5 different ideas: review, test, humor, story, comparison. In automated campaigns creative works like targeting: different ideas find different people. Copies are one idea to the algorithm.

What’s most likely the problem?

The platform can’t see purchases and learns from clicks. The signal is one of automation’s four inputs. Without purchase data the algorithm optimizes what it sees, clicks, and brings clickers.

Add negative keywords and check the search terms report. That’s the fence: negatives cut off irrelevant queries, and the weekly report shows where else money goes.

How should she check the promise?

An experiment: campaign with AI Max vs without. The official number is an average across all clients from the seller itself. Only an experiment with one difference gives an honest answer for you.

Automated campaigns: Advantage+, PMax, AI Max, Smart+

Lesson 12 of 16 in the course “Marketing 2027”. Free, with animations and a quiz.

The machine picks the audience now

Just a few years ago, setting up ads looked like this: you picked the age, city and interests like “loves streetwear” yourself. Now platforms increasingly do it for you. You say what you want, for example purchases, and the algorithm looks for people similar to those who already bought. At Meta, the system behind this is called Andromeda. Meta’s engineers described it in 2024: it picks a few thousand suitable ads out of tens of millions for a specific person in a fraction of a second. Why? The platform has far more signals about people than you do: what they watch, where they linger, what they buy. Your job changes: fewer manual checkboxes, more work on what the algorithm gets as input.

The four main automated campaigns

Let’s look at what the big platforms offer. Meta Advantage+ picks the audience, placements and creative combinations by itself. Google Performance Max, or PMax, has run a single campaign across all Google channels since 2021: Search, YouTube, Gmail, Maps and more. Google AI Max for Search arrived in May 2025: it finds queries broader than your keywords, writes headlines from your site and chooses which page to send people to. TikTok Smart+ launched in October 2024 and automates audience, bids, creative and budget allocation. The names differ but the logic is the same: you set the goal and provide assets, and AI decides who sees what and where.

What’s still in your hands

If AI picks people and placements itself, why do you need a marketer at all? Because automation has four inputs, and you decide their quality. First, creatives: the algorithm can only show what you uploaded. Second, the signal: if the platform can’t see real purchases, it learns from clicks and brings clickers, not buyers. Third, goal and budget: optimize for cheap views and you get cheap views. Fourth, guardrails: exclusions, negative keywords and rules so the AI doesn’t wander off. Think of a self-driving taxi: it steers itself, but you set the address, the fuel and the no-go streets. Automation won’t fix a bad input. It will just scale it faster.

Creative is the new targeting

Since you no longer pick interests, how does the algorithm know who to show the ad to? Largely from the creative itself. A video about a hoodie surviving 30 washes finds practical people. A funny sketch finds people who love humor. A real customer’s review finds the doubters. That’s why five copies of one video with different button colors are nearly useless: to the algorithm it’s one idea, and it finds the same people. Five different ideas, though — review, test, humor, story, comparison — work like five different audiences. That’s why the lessons on hooks and ad anatomy spent so long on different angles. In the age of automated campaigns, that’s your main lever.

Check platform numbers yourself

Every platform promises growth. Google says advertisers using AI Max typically see 14% more conversions at a similar cost. Meta reported that turning on Advantage+ AI targeting gave advertisers a 22% increase in ROAS; ROAS is how much revenue each dollar spent brings back. TikTok said that in a closed test of Smart+, 71% of advertisers saw better results in catalog ads. These are real claims, but they’re made by the seller about its own product, and averaged across all clients. Your results may differ. The fix is simple and familiar from the last lesson: run an experiment “with automation” vs “without.” Google Ads has experiments for this, Meta has A/B tests. Trust your own numbers, not someone else’s averages.

Build a fence

Automation searches wide and sometimes finds the wrong thing. This example is made up but very realistic. Nika’s brand is called “Pushka”, Russian for “cannon” and slang for a gun, and AI Max could easily decide that a search like “pushka buy a barrel” is about her hoodies. Money goes to people who don’t need clothes. So build a fence. Negative keywords: “weapon”, “barrel”. Brand and URL controls: where you can appear and where to send people. Text rules: since April 2026 Google has AI Brief, where you write in your own words what ads may and may not say and see examples before launch. And once a week, open the search terms report: it shows which phrases actually triggered your ads. Five minutes a week saves days of budget.

Do it yourself: launch an automated campaign

Before launch, check the signal: make a test purchase and confirm the platform saw it. Without that, automation learns from noise. Upload at least five different creative ideas, not five variations of one. Build the fence: negative keywords, exclusions, text rules. Run an experiment “with automation” against your usual campaign, and decide in advance how long it runs and which metric decides. Once a week, open the report: which queries and placements showed your ads, which creatives got the money. If something looks odd, adjust the fence. In a month you won’t have faith in the platform’s promises, you’ll have your own answer about whether automation works for you.

Check yourself

What will help more?
Correct answer: 5 different ideas: review, test, humor, story, comparison. In automated campaigns creative works like targeting: different ideas find different people. Copies are one idea to the algorithm.
What’s most likely the problem?
Correct answer: The platform can’t see purchases and learns from clicks. The signal is one of automation’s four inputs. Without purchase data the algorithm optimizes what it sees, clicks, and brings clickers.
What should she do?
Correct answer: Add negative keywords and check the search terms report. That’s the fence: negatives cut off irrelevant queries, and the weekly report shows where else money goes.
How should she check the promise?
Correct answer: An experiment: campaign with AI Max vs without. The official number is an average across all clients from the seller itself. Only an experiment with one difference gives an honest answer for you.

← Testing without fooling yourself MCP: an agent connects to your ad account →