Case Study: Adapting Activity to Meta's AI Audience Changes

Over the past year, Meta has increasingly embedded AI into its audience targeting tools. Many interests have been removed from targeting, and it’s no longer possible to exclude interests either. While some organisations have reported that they have seen Lookalike audiences failing to perform as strongly as they had in the past.

This has led to a shift towards broad targeting, with Meta itself often pushing for ad set up that allows the ad system to fully decide who is best suited to see ads. By the end of 2026 Meta has also reported it plans to fully automate advertising with AI. Advertisers will only need to provide a product image and a budget, and Meta's AI will apparently generate the ad, including image, video and text, and then determine user targeting.

The shift to more broad targeting has been noticeable for a number of our clients, with broad national (essentially no audience targeting) accounting for 76% of one-off donor recruitment results in 2024 for a major German aid charity - and achieving the lowest cost per acquisition.

But does it always perform best and how did we have to adapt ad set-up to best utilise the changes? Here are some findings from our work with our client in Germany, who works in the field of international emergency aid:

1. Niche and focused campaigns still performed better with targeted audiences

While donation campaigns with broad appeal worked well with national targeting, we found that more specific campaigns, such as those aimed at recruiting regular donors, performed best with more refined targeting. In particular, narrowing audiences by age yielded strong results.

2. Consolidating activity proved most effective

During our most successful period this client had only three ads live, in two ad sets. And this is a major charity with large budgets! We found that consolidating the budget and slowly phasing in more and crucially high quality creative worked well.

Separating content by ad set allowed us to maintain greater manual control, to optimise more effectively under broad targeting by shifting budgets between content pieces based on performance.

The consolidation and phasing approach may have also reduced the risk of multiple broad ad sets competing with each other, with Meta's AI potentially targeting the same users. Another charity client, for example, found that launching a second broad targeting donate campaign coincided with a drop in results from their main donate campaign.

3. Utilised other ways to 'control' board targeting ad sets.

Over time, ad performance tends to decline as costs rise. To counter this we experimented with adding target costs and bid caps. We found that introducing target costs - especially in long-running and well-optimised campaigns - helped to significantly reduce cost per acquisition while maintaining the average donation value.

4. Separating ad sets by platform worked well - but only on campaigns with large budgets.

Over the years, our client has seen a gradual shift toward Instagram, which now accounts for around 70% of their Meta donor recruitment. This is likely driven by the increasing use of video-based ads, which dominate their creative strategy.

But while the AI driven targeting seems to 'push' this client to Facebook placements more, we found that separating ad sets based on platform helped us in campaigns that were perhaps more suited to a Facebook audience (potentially older demographics).

Conclusion

Utilising the AI targeting features did work well for our international aid client, when implemented with a practice of manual controls. Overall as Germany has an extremely strong interpretation of GDPR and high rates of data opt out (a growing global trend), we especially couldn't rely on Meta to fully optimise. Instead UTM actual result reports were essential to help us manually allocate budget to content and ad sets that performed well.

While audience selection became less important, content was still and increasingly became the leading factor of success, with high-quality videos 30 seconds or more performing the best.

Are you interested in running rigorous, evidence-based tests to see if AI targeting will improve your Meta Ads? That’s something we can help you with - book a call with Nick today.