Netflix Deploys Generative AI to Fight Content Overload

Netflix is using generative AI to help its 325 million subscribers find something to watch faster, chief product and technology officer Elizabeth Stone told the Bloomberg Tech conference in San Francisco on Wednesday. Stone described the technology as enabling “the more personalized, the more interactive, the more immersive” content experience, framing it as the company’s primary answer to a US catalog that has grown past 4,000 movies and 2,000 TV shows and keeps expanding.

The company posted $12.25 billion in Q1 2026 revenue, up 16.2% year-over-year, with profit rising 83% to $5.28 billion. Its ad-supported tier reached 250 million monthly active viewers globally by May, up from 190 million in November 2025, with the company targeting $3 billion in ad revenue for the full year.

Stone’s Argument at Bloomberg Tech

Stone was promoted to the newly created chief product and technology officer role in February 2026, merging Netflix’s product, engineering, and data teams under a single executive. She joined the company in 2020 as vice president of product data science and engineering, was named chief technology officer in October 2023, and now oversees the engineering infrastructure alongside the product decisions built on top of it. Her profile on Netflix’s investor relations site describes her mandate as leading the company’s product, engineering, and data organizations.

Wednesday’s Bloomberg Tech appearance is the clearest public statement yet of where that mandate is pointed. Netflix’s recommendation system already accounts for roughly 80% of what subscribers watch, by the company’s own data. Stone’s case is that generative AI handles the remaining friction: the gap between a subscriber opening the app and committing to something, the moment where the current system’s confidence drops or the viewer’s intent is too vague to parse from a browsing history alone.

Co-CEO Gregory Peters framed the same shift during the Q1 2026 earnings call in April. “We have been in personalization and recommendation for two decades, but we still see tremendous room to make it better by leveraging newer technologies,” Peters said. New model architectures, he added, let Netflix iterate on recommendations faster and add support for different content types much more efficiently than prior systems allowed.

The Library That Got Too Large

Netflix’s US catalog held over 4,000 movies and more than 2,000 TV shows as of early 2025, growing continuously through a pace of original production that has reached roughly one new title per day in recent years. Each addition widens the discovery gap: the catalog grows faster than any individual subscriber can navigate it, which means the recommendation engine’s quality, not the catalog’s volume, determines whether any given viewing session ends with something playing.

Research on choice architecture, including psychologist Barry Schwartz’s widely cited findings on how excessive options induce paralysis rather than freedom, maps cleanly onto the streaming context. Netflix has already tried simpler interventions: a “Play Something” shuffle button, country-specific “Top 10” rows, and category-based sorting all reduce the visible catalog without tailoring suggestions to individual intent in real time. The existing AI system handles the problem roughly 80% of the time. Subscribers who open the app and close it without committing represent the remaining gap.

• 80% of content watched on Netflix is surfaced by its AI recommendation system, per the company’s own data
• 250 million monthly active viewers on the ad-supported tier globally as of May 2026, up from 190 million in November 2025
• 40% of new Netflix signups now choose the lower-cost ad tier, per the company
• $3 billion targeted in ad revenue for 2026, approximately double the 2025 total

Recommendation Data Powers the Ad Tier

Scale of the Behavioral Dataset

Each time Netflix’s recommendation engine moves a subscriber from browsing to watching, it records what moved them: which title was surfaced, which thumbnail was displayed, how long before the click, whether the subscriber finished the episode or left midway. Across 325 million paying subscribers globally at year-end 2025, per Netflix’s Q4 2025 earnings filing with the Securities and Exchange Commission, those signals accumulate into a behavioral dataset that trains future recommendations and informs ad targeting simultaneously.

The ad-supported tier has scaled sharply on the back of that dataset. Netflix’s 2026 upfront presentation reported 250 million monthly active viewers, with more than 80% watching actively every week. The company’s Q1 2026 shareholder letter, filed with the SEC on April 17, 2026, confirmed ad revenue on track for $3 billion this year, doubling 2025, with full-year revenue projected between $50.7 billion and $51.7 billion.

