Advertising Is Being Reinvented by AI

A few days ago, a friend of mine who runs an e-commerce business complained to me.

He said he has a media buyer on his team who, every morning at nine, opens up the computer, checks yesterday's Google Ads spend, looks at how the Meta campaigns performed, and then starts adjusting bids, swapping out creatives, and shifting budgets. Over the course of a day, just watching the dashboards eats up four or five hours.

I asked him: how much ROAS does this buyer deliver for you in a month?

He thought about it and said, around 2.4x.

I said, have you ever considered that this whole process is being reinvented?

What Is the Machine Actually Doing?

When a lot of people hear "AI advertising," their first reaction is: isn't that just programmatic buying? That's been around for a decade.

No, it isn't.

Programmatic buying ten years ago meant humans wrote the rules and machines followed them. You set "bids no higher than 5 yuan, target women aged 25-35," and the machine dutifully operated within those constraints.

What we're calling AI advertising today is the machine deciding for itself how to bid.

Every time it enters an ad auction, it simultaneously looks at tens of thousands of signals. What phone is this person using? What time are they online? What did they just search for? What did they buy last week? Do they resemble your existing customers? After taking all that in, it calculates a probability: if this person sees your ad in this exact second, will they place an order? Then it decides on its own how much to bid.

Google has disclosed a figure: Smart Bidding processes over 70 million signals per auction.

Think about it — how many signals can a single person process in a day? A few dozen? A few hundred? Tops.

That's the essence of the gap.

It's not that the machine is smarter than the human. It's that the machine is orders of magnitude faster and can see far more dimensions.

Eight Jobs the Machine Handles For You

So how exactly does AI break down the advertising process?

Let me tell you — it has essentially compressed an entire media-buying team into a single system.

First, bidding. Humans used to set bids based on experience and intuition. Now the machine bids independently in each auction, tailoring the bid to each individual. Google has tested this internally — compared to manual bidding, it delivers an average of 20% more conversions.

Second, creative combination. You feed it a batch of headlines, descriptions, images, and videos, and it mixes and matches them, then picks the version most likely to get a click for each specific user. Meta's Dynamic Ads and Google's Responsive Search Ads both work this way. Advertisers typically see CTR (Click-Through Rate) that's 15% to 30% higher than with static creatives.

Third, finding people. Not the "I want women aged 25-35" kind of audience targeting. The machine ingests your CRM (Customer Relationship Management) customer data, learns what type of user is most likely to buy, and then goes out into the vast sea of people and finds lookalike audiences. With this approach to new customer acquisition, the customer acquisition cost typically drops 25% to 40%.

Fourth, monitoring. The machine scans all metrics every few minutes. Which ad group has gone off the rails? Which creative are users getting tired of? Which new audience is popping up? It spots these before you do.

Fifth, cross-platform coordination. You're running ads simultaneously on Google, Meta, TikTok, and LinkedIn — the machine watches over all of them. It makes sure two platforms aren't fighting over the same person, and that budget isn't wasted on redundant impressions.

Sixth, creative production. Large language models read your historically high-performing ads, learn what kind of copy actually works, and then generate new headlines, new descriptions, and even product images and short videos. This solves a perennial headache in the advertising industry — creative production capacity can't keep up.

Seventh, fraud detection. Fake clicks, bot traffic, invalid conversions — the machine identifies them in real time and blocks them outright. This is not something you can catch by manually watching the dashboard; the fraud tactics are far too sophisticated.

Eighth, accounting. A single user, from the first time they see your ad to when they place an order, may have crossed three or four platforms and seen your ad seven or eight times. Who gets the credit for this conversion? AI attribution models can trace this winding path and sort it out. Once it's sorted, many advertisers discover their actual ROAS is 15% to 25% higher than what was on the books. They had been underestimating the contribution of upper-funnel efforts.

You see, these eight functions combined are essentially the day-to-day work of an entire ad-buying team.

Everyone's Running, But on Different Tracks

So where do you use these capabilities?

There are mainly five platforms, but each excels in different directions.

