Google Marketing Live 2023 and what to do about AI-generated assets
Google announced generative AI for Search and Performance Max assets. Here is what was actually said, what has no date yet, and how to prepare your accounts before the tools reach you.
At Google Marketing Live on May 23, Google put generative AI at the center of its ads products. Most of the announcements describe tools that are arriving over the coming months, not features you can switch on today. That is a good reason to separate what was announced from what is available, and to decide now how much creative control you are willing to hand over.
What Google actually announced
Four items matter for performance advertisers, according to Google's own recap.
- A conversational experience in Google Ads. You give it a landing page and it summarizes the page and suggests keywords, headlines, descriptions, images and other assets for a Search campaign, which you review and edit before launching.
- Automatically created assets for Search. Google says these began rolling out in 2022 and build headlines and descriptions from your landing page and existing ads. The new part is generative AI that adapts Search ads to the context of each query, which Google says is coming "soon".
- Generative AI in Performance Max. You provide a website, and the system learns the brand and populates text and other assets. Google also says it will suggest AI-generated images across more inventory and formats. The announcement gives no availability date.
- New Performance Max goals. An enhanced new customer acquisition goal and a new re-engagement goal.
Google also described ads in its Search Generative Experience, which was available in Search Labs at the time. Over the coming months, it said, it would experiment with Search and Shopping ads integrated into the AI snapshot and conversational mode. That is an experiment, not a product to plan a budget around.
Treat Google's performance claim as a claim
Google says advertisers using Performance Max see on average over 18 percent more conversions at a similar cost per action, based on its own data from November and December 2022. That is Google's figure and it compares different kinds of setups, so it says little about what will happen in your account. Use it as a reason to test, not as a forecast.
Audit what your account already says about your brand
Automatically created assets already pull from your landing pages and ads, so the quality of those inputs matters today. Open the pages you send traffic to and read them as a stranger would. If the page makes a claim you cannot support, or uses outdated pricing, a system that writes headlines from it can repeat the problem at scale.
Do this before any of the new tools arrive:
- Review the asset reports for responsive search ads and note which headlines and descriptions are system-generated, if any appear.
- List the claims you never want in an ad (superlatives you cannot prove, pricing, regulated statements) and keep it where your team can see it.
- Make sure your landing pages state your offer, price and terms accurately, since they are the source material.
Plan the test before the tool is switched on
When generative asset creation reaches your account, run it as a controlled experiment. Pick one campaign with stable volume, duplicate it using the experiments feature, let the copy use generated assets and keep the original on manual ones. Change nothing else and run it for several weeks.
Judge it on conversions at your target cost, not on click-through rate alone. Generated copy can lift clicks without lifting results. Also read the generated assets yourself every week for the first month. A human check costs minutes and is the cheapest brand safeguard you have.
Decide your control line now
There are two reasonable positions. If your brand voice is tight or your category is regulated, keep final approval on every asset and use the tools only to propose drafts. If you run a broad catalog with thin creative resources, accept more automation but set guardrails: a banned-claims list, a regular review and a clear owner. What does not work is leaving the decision to whoever happens to see the setting first.
The practical takeaway is to use the time before these tools launch to clean up the inputs they will read and to write the test plan you will use to judge them.