AI Driven Retail Media: How Generative AI Reshaped 2025 Advertising

Sensor Tower’s latest State of Mobile data shows a crisp, structural shift: generative AI didn’t just reshape app usage in 2025 — it rewired retail media, turning high-intent shopping moments into a new battleground for brand attention and co‑branded advertising. The effect was dramatic: generative AI downloads doubled year‑over‑year, non‑game IAP revenue surged, and retail media category impressions spiked, with Microsoft’s Copilot campaign emerging as an unmistakable accelerant for the category’s growth.

Background / Overview​

Sensor Tower’s 2026 “State of Mobile” frames 2025 as the year when AI assistants and AI-enabled formats moved from experiment to scale. Across iOS and Google Play, the firm records record totals for downloads, time spent, and in‑app purchase (IAP) revenue — a shift that pushed non‑game monetization past games for the first time and made AI a central driver of attention monetization. Generative AI apps doubled downloads and roughly tripled IAP growth, while social apps continued to dominate raw time spent. That dual dynamic — AI’s rapid adoption + social’s persistent attention share — set the stage for advertisers to reallocate budgets into retail media and video formats where AI narratives could be shown, not just told.
At the same time, independent industry tallies highlighted methodological differences in absolute figures, so the takeaway should be directional rather than fixated on single-number totals: AI accelerated monetization, and retailers and brands rushed to capture those high-value moments.

---

The numbers that mattered​

• Generative AI downloads: +100% YoY (doubled).
• Non‑game IAP revenue: surpassed games in 2025, driven by subscriptions and AI features.
• Retail media category impressions: +165% YoY (driven in part by eight major Gen‑AI brand campaigns across retail networks).

Sensor Tower’s dataset surfaces a clear composition effect: a handful of large Gen‑AI advertisers dominated the incremental impression volume, with Microsoft Copilot, Google AI, and Adobe accounting for the largest shares of activity. Where previous retail media growth was broad-based, 2025’s surge concentrated around a short list of high‑spend AI brands and a strategic media mix centered on video.

---

Microsoft Copilot: the campaign that moved impressions​

Microsoft’s Copilot push — notably a retailer co‑op with Costco promoting the “Copilot + PCs” category — delivered a huge October spike in retail impressions and defined the narrative of AI + hardware for consumers. That execution shows two important lessons:

• Retailer partnerships can multiply reach quickly when a brand enters the retail media ecosystem with deep creative assets and tileable video content.
• Category creation matters: “Copilot + PCs” reframed Microsoft’s hardware partners as AI-native devices rather than incremental refreshes, giving retail media placements a stronger commercial hook during the holiday device-buying window.

The Copilot effort illustrates how a single brand with the right creative, retailer mix, and seasonal timing can lift an entire advertising vertical on retail networks. (Sensor Tower’s reporting highlights this spike and attributes much of October’s uplift to Microsoft’s program.)

---

Who bought the attention — and where​

Top advertisers by impression volume​

• Microsoft Copilot (largest share; dominant October presence).
• Google AI (multi‑hardware messaging across phones, laptops, smart home).
• Adobe (Acrobat Studio launch and creative/productivity positioning).

Eight generative AI brands were responsible for the bulk of the category’s growth, and their retailer choices reveal strategy: Microsoft included Costco; most others concentrated impressions on Amazon and Best Buy. That retail selection reflects two factors — retailer inventory depth for device and software bundles, and the programmatic accessibility of OTT/offsite video inventory tied to those retailers’ ecosystems.

Media channels: why video mattered​

Gen‑AI advertisers skewed heavily into video-first channels — OTT and YouTube-style placements — because video communicates sequential, aspirational, and practical use cases that static banners cannot. OpenAI’s ChatGPT, for example, ran the majority of impressions on OTT within Amazon’s retail media environment, showcasing assistant use cases like trip planning and recipe discovery. Anthropic similarly leveraged Amazon OTT and offsite video but positioned Claude toward advanced problem‑solving scenarios like science and research. These executions show that retail media is not just product tiles and search; it has become a video fabric where brands show the power of AI within everyday flows.

---

What the creatives said: positioning across the AI spectrum​

A closer look at top video creatives shows a striking pattern: most ads try to do three things — educate, demonstrate, and remove friction — but they differ wildly in tone.

