Why your audience settings matter less than they used to
If you have run Meta Ads for a few years, you probably built your results on audiences: interest stacks, lookalikes, layered exclusions. For a long time that was the main lever. In 2026 it is increasingly not. Meta’s ad delivery now leans on an AI system called Andromeda, and the practical message to advertisers is consistent: go broader, restrict less, and let the system find buyers.
That does not mean targeting is dead. It means the thing that does the targeting has moved. Instead of you telling Meta who to show an ad to, the ad itself increasingly tells Meta who it is for. Understanding that shift is the difference between campaigns that quietly stagnate and campaigns that keep scaling.
What Andromeda actually is
Andromeda is Meta’s ad retrieval engine. It sits at the very first stage of ad delivery, before the auction itself. Out of a huge pool of candidate ads, it narrows things down to the few thousand ads that are eligible to enter a given auction for a given person. Only after that do the familiar auction mechanics (bid, estimated action rate, quality) decide the winner.
The important detail is how it chooses. Rather than relying mainly on the audience boxes an advertiser ticked, it reads the creative itself (the visuals, the message, the format) and predicts which users that ad is likely to resonate with. In plain terms: your creative is now doing a big share of the targeting work.
What this means for how you build campaigns
The biggest consequence is that creative diversity becomes a core input, not a nice-to-have. If you run five ads that are the same image with five slightly different headlines, the system sees essentially one idea. If you run five ads built around genuinely different angles (a price angle, a social proof angle, a problem-first angle, a founder-story angle, a comparison angle), you give the retrieval stage five different ways to find five different kinds of buyers.
The second consequence is that heavy manual restriction can work against you. Narrow audiences, long exclusion lists and over-segmented ad sets all limit what the system can learn from. Meta’s own guidance for 2026 leans toward broader audiences and fewer restrictions, and recommends against over-optimising because it limits AI learning. That matches what many advertisers report seeing in practice.
A practical creative framework: angles, not variations
Before you brief a designer or shoot a video, decide the angles. An angle is a distinct reason someone would buy. For a coaching service it might be “save time”, “avoid a costly mistake”, “proof from people like you” and “the method behind it”. Each angle then gets its own treatment: a different hook, a different visual, a different opening line.
Then vary format as well. A static image, a short vertical video, a carousel and a user-style testimonial clip each reach people differently. Mixing formats across a single campaign gives the system more distinct signals to work with than running the same format repeatedly.
- Pick 3 to 5 genuinely different angles per product or offer.
- Make at least two formats per angle (for example static plus short video).
- Keep the first three seconds and the first line of copy distinct across ads.
- Retire ads by learning (what angle fatigued), not just by cost per result.
Structure: fewer campaigns, more learning
A simple structure tends to serve this model best. Fewer campaigns and ad sets, each with enough budget to exit learning and collect real conversion data, usually beat a sprawl of tiny ad sets that each starve. Advantage+ sales campaigns are built around this idea: broad inputs, AI-driven delivery, and a creative library doing the heavy lifting.
This is also where conversion tracking quality matters more than ever. When the algorithm is steering, the signal it steers by is your pixel and Conversions API data. Clean, deduplicated events and a meaningful optimisation event (a purchase or a qualified lead, not a vanity click) give Andromeda something real to learn from.
What to measure instead of audience performance
When audiences stop being the variable you control, reporting needs to shift too. Look at performance by creative and by angle: which message is pulling cost per acquisition down, which is driving new customers rather than returning ones, and which fatigues fastest. Hook rate (how many people watch the first few seconds) and hold rate on video are useful early signals of whether an angle is landing.
Also watch for creative fatigue as a leading indicator. If frequency climbs and results soften, the answer is rarely a new audience. It is usually a new angle.
Common mistakes to avoid
- Launching one hero ad and calling the test “broad”. Broad targeting only works when the creative set is broad too.
- Duplicating the same ad into many ad sets and expecting more learning.
- Changing budgets and creatives constantly, which resets what the system has learned.
- Judging an angle after a day or two of data.
- Ignoring landing-page speed and clarity. Better targeting only exposes a weak page faster.
Example: how one offer becomes five angles
Take a local fitness studio selling a 30-day starter pack. The lazy approach is one ad: a smiling trainer with “Join today”. An angle-led approach produces five different ads. One speaks to beginners who feel intimidated by gyms. One focuses on busy professionals who only have 30 minutes. One leads with a member transformation story. One compares the studio with a typical gym contract. One simply shows a class in action with a limited-time offer.
To a person these are five different reasons to click. To a creative-reading system they are five distinct signals, each likely to be matched with a different type of viewer. The budget, the audience settings and the landing page can stay identical. Only the idea changes, and that is usually where the performance gap comes from.
How long to wait before judging results
A common mistake is changing things too early. Meta’s delivery needs time and conversion volume to learn, so a day or two of data tells you very little. As a rule of thumb, give a new creative at least several days and enough spend to generate a meaningful number of conversions before deciding whether it works. If you are optimising for purchases and each costs a few hundred rupees, that is a different time frame from a high-ticket lead where results arrive slowly.
When you do make changes, make one at a time and note the date. A simple log of what you changed and when turns a confusing account into something you can actually learn from.
Why landing pages and tracking matter more, not less
When the algorithm does more of the finding, the quality of what happens after the click decides whether that traffic turns into revenue. A slow, confusing or mismatched landing page wastes the very buyers Andromeda worked to find. Make sure the page repeats the promise made in the ad, loads quickly on a mid-range phone, and asks for one clear action.
On the measurement side, check that your pixel and Conversions API events fire once per action, carry the right values, and map to the outcome you actually care about. Better signals in mean better learning out.
What to do this week
Audit your active ads and group them by angle. If most of them share one idea, that is your biggest opportunity. Commission or shoot two or three new angles, relax any audience restrictions that are not strictly necessary (existing-customer exclusions are usually worth keeping), and give the new creatives enough runway to be judged fairly.
If you want a second pair of eyes on your account structure and creative mix, I run Meta Ads for startups and SMBs with exactly this approach: a small number of clean campaigns, a diverse creative pipeline and honest reporting. You can get in touch here to talk it through.
Meta rolls these features out gradually and details change, so treat the specifics above as a snapshot from early October 2026 and confirm against Meta’s official announcements and your own Ads Manager before making budget decisions.