Google Ads has gone the deepest. Its Performance Max integrates search, display, YouTube, Gmail, and Maps into a single campaign that runs across the entire network. Within Smart Bidding, Target CPA and Target ROAS let the machine bid based on the target cost and return you specify. For capturing search intent, Google is the first choice.

Meta excels at audience discovery and creative distribution. Advantage+ automatically expands audiences and shifts budgets, while Dynamic Ads serves the corresponding product based on user behavior. Its bread and butter is picking out the people who actually want to buy from within a large crowd that all looks the same.

TikTok places its bet on creativity. It uses AI to identify which elements in short videos drive engagement, then automatically scales the content that performs well. For younger audiences and video creatives, TikTok has a unique edge.

LinkedIn is irreplaceable in B2B. Its predictive audiences and automated bidding, paired with professional background data, are the standard setup for targeting enterprise clients.

Amazon DSP knows best what you want to buy. Shopping intent data combined with lookalike audiences makes it the most effective for e-commerce conversions.

What's the difference? Google captures intent, Meta captures discovery, TikTok captures creativity, LinkedIn captures B2B, and Amazon captures purchase signals. What you need to do isn't pick just one — it's to figure out which of these your business depends on.

Getting Started: Five Steps, In Order

So how do you actually put this into practice? I've laid out five steps for you. Follow them in order — don't skip ahead.

1. Get your tracking in order first.

AI runs on data. Feed it garbage and you'll get garbage out. Start by checking Google Analytics, the Meta Pixel, and all platform conversion tracking. Turn on enhanced conversions on the Google side, and set up the Conversions API on the Meta side. The dollar value of your conversion events must reflect real business value — don't just set a one-dollar placeholder for "purchase completed."

If this step isn't done right, everything after it is wasted.

2. Start with the AI built into the platforms.

Don't bring out the heavy weapons right away. First, enable Target CPA or Target ROAS on your existing campaigns, and turn on automatic placements on Meta. Let the AI start running on a small scale, and observe for two to three weeks.

3. Move to advanced campaigns.

Once basic automation is producing results, launch Google's Performance Max and Meta's Advantage+. These campaign types require almost no manual management, but they demand high creative quality and clear conversion goals. I recommend starting with 20% to 30% of your budget as a test and keeping the bulk of it in your proven existing campaigns as a control.

4. Shift your budget toward AI.

Once AI performance consistently beats manual results, move more budget over. A healthy structure is 70% to 80% in automated campaigns, with 20% to 30% left for manual work to test new strategies and new audiences. The machine learns from history — new territory still needs to be explored by humans.

5. Only consider third-party tools if you really need them.

If you're running multiple platforms simultaneously and cross-channel budget allocation, unified attribution, and cross-platform reporting are giving you headaches, then you can look into specialized third-party AI advertising platforms. But be prepared — these tools typically require a monthly ad spend of $10,000 to $50,000 to be worth the investment.

What Now?

A lot of people in ad buying ask: am I about to be replaced?

Honestly, what's being replaced isn't the media buyer role — it's the work style of "manually watching dashboards and adjusting bids by hand."

A media buyer who used to spend four or five hours a day monitoring dashboards might now only need half an hour to review the machine's recommendations. The time freed up should go toward the things machines can't do. Figuring out which audience you should actually be targeting. Understanding where your product differs from the competition. Defining the strategy for this campaign.

AI takes over execution, not judgment.

If you're someone who only knows how to execute, then yes, you should be worried. But if you can make judgment calls, set direction, and define what "good performance" means, then this wave is a good thing for you.

The machine does the grunt work for you. You finally have time to think about where your money should actually go.

But one thing needs to be said upfront.

AI is not a free lunch. It needs data to learn and budget to run. Accounts with a monthly budget under one or two thousand dollars that can't even scrape together 15 conversions a week — the machine simply can't learn. An algorithm running on an empty stomach produces results worse than a human.

So don't treat it as gospel, and don't resist it either. Build a solid data foundation first, then let the machine take over bit by bit — let the results speak.

These days, the worst mistake you can make in marketing is fighting today's war with a decade-old playbook.