When the ad is the product.

I started with a high-distrust audience, a high-volume ad load, and no existing practice to build from.

I joined Crunchyroll as the only content designer in the entire organization, not just for ads but for every member-facing surface across the product. A Q4 2024 content discovery survey of 77 active members had already told us what we were working with: only 18% of members trusted the platform's recommendations. Eighty-nine percent required social validation before committing to something new. The survey also surfaced the top frustrations in open-ended responses: guidance and context were lacking, recommendations were not seen as trustworthy, and browsing the catalogue felt too difficult. These weren't ad complaints. They described the baseline state of the platform before a single ad decision had been made.

That distinction matters. Ad content design on a streaming platform is, at its core, a discovery problem. Every ad moment is a moment where the member's attention has been pulled from content they chose toward content they did not. Getting them back to the show, or getting them to genuinely engage with a promotion, requires understanding how they make content decisions in the first place. The discovery research did not ask about ads, and that's precisely what made it useful. It told us how members think about content before they have been asked to sit through advertising, which is the mental model they bring into every ad pod. Using it to ground ad content decisions wasn't a stretch, because it was the most honest evidence base available.

Worth noting: because this was independent research rather than a study commissioned to support the ad system, the findings were harder for stakeholders to dismiss. They didn't come from the ad team's assumptions about its own work. They came from members talking about their experience of the product as a whole.

The ad load was not small. Each episode included roughly 30 seconds of pre-roll followed by three mid-roll breaks of about 80 seconds each, which puts close to five minutes of ads inside a 23-minute episode. Inside each break, an ad pod of two or three ads runs back-to-back before the show resumes. Every piece of content I wrote sat inside that constraint, and every decision had to earn its real estate against a member who was already counting down.

There was no content practice, no design guidelines, and no precedent. I built the system, the surface-by-surface decisions, and the documentation from scratch. Alongside engineering and product, I worked across ad sales, legal, and research, a team that had competing priorities and clear opinions. The content decisions needed to hold under that pressure, not just make sense in a design file.

The first trust moment had to earn its keep, or the rest of the session never would.

When a member hits play, they expect to see what they chose. The SVOD promo runs instead, and a member who feels ambushed by it starts hunting for the exit before the first show frame has loaded. That's a trust violation built into the product architecture. Content design cannot remove it, only manage it. The question was: how do you earn the member's patience through a moment they did not choose, without making them feel manipulated?

The answer matters beyond the experience. A member who feels respected through the promo is more likely to stay through the pre-roll that follows, and more likely to sit through the mid-roll ad pods after that. Completion rates compound across the session. The SVOD promo is where session-level trust is established or lost, and trust is what determines whether the ad inventory holds its value.

SVOD Promo (in-player) · first frame before content starts CTV · Mobile · Web

Every content decision on this screen was made against the research and built to serve both the member and the business at the same time. The survey showed that 79% of members use story description to decide whether to watch something. The SVOD promo had to do that job in a compressed, trust-borrowing format, without the space a full series page gets.

The counter is a contract, and the skip carries the signal.

Once playback starts, ads run through GAM (the ad delivery service) and Velocity (the video player), two separate systems. My job was designing the content layer on top of both: the counter pattern, skip states, upsell placement, and the bridge copy that returns the member to the show.

The core principle: on surfaces where you cannot offer meaningful agency, transparency is the only honest trust signal available. The counter is the contract. When I tell a member "Ad 2 of 3," I have made a promise. Breaking it (through a miscount, a missing break, or a wrong total) is a trust failure that poisons the rest of the session.

The surface differences between CTV and web go deeper than interaction model. CTV is lean-back, with no cursor, no skip expectation, and completion rates already higher. Web is lean-forward, where a member at a keyboard has more perceived control, and withholding all agency at that moment creates more friction than it resolves. But explaining skip as a surface decision understates the actual logic. The skip on web functions as a conversion surface with a defined lifecycle. The member most motivated to upgrade is the member sitting through their first mid-roll break. That moment, peak frustration and peak conversion intent, is exactly when the skip should exist, and exactly when it should surface the upsell. Once the conversion window closes, the skip has done its job, and what follows is a different design problem entirely.