• Microsoft Copilot: productivity + education — emphasis on work and school scenarios showing AI as a practical productivity layer on laptops.
• Google AI (Gemini): integration + daily companion — messaging that presents Gemini as a helper across hardware lines (phones, laptops, smart home), emphasizing breadth.
• Adobe (Acrobat Studio): creativity + workflow — agentic AI supporting creative production across Adobe’s toolset, pitched at creators and power users.
• OpenAI / Anthropic: use-case storytelling — scenario‑based creatives showing the assistant solving real tasks, from trip planning to technical problem solving.

The result is a spectrum: some brands sell utility (Microsoft), others sell integration and convenience (Google), and others sell capability and creativity (Adobe). That differentiation matters in retail media where the same viewer might encounter product tile, OTT spot, and sponsored search within minutes — coherence across those touchpoints improves selection and conversion.

---

Why retail media was uniquely fertile ground for AI advertising​

Several structural changes converged to make retail media the natural amplifying channel for generative AI in 2025:

• First‑party intent signals: Retailers have transaction history, product availability, and expressed intent — data that makes AI-driven messaging more actionable. This advantage has been true for years, but AI assistants sharpen the value of intent signals by compressing funnel steps into a single conversational decision.
• Agentic commerce experiments: Retailers and platforms are piloting “in‑answer” or “agentic” ad primitives — sponsored suggestions surfaced inside AI-driven shopping flows — which reward advertisers who optimize for selection by a model rather than rank on a grid. Early pilots at major retailers pointed to this future.
• Video + demonstration value: AI marketing benefits from sequential demonstrations; OTT and offsite video on retailer networks let brands show workflows (e.g., Copilot boosting laptop productivity), reducing perceived risk and improving intent-to-conversion ratios.

Taken together, these forces made retail media not just a place to place ads, but a place to validate a brand’s AI proposition in a high‑intent, transaction‑ready environment.

---

Critical analysis — strengths, constraints, and risks​

Strengths (what the data signals marketers should celebrate)​

• Higher‑quality impressions: AI-assisted sessions compress intent and increase the value of a single impression; fewer touchpoints can lead to faster, more attributable outcomes.
• Creative clarity: Video-first retail media allows showing agentic behavior rather than claiming it; brands that used narrative demos lowered cognitive friction in buyer decisions.
• Retailer participation expands ad inventory: Partnerships between brands and large retailers (e.g., Microsoft + Costco) unlocked premium seasonal inventory and scaled reach beyond standard DSP buys.

Constraints and measurement headaches​

• Attribution breaks: Traditional last-click and click‑through attribution are less meaningful when assistants synthesize and complete transactions inside an agentic flow. Measurement must evolve to conversion-lift, holdout tests, and server‑side event logging. If advertisers fail to adapt, they will under‑credit or over‑credit channels in a way that distorts investment decisions.
• Methodology variance across trackers: Absolute download and revenue figures vary by vendor; Sensor Tower’s totals differ from Appfigures’ totals. Marketers need to triangulate and focus on directional signals (growth rates, category shifts) rather than exact dollar or download totals.

Risks for brands and retailers​

• User trust and helper credibility: Agentic ad formats — in‑answer sponsorships or sponsored follow‑ups — must feel helpful. If ads feel manipulative or lower the assistant’s credibility, backlash is immediate and measurable. Platforms and advertisers must design ads that augment, not interrupt, the assistant role.
• Privacy and data governance: Retailer agents rely on first‑party data to power recommendations. The monetization of those signals raises regulatory and reputational risks if consent, data minimization, or transparency are mishandled. Retailers and brands must build privacy-preserving measurement and clear user choice into agentic commerce.
• Concentration and creative fatigue: With a few large AI brands driving the impression surge, smaller advertisers may struggle to compete for premium slots, and consumers may experience ad fatigue if the same AI narratives dominate the feed. This concentration can raise CPMs and compress smaller-category visibility.

