Performance marketing in the AI era: human versus machine

Ten years ago, a large part of my working day consisted of adjusting bids, refining keywords and excluding placements. Those buttons have largely disappeared. In Performance Max, AI Max and Advantage+, the algorithm decides where, when and to whom your ad appears, and it does so at a scale no human can follow. So the question is no longer whether you advertise with AI, but what work is left and who does it. This article sets out that division of labour: what the machine takes over in the major advertising platforms, which work becomes heavier as a result, and how to monitor such a campaign without falling back into micromanagement.

Chatbots, conversational marketing and visibility in AI search engines also belong to the AI era, but they are a different subject. Those are covered in AI chatbots for customer service and in our GEO strategies.

What does AI actually change about performance marketing?

AI moves the work from execution to steering. Bidding, search term matching, audience selection and placement are automated. What remains is the quality of your inputs: which conversions you measure, which value you pass on, which creatives you supply and what you exclude. According to Google, in the first quarter of 2026 more than 30% of customers’ search advertising spend ran through AI campaign types.

That sounds like less work, and it rarely is. The work shifts to the places where mistakes have become more expensive. A badly configured conversion used to produce a skewed report; today it drives the entire bidding behaviour of your campaign. A weak ad text used to cost click-through rate; today it is one of the few signals you still fully determine yourself.

ClickForest manages performance marketing for SMEs in Flanders and keeps seeing the same split: accounts where someone guards the inputs perform better than accounts where someone guards the settings. How this fits into the wider field is set out in our pillar on what performance marketing is.

Which AI campaign types are you steering today?

Four names cover most of a Flemish SME’s budget. At Google those are Performance Max, AI Max and Demand Gen; at Meta it is Advantage+. They differ in reach and in how much you can still set manually, but they share the same principle: you supply goals, data and creatives, the system decides on bidding, matching and placement.

Campaign type	What it is	What you supply
Performance Max (Google)	One campaign that automatically advertises across all Google channels at once, from Search and Shopping to YouTube, Display, Gmail, Discover and Maps	Conversion values, asset groups per theme, search themes, exclusions
AI Max (Google)	A feature set within your existing Search campaigns with three parts: broader search term matching, automatically customised ad text and automatic landing page selection	Keywords as a basis, exclusion lists, landing pages
Demand Gen (Google)	Visual campaigns on YouTube, Discover and Gmail (according to Google’s documentation now also Maps and the Display Network), aimed at creating demand rather than harvesting it	Image and video, audience signals, conversion goal
Advantage+ (Meta)	Automated campaigns on Facebook and Instagram in which audience, placement and creative combinations are determined by the system	Catalogue or offer, creatives, audience signals and exclusions

The direction is clear: Google is retiring the classic alternatives. AI Max has been generally available since April 2026, automatically created assets and the campaign-level broad match setting migrate to AI Max from September 2026, and for Dynamic Search Ads Google pushed that automatic migration back to February 2027 in June 2026. That delay followed pressure from advertisers:

“We’ve heard your feedback loud and clear: you need more time to transition from Dynamic Search Ads (DSA) to AI Max.”

— Ginny Marvin, Google Ads Liaison at Google, June 2026

So you are not choosing whether you work with this, only when and how controlled. The practical rollout order is in our guide to Google Ads optimisation.

What does the AI take over and what do you keep doing?

The machine takes over everything that has to be repeated across millions of auctions a day: bidding, matching, finding audiences, allocating budget, combining variants. The human keeps everything that requires meaning: deciding what counts as a result, what that result is worth, what your brand says and where your limit lies. That split is more useful than the question of whether AI replaces marketers.

Task	The AI does	You do
Bidding	Sets the bid per auction based on signals such as device, location, time and query	Choose the bid strategy and set realistic target values
Search term matching	Automatically broadens to related queries and phrasings	Read the search term reports and exclude what does not fit
Defining audiences	Looks for profiles resembling those who convert	Supply first-party lists, signals and exclusions
Placement and channel mix	Distributes budget across channels, formats and moments	Check the channel reporting and correct any drift
Ad text and assets	Generates variants and combines what performs best	Set brand voice, offer and proof, and approve or reject
Conversion measurement	Models what consent or tracking loses	Decide what a conversion is and what value it gets
Profitability	Optimises on the value you pass on	Work through margins and steer on profit instead of revenue
Strategy and positioning	Nothing	Choose offer, price, audience, channels and what you deliberately skip

Note the bottom row. An algorithm optimises within the brief it is given, it does not invent that brief. Anyone who puts a weak offer or an unclear positioning into an AI campaign gets an efficiently distributed weak offer back. That is exactly why the debate about AI and human creativity also plays out in advertising.

Why has your measurement become the most important button?

Because the AI optimises on precisely what you pass on as a conversion. Measure the wrong action or the wrong value and the system will scale that mistake with full conviction. Measurement is therefore no longer a precondition but the steering itself: it is where you tell the algorithm what success means.

Three things decide whether that steering is right. One: what counts as a conversion. Putting a newsletter sign-up and a quote request in the same conversion action means the system chases both equally hard. Two: whether your measurement holds up. For European traffic Google requires consent signals via Consent Mode to keep feeding conversion measurement, personalisation and remarketing, and what drops out is modelled. Three: which value you send along.

That third one is the biggest lever, and the most misunderstood. Google Ads has no setting to optimise for profit: no POAS bid strategy, no checkbox. The route that works is to pass profit value rather than revenue through as conversion value and combine it with Target ROAS. If you want to verify those profit figures independently, link your Google Merchant Center and fill in the cost_of_goods_sold attribute in your product feed; via conversions with cart data, Google Ads then reports gross profit per campaign and per product. The worked example is in POAS versus ROAS, and why POAS is a fairer compass than ROAS becomes clear as soon as your margins differ per product.