CTV · counter only, no false agency LRX / TV

Content decision

Ad 1 of 2 with an 80-second orange countdown circle, positioned top-left where CTV eyes naturally scan. Two pieces of information, doing different jobs. The ad count sets scope: the member knows this is one of two, not an open-ended interruption. The countdown sets pace: 80 seconds, visibly moving, not abstract. Together they make the pod legible without a single word of explanation. No skip. No upsell. The counter is the only promise made on this surface, and it has to be kept exactly.

Web · counter plus agency plus upsell Web

Word order is the content decision

"Skip ads, go Premium", not "Go Premium, skip ads." Member benefit leads. Business mechanism follows. Reverse the order and it reads as a sales pitch with a reward buried inside it. This order reads as agency with a path available.

Why the skip exists, when it exists, and when it stops.

The objective of the ad platform is to sustain the conditions under which members remain willing to watch ads across sessions. A member who cancels or downgrades generates zero inventory. The content system has to serve completion rates and tier retention simultaneously, and those two goals require different responses at different points in the member lifecycle.

The table below maps member state signals to content decisions. CTV holds constant across all states because the interaction model and lean-back behavior make the progression unnecessary. Web and mobile adapt because the member's relationship to agency on those surfaces changes as they move through the lifecycle.

I mapped every ad moment, surface, and content response.

Before writing a single word of copy, I mapped the complete player state as a sequence flow: from playback request through entitlement check, SVOD promo, ad pods, dub card, and end slate. Every decision node, every branch condition, every terminal state. The map made the content design problem legible to engineering, product, and legal without requiring a separate conversation for every edge case. It also made the scope of the work visible in a single view, which mattered for stakeholder alignment in an organization where no one had done this before.

Sequence flow · full playback journey · all surfaces Systems documentation

The gate taxonomy (6 gate types across 4 surfaces) was a direct output of the mapping process. Gates are the moments where the member cannot proceed without taking an action or being redirected. Naming them, categorizing them, and documenting the content response for each one meant that when a new gate state appeared during development, the team had a framework to drop it into rather than solving it from scratch each time.

Here's where the exchange pays off.

Fifty-four percent of members sought new content immediately after finishing a series, making the post-completion moment the single highest-intent discovery touchpoint in any session. It is also where the flywheel closes.

The survey gave us something more specific than the raw percentage: 61% of those members wanted something with similar story or themes, and 51% wanted something from their own watchlist. That means a generic recommendation at the end slate is a missed opportunity even when it lands. The content design decision on that surface is to surface the right signal in the right order: watchlist first, then similarity, because both are validated by the member's own behavior rather than the platform's algorithm.

A member who saw a relevant SVOD promo, added it to their watchlist, sat through mid-roll ad pods, and arrived at the end slate has demonstrated intent at every stage. The end slate serves them the next show. The exchange is complete: they gave attention, they got content, and trust held through the whole transaction.

The flywheel only closes if every earlier stage earned it. The promo earns watchlist signal. The counter earns patience through the pod. The bridge earns the return to content. The end slate earns the next session. If any stage breaks trust, the flywheel stops, and when the flywheel stops, so does the inventory value of everything that follows.

End slate · post-completion · Web Phase 03: Post-play

I made the privacy controls feel like product features.

Being the only content designer in the organization meant the ad experience did not exist in isolation. Every decision I made in the player had to be consistent with what members encountered in account settings, in onboarding, and in the privacy controls that governed how their data was used for advertising. The Data and Privacy screen was one of the highest-stakes surfaces I owned, not because members visited it often but because every word on it was load-bearing. Legal compliance, advertiser capability, and member trust were all sitting on the same page.

Account Settings · Data and Privacy · web Privacy and consent UX

Navigation placement: General, not Account, not a standalone Privacy section.

"Data and Privacy" sits fifth under the General navigation group, after Membership Info, Preferences, Activate Device, and Device Management. That position was deliberate. Placing it under Account (alongside Email, Phone, and Password) would have framed it as administrative and potentially alarming, a legal compliance checkbox rather than a member setting. Placing it under General alongside Preferences frames it as something the member controls as part of how they use the product.

