AI advertising in 2026: what to automate and what to keep in hand

Where Meta and Google stand

At Meta, Advantage+ automates targeting, budget allocation, placements and part of the creative work. In June 2026, Meta announced an end-to-end creative space with a brand memory, and its Muse Image model is coming to Advantage+ creative.

At Google, AI Max for Search campaigns automates query matching, text customisation and landing page selection.

What AI does well

Platform systems are effective on fast, repeated decisions:

• Finding audiences from the conversions already obtained.
• Producing and testing creative variations.
• Spreading delivery across placements and formats.
• Adjusting bids in real time.

What remains a human decision

Other decisions need context the platform does not have:

• The business goal and the conversion event to optimise for.
• The offer, the price and the terms.
• Advertising claims and compliance, especially in regulated sectors.
• Brand consistency.
• The budget split between channels, based on actual margin and revenue.
• How sales handles leads after the form.

The rule that ties it together: signal quality

Every algorithm optimises what it is sent. If the only signal is a completed form, the AI will find more forms, not more customers. Sending qualified leads and sales back to the platforms, through Meta’s Conversions API and Google Ads conversion imports, is the setting that makes all the automation useful.

A grid to decide

Two questions are enough to sort a decision. Can it be undone easily? And how much does a mistake cost?

• Reversible and cheap (a bid, a placement): automate.
• Reversible but visible (a text, a visual): automate with a review step.
• Hard to undo (a claim, a price, a brand image): keep human.
• Structural (the goal, the total budget, the choice of channels): decide as a team, with CRM numbers in hand.

Reorganise roles in the team

When manual settings disappear, three responsibilities gain value: guaranteeing the quality of the signal sent to platforms, defining the creative strategy and the angles to test, and reading results all the way to revenue.

These roles can be held by one person in a small organisation, but they must be named. Automation without an owner always ends up optimising something nobody chose.

Keep a decision log

Automated systems change many things without saying so. A simple dated log makes results readable: what was changed, why, and what was expected.

• Changes of goal or conversion event.
• New automation features turned on.
• Attribution changes announced by platforms.
• Changes to the site, the form or the CRM.

What to keep learning

Delegating to a machine does not remove the need to understand what it does. Reading attribution, computing a margin, checking tracking and writing a brief remain essential: these are the skills that let you control automation, and contradict it when it is wrong.

Questions