At the end of 2025, I went back through that search company's ad lineup

A few days ago, a friend who runs ad buying grabbed coffee with me. He held his phone, brow furrowed tight.

He said he manages a dozen-plus accounts, and the past year he felt like he was chasing a train he could never catch. A new feature today, a dashboard redesign tomorrow, an English acronym he'd never heard of the day after.

I didn't rush to answer him. When I got home, I went through every ad feature that search giant released this year, from top to bottom.

After I finished, I felt oddly calm.

Why? Because on the surface, the year produced dozens of new things — flashy and scattered. But line them up next to each other, and you realize they're all saying the same thing. That thing, I'll get to slowly.

Start with Search

Search is this company's root.

You used to search a word, it gave you ten blue links. Now? It has AI write you a paragraph first, chewing up the answer and spoon-feeding it to you.

So where do the ads go?

Inside that AI-written answer.

Think about what that placement means. The user asks a real question, the AI gives a tidy answer, and the ad sits right underneath it. Head-on collision, nowhere to dodge.

And this year they pushed it one step further with something called AI Mode. What does that mean? The queries users used to dig deep on — the long, complex ones that ads could never reach — now they can carry ads too.

That was the first thing this year that stopped me for a second.

The deeper the user asks, the better you can catch them.

One button hands Search ads to AI

This next one — I think it's the most worth thinking seriously about for anyone in ad buying.

They built something called AI Max. Put plainly: you hand the big levers of Search ads to AI. Which keywords to bid on, what copy to write, how to combine them — AI decides.

It's now the fastest-growing product line inside this company's Search ads.

Why the growth? Because it's genuinely making decisions for you. You don't have to climb out of bed at midnight to tweak keywords anymore, and you don't have to agonize over whether to add that one long-tail term.

But just handing it over isn't enough — they left you a brake pedal too. This year they added an "experiments" feature, so you can test on a sliver of traffic first and scale only if it works. On the copy side, you can start reining in AI when it gets too far out of line.

There's another data point I made a note of.

A feature called Smart Bidding Exploration lets you loosen your ROAS (return on ad spend) target a bit, letting AI reach for traffic it didn't dare touch before. The result? Accounts that used it picked up 18% more new converting keyword categories on average, and 19% more conversions.

18. 19. Those two numbers aren't big. But in ad buying, that's found money.

Give AI a little slack, and it hands you a field you never plowed.

The product that felt most like a black box cracked open a few windows this year

Done with Search, on to the next.

There's a product in this industry I've always called "the black box" — Performance Max. You throw money in, it runs across every channel for you, and you have no idea where the money actually went. You only know how much came back.

Convenient, sure. But you don't sleep well at night.

This year, that black box threw open several windows at once.

First, you can now see which channel your money went into. Search, Shopping, video, display — broken out line by line. You can even drill down into each channel's format, metrics, and where the problems are.

Second, the search terms report is fully open. You used to have no visibility into which terms AI was buying for you — now it's crystal clear, as transparent as the Search ads you run yourself.

Third, you can add negative keywords now. Up to ten thousand per account. Any term that has nothing to do with your business — block it outright.

Fourth, "search themes" expanded from 25 to 50. The more hints you give AI, the tighter it runs.

Fifth, device, audience, and existing customers — you can start picking. You can even tell it: customers who lapsed but used to have a high average order value — bid more to win them back.

The black box didn't disappear, but it learned to report.

There's also a small move I found interesting. If you advertise for brick-and-mortar stores, this year the black box can make you show up on a navigation app's map. When someone's driving, your store becomes a pushed pin.

I smiled when I read that one.

Ads made it all the way to the steering wheel.

A video site that turned itself into a mall

One layer up: video.

This company owns a video site so big it has no edges. People watch it on TV, scroll short clips on their phones, watch and search at the same time, search and buy at the same time.

This year it did one thing: it made the TV screen shoppable. You're watching a show, the product on screen is one tap away from your cart. Screen feels too far? One tap sends it to your phone to keep browsing.

The most expensive ad slot — the homepage — got an upgrade too. It used to be a banner. Now it fills the whole screen, cinematic, taking over the biggest screen in your living room.

Then there's sports and culture. They packaged sporting events and cultural moments into "sponsorship bundles" — surrounding the viewer pre-roll, mid-roll, post-roll. You watch the game, the concert, the New Year's Eve special, it's there.

What caught my eye more: they built a Creator Center. Brands used to chase creators across tools and platforms to collaborate and reconcile data. Now it's all in one dashboard — discover creators' videos, see paid vs. organic performance, search for people, reach out.

The video site isn't satisfied with letting you watch. It wants you to buy, to find, to settle in and stay.

Demand Gen: the awkward middle child grew 26%

Demand Gen — honestly, I used to find the name a little awkward. Not quite Search, not quite video, stuck in between.

But this year I went back and looked at the data.

Over the past year, conversions per dollar spent grew 26%. Behind that: more than sixty AI adjustments.

