30. Apex went the other way. Season 16’s Orientation Matches said it out loud: new players start in disclosed, bot-majority matches and graduate into the real pool. I led Experience Design at Respawn through that era, and I’ll defend the disclosure call without hedging. Telling the truth costs you a little theater on day one, but it buys the only currency that compounds across every chapter of this manual.

After my time, Respawn went further. EA’s June 2026 matchmaking update states that “bots have a key role to play in the extended onboarding experience” and describes bots woven directly into live low-skill matchmaking. Onboarding dissolved into matchmaking itself, and I don’t think that’s wrong. It only goes wrong on the day a player can’t find out who they actually beat.