What is left of audiences and targeting?

Less than you might think, and that is largely intentional. Google’s similar audiences have been phased out in favour of optimized targeting, where the system itself looks for profiles resembling those who convert. At Meta, the Advantage+ audience takes over that role: your audience setting becomes a starting signal rather than a hard boundary.

That shifts your work to the input. An AI that finds its own audiences needs material to start from, and that material is your first-party data: customer lists from your CRM, high-value buyers, newsletter subscribers, visitors who requested a quote. A list of your thousand best customers now carries more weight than the sharpest interest targeting of five years ago.

The other side matters just as much: exclusion. Existing customers you do not want to pay for again, applicants, suppliers, regions you do not deliver to. The AI does not know your business logic, so what you do not exclude, it may well buy. How to spread those layers across the full funnel is set out in our guide to advertising strategy through the marketing funnel; the division of roles between paid and organic is in integrating SEA and SEO.

Why do creatives become your biggest lever?

Because they are one of the few inputs the algorithm cannot invent for you. In AI campaigns the system decides which combination of text, image and video it shows, but it can only choose from what you supply. A campaign with three weak images and one ad text gives the AI almost nothing to optimise with, however well your measurement is set up.

The platforms are pushing hard in that direction themselves. For the first quarter of 2026, Meta reported that more than 8 million advertisers use at least one of its generative AI tools for ad creative. Useful for producing variants, but it moves the bottleneck: if everyone uses the same generator, the difference between ads becomes a human choice again.

In practice that means three things. Supply variation that genuinely differs, so a different offer, a different objection, a different format, and not the same message five times in another colour. Refresh on a rhythm, because creative fatigue in automated campaigns is a matter of weeks rather than months. And keep the final edit human: a generator does not know your claims, your legal limits or your brand voice. What an ad is allowed to say remains your decision.

That connects to what happens after the click. An AI campaign performing well on a weak landing page mostly pushes money through your account, and there conversion optimisation helps more than an extra bid strategy. How personalisation and CRO strengthen each other is covered in CRO with AI and personalisation.

How do you monitor an AI campaign without losing control?

With a fixed rhythm instead of daily intervention. AI bid strategies have a learning period, so making major changes every day resets that learning process again and again. Judge on weekly averages instead of daily figures, never compare a learning period with a stable one, and keep one structural experiment open per quarter instead of five at a time.

The five checks that matter most in practice:

• Search terms, weekly. AI Max and Performance Max broaden noticeably. Read the reports and update your exclusion lists before the money leaks away.
• Keep brand traffic separate. Searches on your own name convert anyway. Do not let an AI campaign claim them as its own achievement without you noticing.
• Channel mix, monthly. Google reports per channel what a Performance Max campaign spends and returns. A campaign that gradually becomes a YouTube campaign is a different campaign from the one you thought you were buying.
• Recheck your values, quarterly. Margins change, shipping costs change, return rates change. Your submitted conversion values cannot ignore that.
• Put profit next to platform figures. A campaign that looks like it performs better while your gross profit falls is not an improvement.

What such a routine looks like month after month at a Flemish SME is shown in the advertising story of Bastiano. If you doubt you can keep up that rhythm yourself, outsourcing is a legitimate consideration.

Where does AI steering go wrong?

Rarely in the settings, almost always in the assumptions around them. The four mistakes we see most often: believing the campaign will work out for itself what is profitable, drawing conclusions over a period that is still learning, reading dashboard figures as bookkeeping, and judging the campaign without looking at what happens in the business.

The platform figures themselves also call for a critical eye. In its quarterly call for the third quarter of 2025, Meta reported that advertisers running Advantage+ lead campaigns see on average 14% lower cost per lead than advertisers who do not. That is a comparison between two groups of users, not a controlled test, so it says little about what would happen in your account. Such figures are direction markers, not promises; our broader reservations about the channel are in the downside of digital advertising.

Put the independent measurements next to that. Google itself states that advertisers activating AI Max typically see 14% more conversions or conversion value at a similar CPA, based on its own internal data. Smarter Ecommerce analysed a million AI Max impressions across 600 active accounts and found a median 13% rise in conversion value, but at a 16% higher cost per acquisition, with outliers from 42% above to 35% below baseline. Their Head of Ecommerce Insights sums it up as a coin toss. The same reverse lens applies there, because that company sells automation software itself. The point is not which figure wins, but that the gap between “similar CPA” and “16% more expensive” is exactly the space your own measurement has to fill.

Finally, a limit no platform guards for you. Since 2 August 2026 the transparency obligations of Article 50 of the European AI Act apply: anyone deploying an AI system that interacts directly with people has to make that clear, and generative output has to be marked in machine-readable form as AI-generated, with a transition period until 2 December 2026 for systems already on the market. The marking obligation sits with the providers of those systems, but the claims in your ads remain your responsibility either way. ClickForest therefore keeps the final edit human on all ad copy that goes live, even when the first version was written by a model.

What should you remember about performance marketing in the AI era?

The machine has taken over execution: bidding, matching, finding audiences, allocating budget. What stays with you is everything that carries meaning: deciding what a conversion is and what it is worth, passing on profit instead of revenue, supplying creatives that genuinely differ, excluding what does not fit your business, and putting the whole thing next to your real figures on a fixed rhythm. Anyone who does those five things gets more out of an AI campaign than someone stacking advanced settings on a messy foundation.

ClickForest turns performance marketing into measurable growth for SMEs in Flanders, from measurement and profit steering to campaign management and conversion optimisation. Curious whether your AI campaigns steer on the right value? Book a video call without obligation and we will look at your account together.