26%.

I stared at that number for a while. This thing really took off.

What it filled in this year is all the stuff that actually matters in the trenches: you can set new-customer goals now, bid with target CPC (cost per click), run on third-party display networks, auto-generate video, choose to run only short-form video. You can attach a product feed and turn the ad into a little storefront. You can do cross-channel bidding that ties online and offline together. You can drop a checkout link straight into the cart.

One more detail — I think people running cross-platform campaigns will love this. It added a "platform-comparable conversions" column, with methodology aligned to other platforms. You can finally compare fairly in one table.

It didn't add gimmicks. It filled the holes that media buyers lose sleep over every day.

App: the iOS nut, cracked this year

App ads — the past two years, everyone got beat up by iOS. Attribution broken, conversions invisible, models guessing in the dark.

This year the company came back with three cuts.

First, on iOS, it uses on-device, privacy-safe event data to calculate conversions — optimizes on it, reports on it, without crossing the line.

Second, it connected with third-party attribution vendors — called ICM (integrated conversion measurement) — more real-time, more complete data.

Third, App installs coming from Search, Shopping, and the black box can now be attributed properly.

And on iOS, App ads finally got target ROAS bidding. That used to be Android-only.

More aggressive: App installs from web ads can now be measured too. You run a web campaign — how many people installed the app, what they did after installing — all strung together in one line.

The attribution black hole on iOS is being filled in, piece by piece.

Retail: you're not just slapping up a product image anymore

Retail — this year's direction is clear: let the product speak for itself.

They gave merchants a "Brand Profile" — in search results it now shows lifestyle images, video, promos, shipping, return policies. A dry line of text turned into a complete storefront facade.

Inside Merchant Center, generative AI proactively pushes product images and product videos that fit your brand's tone — one click and you're using them.

And memberships. Member pricing used to only work inside the merchant's own app. Now it can be attached directly to ads. When AI bids, it'll specifically bid up for high-value existing customers.

What ads do has changed. It went from pushing product out, to pulling the storefront in.

Measurement: how good your AI is depends entirely on how clean your data is

This part, I want to spend an extra beat on.

This year the company open-sourced something called Meridian. It's a marketing mix model — MMM for short.

What's a marketing mix model? It's reverse-accounting: you spent this much on TV, Search, video, out-of-home — which slice actually drove incremental results? Not clicks, not impressions — without that ad, how much revenue would you actually have lost?

This kind of model used to be a big-company luxury — a team of data scientists running it for half a year. Now it's open source, and it can calibrate against your incrementality experiments.

This is big-company weaponry, handed to small companies.

They also lowered the bar on incrementality testing. You used to have to spend a certain volume to measure causality — now you can test with less budget, and test an entire account at once.

On the data side, they built a "Data Manager" that pools data from your website, app, stores, and CRM into one reservoir to feed AI. The logic is simple: AI can be brilliant, but if the data you feed it is garbage, what it gives you back is garbage too.

One number I noted: accounts that connected their tagging gateway saw 14% more signal on average.

14%. Not a lot — but in ad buying, every percentage point is money.

Creative: if you write copy for a living, you should be nervous

Used to be, making a single ad creative meant finding a designer, a photographer, editing video, rewriting copy. Now, inside a backend called Asset Studio, you type into a chat box and get images and video out.

They stuffed a generative image model in there. Upload a product photo and it'll swap seasons, scenes, styles for you — up to five products arranged together at once, with textures and lighting filled in.

Image edits can be batched. Video can be auto-generated. When you're done, generate a link and send it to the team for review — no more files bouncing back and forth.

The barrier to creative got compressed down to: can you type?

Finally — the thing I was holding back at the start

Back to that friend from the beginning.

I told him: don't focus on the dozens of features this year. Group them together and you'll see they're all answering the same question.

What question?

"Can AI actually run my business into every crack and corner?"

In Search, AI catches the deepest questions your users ask. In video, AI turns the screen into a shelf. In the black box, AI runs across every channel for you — and then honestly tells you where the money went. On bidding, AI loosens things up, tests new keywords, wins back lapsed customers. On attribution, AI calculates the incrementality of every cent. On creative, AI makes the images and the video. Even in the backend, there are now two AI advisors — one for ads, one for analytics — and you can just chat with them by typing.

Dozens of features. One big net. What it catches is one thing: liberating people — or, more bluntly, squeezing them out — from every decision that's repetitive, mechanical, or based on gut-feel experience.

So I told that friend: you're not chasing that train because you're too slow.

It's because that train doesn't need a driver anymore.

Or more precisely — the driver it needs has changed. From someone who knows which buttons to push, to someone who knows which direction to go.

Figure out the direction. Hand the buttons to it.

That's the biggest thing I took away from reading this year's lineup.

Where it'll take the industry — I don't know. But I'm fairly sure of one thing: the people still agonizing over "will AI replace me" have probably already been replaced.

The question that actually matters is always the next one: so what step do I take now?