AI Agents in the Ad Stack

Netflix is also embedding AI in the buying layer. The company tested AI purchasing agents, software that autonomously manages and buys ad campaigns without manual input, with DoorDash, Target, and TurboTax ahead of its 2026 upfront. Nearly 50% of Netflix’s non-live ad inventory now moves programmatically. The active advertiser base surpassed 4,000 companies, up 70% year-over-year.

Netflix is separately expanding contextual advertising capabilities that match brand creative with specific shows and viewing environments. The feature was piloted with the same initial advertisers and is scheduled for full rollout across all ad-supported markets by year-end 2026.

Better recommendation precision generates richer behavioral signals, and richer behavioral signals make ad inventory more targetable. More targetable inventory commands higher cost per impression, and AI purchasing agents reduce the friction for advertisers accessing it. Each browse session the recommendation engine resolves into a viewing commitment adds a data point the ad stack did not previously have.

The Vertical Feed and Conversational Search

Netflix launched a TikTok-style vertical video feed at the end of April 2026, confirmed in the Q1 2026 earnings release. The feature shows short clips from shows and movies in a scrollable vertical format. Subscribers can tap any clip to start the full title immediately, save it to their watch list, or share it, and the feed includes video podcasts alongside films and series.

The feed is a distinct behavioral surface from the main browse grid. The grid records which title a subscriber selects from a row of thumbnails. The vertical feed records which 15-to-30-second clip prompted a full watch or a save, a finer-grained measure of what moved someone from scrolling to committing. Intent captured at the clip level is more specific than intent inferred from which genre row a subscriber browsed.

Netflix launched a ChatGPT-powered natural-language search feature in 2025, letting subscribers describe what they want instead of navigating genre filters. The vertical feed, the conversational search, and the generative personalization described at Bloomberg Tech each feed the recommendation system with behavioral data that the static grid interface alone cannot capture.

The $600 Million Production Bet

Netflix closed its acquisition of InterPositive in Q1 2026. Bloomberg reported the deal could reach up to $600 million in total consideration; the actual cash paid was less, per people familiar with the terms. Netflix’s Q1 cash flow statement shows $585.7 million in acquisitions spending for the quarter, consistent with that range. The acquired company, co-founded by actor and filmmaker Ben Affleck, built a toolkit for post-production work, letting filmmakers alter and refine existing footage using AI models. Filmmaker David Fincher used the tools on an upcoming project before the acquisition closed.

Netflix had used generative AI in production for the first time on El Eternauta, a Netflix original, before the deal came together. Stone announced the acquisition with this statement:

Our approach to AI has always been focused on meaningfully serving the needs of the creative community and our members. The InterPositive team is joining Netflix because of our shared belief that innovation should empower storytellers, not replace them.

Stone made the statement as Netflix’s chief product and technology officer at the time of the deal’s announcement in March 2026. The tools operate under published constraints: the system trains only on existing footage, requires creator consent before any content is used for training, and cannot generate new projects from scratch without base material. An eMarketer analysis of the acquisition described Netflix’s positioning as one focused on optimization over automation, a read that Affleck’s ongoing role as an active filmmaker helps reinforce with the production community.

Where Rivals Stand

Netflix’s moves on AI discovery, ad-tier scale, and production tools are running in parallel with investments at YouTube and Amazon, with each platform having taken a distinct shape.

The furthest public commitment on AI-generated content belongs to iQiyi, the Beijing-based streaming platform. In April 2026, iQiyi founder and chief executive Gong Yu said AI would create the bulk of the platform’s films and shows within five years, a commitment that goes beyond anything Netflix has signaled. iQiyi launched the Nadou Pro AI toolkit, covering scriptwriting, storyboards, and final rendering, and is reorienting its video platform toward a social-media model built primarily on AI-generated content.

Netflix has organized its AI investments around improving and surfacing human-created content. The personalization strategy outlined at Bloomberg Tech, the production tools acquired with the Q1 deal, and the new discovery surfaces share that aim. Netflix’s Q2 2026 earnings report is the next public measure of whether the AI-driven discovery push is translating into ad pricing gains alongside the engagement numbers.