SEM in 2026: what actually changed when the platforms went AI-first

Google's Performance Max and Meta's Advantage+ didn't just add automation to campaigns. They moved the important decisions inside the machine. Performance Max is Google's campaign type that runs across Search, YouTube, Maps, and the rest of their inventory from a single campaign, with the system choosing placements and bids. Advantage+ is Meta's version of the same idea. In both cases, the targeting and bidding levers that used to fill an operator's week are now handled by the platform.

The old playbook of tight keyword control and manual bid adjustments doesn't win anymore. But "hand it to the AI and walk away" isn't the answer either. The operators doing well in 2026 are the ones who understand which inputs the algorithm actually learns from, and who spend their time there.

The input layer matters more than the campaign layer

An automated campaign is a function of what you feed it. Three inputs dominate.

Conversion quality. If the account optimizes for form fills, the system will find you more form fillers, not more customers. The fix is feeding real sales outcomes back into the platform, which I covered in detail in the offline conversion import piece. This used to be a nice-to-have. In an AI-driven account it's the steering wheel, because the conversion data is the only place you still tell the machine what "good" means.

Audience signals. First-party data, meaning your email list, your purchase history, your CRM segments, gives the system a starting picture of who your customer is instead of making it guess from scratch. No competitor can copy your customer list. It's one of the few durable advantages left in the auction.

Creative variety. The platforms now assemble and test combinations on their own, which means they need genuinely different raw material to work with. Several distinct angles beat several polished variations of the same angle, because near-duplicates give the system nothing to learn from.

What stopped working

The habits that defined a well-run account a few years ago now mostly starve the algorithm.

Tight match-type control on Search made sense when you were doing the targeting. Now it fences off the queries the system would have found for you. Granular ad set segmentation on Meta splits your conversion data into slices too thin to learn from, so every slice performs worse than the whole would have. Manual placement curation, dayparting spreadsheets, and bid adjustments layered on top of automated bidding fall into the same category: effort spent overriding a system that has more data than you do.

None of this means the platforms are always right. It means the way to correct them changed. You correct them with better inputs, not with more settings.

What started working

The leverage moved to three places, and all of them sit outside the campaign settings screen.

The offer. What you're actually selling, how it's priced, and how the landing page presents it now matter more than any toggle in the account, because the machine can find your audience but it can't make your offer worth clicking. Creative testing became the main form of optimization, since messaging is the biggest input you still fully control. And tracking architecture became a foundation rather than an afterthought, because every automated decision downstream inherits the quality of the conversion data upstream.

The playbook, in order

1. Rebuild conversion tracking first. Every action that matters to revenue gets tracked and valued, and outcomes from your CRM flow back into the account. Nothing else on this list works without it.
2. Consolidate. Fewer campaigns with more data each learn faster than many campaigns splitting the same budget. Consolidation feels like giving up control. What you're giving up is fragmentation.
3. Feed it distinct creative angles. Different hooks, different formats, different arguments for buying. Judge them on cost per real conversion, not clicks.
4. Let budget move. Allow the system to allocate across audiences and placements inside a campaign instead of fencing every dollar by hand.
5. Review at the asset level, weekly. The campaign-level numbers smooth everything into mush. The asset and search-term reports are where you catch the machine drifting, and the review has to happen on a schedule.

Most of this list is standard paid ads management work now. The weekly reviews and monitoring are the part I hand to AI agents and automation, so they run on a calendar instead of a memory.

What this means for the operator

The job description changed. Less time pulling levers inside the account, more time on the things the account depends on: the tracking, the offer, the creative pipeline, and the judgment call of when the automation is drifting and needs correcting. That last one is where experience still pays. The platforms optimize toward the goal you set, ruthlessly, including when the goal is set wrong.

I don't think of this as believing in AI or not believing in it. The machines run the auction either way. The only question is whether your account is feeding them signal or noise, and that part is still entirely up to you.