Bananas, Breakthroughs and Backlash: AI's Wild Year in Advertising

Trends and Insight 2025: Making Sense of a Year in AI

2025 hit hard. AI moved from side project to standard practice across advertising, media, and production. By year's end, fully AI-generated or AI-enabled spots weren't the anomaly - they were everywhere. Big brands went public, budgets stretched further, and the comment sections lit up.

Tech Acceleration at Full Tilt

The AI video arms race escalated. Google's Veo 3 and OpenAI's Sora 2 added synced audio and better physics. Sora 2, Veo 3, and Midjourney V7 shipped character consistency features, though keeping a character locked across scenes still takes finesse.

Rollouts weren't uniform. Sora 2 remained limited to a handful of markets, with European data rules slowing expansion. If you work in Europe, you live this reality - regulatory pressure is real. For context on the privacy backdrop, see the EU's data protection framework here.

On usability, Runway's Act Two (July) let creators steer characters via simple, real-time performance capture. In September, Luma's Ray 3 brought multimodal input and better scene control - draw framing, specify camera moves, place characters. Then December happened: Google's Nano Banana Pro churned out near-photoreal images that fooled veterans. Meanwhile, OpenAI reportedly hit "code red" in response to new moves from Google's Gemini 3. Signals from investors pointed to froth; that's a 2026 concern, but the noise got louder.

Production Found a Hybrid Rhythm

The myth: type a prompt, get a finished ad. The reality: bespoke workflows that mix AI with traditional craft. VFX teams use machine learning to speed up parts like CG lighting; editors and designers shift more time to creative polish.

Hybrid also means keeping human rituals that build alignment. Casting, location scouting, and look-dev still matter - even when the final is heavily AI. A recent Lamborghini film (Televisor Studios, Iryna Nalyvaiko) blended AI for pre-viz and mood with 3D modeling for car control. Director Julie Reali's take: the edge is taste - knowing when to use AI, and when to put it down.

What Works - and What Doesn't

Adoption swelled. At Ogilvy EMEA, roughly 80% of people used AI to process research, iterate faster, and test scripts with virtual audiences. Useful? Yes. A silver bullet for ideas? Not really. Their work suggests AI still struggles to originate strong creative concepts.

Agentic AI became the buzz term. Teams trained agents to accelerate new business, strategy, UX, and production tasks. Yet many pilots stalled. A widely cited 2025 report from MIT argued most companies saw little financial return; media and tech were the exceptions. The takeaway: build guardrails, measure ROI early, and expect human oversight to make or break results.

On that note, several organizations learned the hard way that automated citations can hallucinate. Consulting, legal, news, and public sector teams were all publicly embarrassed at least once. If your workflow outputs references, treat it as suspect until verified by a human.

Holding Companies Went All-In

With choppy share prices, leadership leaned into AI. Publicis' 2024 investment in its Core AI platform set up a strong 2025. WPP elevated technical leadership, then launched Open Pro - tools without forcing agency services. It's a swing at in-house teams and SMBs who'd never hire a big network.

Omnicom focused on the IPG deal and Acxiom's data muscle, hoping better identity graph + Omni = sharper personalization. Independents didn't sit still either. Labs like Silverside partnered with consultancies; production shops launched AI divisions; new AI-native outfits hit the market. The tension ahead: talent pipelines. AI absorbs repetitive junior tasks - the same work where instincts are built. Cutting those roles might save expense and hurt the future.

AI Showed a Human Face

Irony of the year: AI giants leaned on human-led, live-action storytelling for their brand campaigns. OpenAI, Meta, Anthropic, Perplexity - all spotlighted relatable stories and faces, not pure product fireworks. The read is simple: trust beats tech demos.

Budget Stretch, Public Scrutiny

More brands disclosed AI use, especially in H2. Some embraced full AI production; others blended AI shots into traditional workflows. Either way, money went further: more versions, more ambition, faster localization. Most of it quietly worked.

But when it didn't, it got loud. McDonald's Netherlands pulled a Christmas ad after backlash. Coca-Cola took the hits and kept iterating - accepting criticism as the fee for moving early. For every pile-on, dozens of AI-assisted spots slipped by without comment. Localization remains a strong, low-drama use case - as long as native expertise checks the cultural details.

The Year's Messy Debates

Two themes stuck. First, cognitive costs. A June MIT study suggested tools like ChatGPT can reduce critical and creative brain activity. Second, "workslop" - AI-churned memos and reports created to look busy, read by no one, helpful to nothing. If you lead teams, watch for both.

AI influencers, models, and actors scaled up, igniting debates about authenticity, ownership, and transparency. One high-profile "AI starlet" dominated headlines, though the tech behind the persona was less impressive than the PR. Environmental impact moved up the agenda too. Data centers use serious energy and water, pushing AI into sustainability frameworks like Ad Net Zero. Bias, beauty standards, and what's "real" also got thornier as models like Nano Banana Pro blurred the line.

Legal Grey Zones and New Guardrails

Legal and ethical questions kept many brands cautious. Walled gardens trained on approved assets reduced risk - at the cost of creative range. Mid-market brands sometimes took a looser stance; others avoided the tools entirely.

Recent legal decisions and high-profile licensing deals suggested a shift toward more permissive use - with big conditions. Some studios reportedly licensed IP to AI platforms under strict terms, a notable turn for historically protective rights holders. National stances diverged too: proposals in some countries to restrict scraping contrasted with more permissive signals elsewhere.

Bubble Talk vs. Real Work

Yes, there's bubble chatter. But even if capital cools, the practical use cases aren't going away - especially in media and production. Expect better visuals and audio, fewer clunky prompts, and more multimodal control. Expect stronger governance and compliance checklists. And expect the argument to keep going.

What Teams Should Do in 2026

Adopt hybrid workflows: pair AI generation with traditional pre-viz, edit, and VFX guardrails.
Measure what matters: define ROI per use case (time saved, versions shipped, quality scored by humans).
Enforce human-in-the-loop for anything factual, legal, or brand-sensitive - especially citations and claims.
Pilot agentic systems in narrow, high-volume tasks; kill pilots that don't scale within 90 days.
Protect your pipeline: keep junior learning loops alive; rotate AI tasks to prevent skill atrophy and "workslop."

If you're upskilling teams on multimodal video, agent workflows, or prompt craft, browse current programs here: Latest AI Courses.