---

How to act: practical playbook for advertisers and retailers​

For advertisers​

• Map creatives to the funnel — produce short, sequential video assets that show AI solving specific, high‑intent tasks (device setup, workflow automation, shopping lists). Video sequences should be tileable for OTT, in-stream, and retailer offsite video inventory.
• Optimize product data for model selection — agentic flows will favor products that present clean, structured attributes (availability, price, SKU-level data). Treat product feeds as model input rather than just search keywords.
• Invest in lift measurement — move budget toward experiments that use holdout groups, incremental lift tests, and server‑to‑server conversion measurement rather than last-click reporting.
• Prioritize retailer partnerships — co‑op buys or creative collaborations with retailers can unlock premium placements and co‑branded messaging that feels native in an agentic flow.

For retailers and platforms​

• Build clear monetization primitives for agentic commerce: in‑answer sponsorships, sponsored follow‑ups, agentic cards with direct purchase hooks, and revenue-share models for publishers cited by agents. These primitives must preserve the assistant’s credibility and offer transparent user controls.

---

What success looks like — short, medium, and long horizon​

• Short term (next 6–12 months): expect more targeted OTT and retailer‑video campaigns from AI-first brands testing product bundles, with measurement moving to conversion-lift proofs. Creative that demonstrates immediate utility will outperform aspirational puffery.
• Medium term (12–24 months): ad formats inside answers and agentic selections will become more formalized. We will see emerging pricing models pay-per-assist or revenue share for agent‑driven transactions. Retailers that instrument these primitives cleanly will capture a larger share of advertiser spend.
• Long term (24+ months): if assistants become dominant discovery surfaces, brands will compete to be selected by models. That competition favors companies that optimize product data, syndicate robust model‑friendly content, and sustain user trust through transparent ad design and privacy protections.

---

Creative and technical considerations — checklist​

• Creative: produce 6–15 second demo cuts + 30–60 second narrative spots; always show the problem → AI assistant → solution sequence. Include clear calls to action that map to retailer availability.
• Data: ensure SKU-level availability, shipping time, and structured specs are pushed into retailer feeds and merchant centers — these data points are the features models use when recommending items.
• Measurement: implement server-side event collection and randomization to enable lift tests; use revenue-share pilots and holdouts to allocate credit where agents complete the transaction.
• Governance: publish clear disclosure for in‑answer sponsorships and implement privacy-first signal flows that permit measurement without exposing raw consumer data.

---

The macro view: AI rewrites ad economics​

Generative AI’s rise changes ad economics in two fundamental ways: it concentrates intent into fewer, higher‑value moments and it changes the unit of value from pageviews and clicks to agent selections and assisted actions. Retail media — with its first‑party commerce signals and growing video surface — has become the natural amplifier for this new economics triangle: brand value proposition, retailer availability, and assistant selection. The year’s data show that being present, coherent, and demonstrative in that environment yields outsized returns.
At the same time, the industry faces new measurement and trust challenges. If platforms and advertisers can solve for privacy‑preserving attribution and create ad formats that augment trust rather than erode it, retail media will mature into a higher‑quality ad ecosystem that rewards relevance over reach. If those problems aren’t solved, the surge of 2025 risks collapsing into wasted spend and consumer cynicism.

---

Final takeaways​

• The Sensor Tower data make one thing clear: generative AI was not a marginal driver in 2025 — it was a structural accelerator of mobile monetization and retail media growth. Marketers must treat assistant‑driven commerce as a distinct channel requiring different creative, measurement, and data hygiene.
• Video-first retail media campaigns — especially those that demonstrate concrete agentic value — outperformed static approaches in creating intent and accelerating selection. Brands should prioritize short-form demonstration assets and tight retailer integration.
• Measurement must evolve. Advertisers who cling to last‑click rules will misread the new economics; those who invest in lift testing, server‑side instrumentation, and partnership pilots will gain clarity and competitive advantage.
• Finally, trust and governance will determine whether agentic advertising becomes a sustainable uplift or a short-lived spike. Retailers and brands that bake transparency, privacy, and clear sponsorship disclosures into agentic ad formats will win the long game.

The surge is real, but its permanence will be decided by measurement rigor, creative craft, and how well the ecosystem balances monetization with the helper role of AI. The 2025 data give marketers the playbook — now execution will test who can translate impressions into long‑term preference and revenue.

---

Source: Sensor Tower How Gen AI Prompted a Retail Media Surge