AI can make more ads. Your brief still needs to make sense.

What changed

In May 2026, Google announced new Asset Studio capabilities for building and refining advertising creative from a brief, brand guidance and business context. Availability and supported features should be checked in the relevant advertising account. The announcement makes a useful discussion timely: producing more variations is only valuable when the variations test an idea.

Write the question before the prompt

Our recommended starting point is a short test brief: who is the audience, what decision are they making, and what do we want to learn? For a publishing business, one concept might explain package clarity while another introduces the production process. Those are different reasons to enquire. Changing only a background colour may be less informative if the underlying message is still unclear.

Keep the product promise accurate

AI-assisted visuals and copy need human review. Check that the depicted product, service scope, price context and offer conditions match reality. Do not turn an illustrative image into supposed evidence of a completed client project. For healthcare, education and other specialist services, have the appropriate subject expert review factual claims before the campaign is released.

Organise variants so the result is readable

Give each concept a clear name and record its hook, format and intended audience. Keep a record of which assets actually ran. Review the available conversion and lead-quality feedback alongside attention metrics. If the landing page or sales process changes during the test, record that too; otherwise it becomes difficult to interpret what caused the difference.

Keep the next experiment small

At the end of a review, decide what deserves another test, what should be revised and what no longer supports the offer. A useful creative library captures learning as well as finished files. First Thought’s AI workflows are an emerging area: any pilot should have a clear scope, human review and a practical definition of usefulness.

A working brief for your next experiment

Write the audience situation in one sentence, followed by the objection or question the creative should answer. Then list the approved facts, visual assets and claims that must remain accurate. Define the action the viewer can take and the page or form they will reach. This becomes the reference against which both human-created and AI-assisted variations are reviewed.

For a hypothetical product demonstration, the team might compare an opening focused on how the item is used with an opening focused on a specific feature. Keep the product description and offer consistent. If one version changes the price, the audience and the landing page as well, the comparison becomes much harder to interpret.

Build a review gate before publishing

Have a person check product identity, readable text, brand fit, factual accuracy and the relationship between the image and the offer. Record the approved version so the published asset can be traced back to the brief. Do not assume an attractive generated image accurately represents dimensions, materials or the service experience.

The review should also consider whether an asset could be mistaken for real project evidence. A conceptual illustration may support a service explanation, but it should not imply that a pictured building, person or customer outcome belongs to the business. Good creative direction includes deciding what an image should not claim.

Capture learning, not just finished files

Maintain a simple concept log: the question, the assets, the running period, the available outcomes and the next decision. Note limitations such as low volume or incomplete sales feedback. Over time, this gives the team a more useful resource than a folder of unexplained designs. The value of faster production is the room it creates for better testing and judgment.