A Review of 2025. How My Predictions went and why AI believes its own Hype!

A year ago, I made eight predictions for marketing and media in 2025. I made them deliberately bold, against the consensus of the day.

At the time, AI was in a trough of disillusionment. Thinking models weren't yet mainstream. Image generation couldn't spell. The hype cycle had peaked and the hangover had set in. Most commentary was either breathless enthusiasm disconnected from reality, or premature obituaries for technologies that hadn't yet hit their stride.

So I tried to ask a different question: what do I know, from actually doing this work, that most others don't? Where is the consensus wrong? Making predictions is a useful discipline - it forces you to commit to a view, and twelve months later reality marks your homework.

I asked Claude to score me. Six hits, one partial, one miss. I'll take that. But the more interesting story isn't the scorecard - it's what the year revealed about the gap between people who understand AI from implementation and people who understand it from press releases.

The Survivors

My clearest hit was predicting Google would keep its limbs. Fourteen months ago, the consensus was existential threat: DOJ lawsuits, AI search competitors, the end of an era. In print and on stage, industry luminaries told me a breakup was coming. I disagreed publicly. Some of them now owe me another lobster dinner.

What I didn't predict was Sergey Brin emerging from whatever yacht he'd been meditating on to personally write code again. But of course he did - this is THE engineer gold rush. When your existential threat is also the greatest technical challenge of a generation, you don't send middle management. You send the founder who still remembers what a compiler is.

2025 showed us Google at its worst and its best. Privacy Sandbox was the worst - years of consultation, endless delays, an outcome that pleased nobody while conveniently protecting Google's interests. Nano Banana was the best - ship fast, let the internet name it something ridiculous, watch it go viral with 200 million image edits in weeks. When Google is scared, they remember how to be Google.

AI Overviews decimated SEO traffic. Content farms that spent years gaming the algorithm discovered the new algorithm doesn't need to send you traffic - it's already eaten your content. Google protected paid search revenue while letting organic publishers take the hit. Their valuable partnership with the open web (yeah right).

Google isn't just surviving. They're potentially the first ten trillion dollar company this decade if they win the AI race. Diversified, great models, unmatched distribution, the only credible challenger to Nvidia on chips. The obituaries were premature.

Microsoft had a different kind of great year. They faced their own breakup threat thirty years ago - they know how this works. Now they own roughly a quarter of OpenAI, they've hedged with Anthropic, and Copilot is everywhere. Most people will tell you Copilot is rubbish - and it was, but it's improving quickly and it's genuinely strong for infosec. The fact that it's deployed across enterprises even when mediocre tells you something about Microsoft's distribution power. Not sexy, but relentless. Classic Microsoft.

Here's the thing about both of them: the current administration has bet the farm on AI succeeding. They need American champions. Breaking up your best horses mid-race is a hard sell when you're trying to beat China.

AI Believes Its Own PR

Here's my thesis for what we learned this year: AI believes its own PR. And not just AI PR - all PR.

These models were trained on the entire internet - decades of corporate communications, press releases, marketing copy. Ask an AI about AI transformation and it'll tell you it's revolutionary. Ask it about Ryanair and it'll explain their commitment to customer experience. Ask it about Google and publishers and it'll describe their valuable partnership (yeah right).

Companies are asking AI how to implement AI, and AI is telling them it'll be transformative and straightforward. The models learned that everything is 'innovative' and 'customer-centric' because that's what every press release says. They're the world's most confident interns - they've read every company blog post ever written and none of the post-mortems.

This explains the gap between AI investment and AI results. Companies spending millions on tools and nothing on training. Buying the Ferrari, refusing to learn the gears. The technology works. The humans operating it are the bottleneck. But 'invest in boring internal training' doesn't get you on stage at Cannes.

Where AI Actually Landed

Different AI applications matured at different speeds this year, and the pattern is instructive.

Image and video creation had their breakthrough year for provability in 2024 - you could finally show boards and clients that this stuff worked. In 2025, they became sophisticated. The quality gap closed. The controls improved. What was a novelty became a production tool.

Media planning followed a similar but later curve - and this was my clearest 'hit' on substance. WPP Open, Publicis Marcel, Viant - it was all anyone discussed at Cannes. Not yet fully rolled out across the industry, but more than proven. The transformation is real. 2025 was the year media planning showed the world what AI can do. 2026 and beyond is when it dominates the discipline.

I was partially right on AI advertising. The giants are making hay - Google, Meta, Amazon, TikTok all have AI deeply integrated into their ad products. The service narrative from holding companies is now entirely AI-led. The mainstream impact has materialised. What hasn't happened yet is the rollout and integration across the industry. Early adopters are generating real value. Everyone else is still talking about it.

On agents: they were mentioned seven thousand times at Cannes and Adweek NYC. Specifics were harder to find. Amazon's Rufus has 250 million users and is driving ten billion dollars in incremental sales - that's not a demo, that's production at scale. But Rufus is Amazon, with their own chips, models and captive ecosystem. For most marketers, agents remain powerful for early adopters, promising for everyone else. The 'army of AI employees' narrative is seductive but simplistic. Watch this space with realistic expectations.

The Giants Who Missed

My big miss was Apple. I predicted they'd finally launch their major advertising product - they own the most valuable real estate in tech, they've been hiring for it, the market wants it. I knew they were working on it. At this point, Apple's strategy reminds me of someone who pays for an expensive gym membership and never goes. The intention is there. The equipment is world-class. And somehow another year passes. I should know - I too pay for a very nice gym.

But Apple weren't the only giants to miss the moment. Meta started the year strong - Llama looked genuinely promising, open source seemed like a credible strategy. Then it fell apart. Hiring disasters. An unravelling strategic narrative. The gap to frontier models kept widening. They'll be consoled that their ad product is still humming and they're selling some snazzy glasses, but they have a long way to run to win the AI war.

My biggest blind spot wasn't either of them though. It was the AI coding revolution. Cursor, Claude Code, GitHub Copilot - this is where AI actually delivers at scale right now. Not because the technology is different, but because the feedback loop is immediate. You write code, run it, it works or it doesn't. No ambiguity, no committees. The lesson for marketing: AI succeeds where feedback is fast and unambiguous.

Faking It Has Never Been Easier

2025 was the year of the AI consultant. Some of us have been doing this work for years. Others discovered the space six months ago, asked ChatGPT to write them a LinkedIn bio, and now charge significant day rates to tell enterprises to 'leverage synergies with large language models.'

Here's the uncomfortable truth: faking AI expertise has never been easier. The tools are articulate. The jargon is learnable. The demos are impressive. But adding real value - actually making this stuff work in production, at scale, generating returns - that requires the kind of knowledge you only get from doing the work.

Simple test: ask them how they overcame a serious problem on a project. If they don't have a story, they haven't actually done anything yet.

The problems matter because that's where learning happens. With a good team, initial setbacks lead to iteration and success. Companies winning with AI have invested in training, built feedback loops, understood this isn't software you roll out once. The ones struggling bought the PR - often from consultants who learned everything they know from AI that believed its own PR.

What's Next

The prediction game continues. Look out for my 2026 predictions in the next week or so. There are going to be some bold calls you won't read anywhere else.

In the meantime, the AI PR machine that believes its own hype thinks it's close to AGI. Wisdom would suggest that a system which can beat almost any maths exam but can't tell when it's being manipulated - or build a good-looking on-brand PowerPoint deck - might recognise it still has a thing or two to learn.

AI is a tool to help people do more. Not one to replace us. Not just yet.