The hierarchy communicates: this matters and we are not hiding it, but it is not an emergency. Members who want to find it can. Members who are not looking for it are not made anxious by its prominence. That balance required a navigation architecture argument, not just a content decision. It also required legal sign-off that the placement met disclosure obligations without requiring it to be the first thing a member sees.

Default states: both toggles off, opt-in rather than opt-out.

Both Profile-Based Ads and Personalized Ads ship with their toggles off. This is the single most consequential content design decision on the screen, because of what default states communicate at scale. An opt-out default says: we assume your consent. An opt-in default says: we will ask for it.

For a member base where only 18% trusted platform recommendations, opt-out would have been a trust catastrophe. The business case for opt-out is real: higher participation rates in behavioral targeting means richer audience segments, which means higher advertiser CPMs. The content design argument against it is also real: a member who discovers they were opted in without asking is not a member who stays opted in. The default state is a retention decision dressed as a privacy decision.

Toggle hierarchy: email identity before behavioral data.

Profile-Based Ads leads. Personalized Ads follows. Profile-Based uses email identity matching, meaning the member already knows they gave Crunchyroll their email. Personalized Ads uses behavioral data: viewing history, genres watched, time spent. The more concrete and already-understood data type leads. The more abstract type follows.

This is a progressive disclosure decision applied to a consent surface. Starting with the familiar lowers the cognitive load before the member encounters the unfamiliar. If the order were reversed, with behavioral data first and email second, the member would have to process a more abstract concept before they had context for what kind of data the platform was even talking about.

Terminology: plain language for the ask, legal language where required.

The page title is "Data and Privacy" rather than "Ad Preferences" or "Privacy Settings." Broader framing, more honest, less optimized for advertiser framing. The subtitle, "Decide how your data and personal info is being used," opens with a verb that gives the member agency. The member is deciding, not configuring.

For Profile-Based Ads, the description names exactly what is being shared ("your email"), exactly who receives it ("Crunchyroll ad partners"), and what it produces ("ads based on your interests"). Three specific answers to the three questions a member has when they see a toggle they do not understand. This was the result of working closely with legal to translate compliance requirements into plain language without losing their legal meaning.

"Do not sell or share my personal information" at the bottom is CCPA-required copy, legally mandated verbatim language. It is also the only phrase on the screen that a member cannot act on without already understanding what it means. It reads as a declaration rather than an action. That tension is where legal requirements and member clarity genuinely conflict, and the resolution here was to isolate the legal language at the bottom of the screen, visually separated from the plain-language controls above, so it reads as a reference rather than the primary ask.

Where legal and member clarity were in tension, and how those negotiations went.

The two toggle labels ("Profile-Based Ads" and "Personalized Ads") were not the first versions legal and I agreed on. The original legal drafts used terms like "targeted advertising" and "interest-based advertising," which are accurate within the industry but meaningless to a member who has never worked in adtech. My initial push was for labels that named the data type directly: "Email-Matched Ads" and "Behavior-Based Ads." Legal's concern was that these were not established terms in their disclosure framework, and that introducing non-standard labels could create ambiguity in a regulatory context.

The resolution was a negotiated middle: labels that were plain enough for members to read without a glossary, but stable enough for legal to defend. "Profile-Based" anchors to identity. "Personalized" anchors to behavior. They are still closer in meaning than I would have chosen without legal constraints, and a member scanning quickly might not register the difference at the label level. But that is what the descriptions are for, and the description-level work is where I was able to win the plain-language argument more completely.

The Personalized Ads description in the first screenshot read "Allows sharing of behavioral data with ad partners." Legal owned that language and treated "behavioral data" as a defined term. My argument was that defined terms belong in privacy policies, not in product UI. I pushed for language that named what behavioral data meant in practice: what the member had watched, how long they watched it, what genres they returned to. The updated version, visible in the screenshot, reflects that negotiation. The category term ("behavioral data") is gone. The actual content is there instead. Legal signed off because the plain-language version was, if anything, more accurate than the category term, and accuracy was their primary concern.

This is what content design work inside a legal constraint looks like. The work is building enough trust with legal to make the plain-language argument on its merits, iteration by iteration, until the copy on screen serves members without creating compliance risk for the company.

Here's why this matters now.