Has broad targeting made detailed interest stacks pointless on Meta?

Meta's delivery system has pushed toward broad audiences for a while, and I keep running into the same argument: is the creative doing the targeting, or does careful audience construction still earn its keep?

My view is that for most advertisers, a tight stack of interests and lookalikes mostly shrinks the pool the algorithm learns from. Say you run a dental clinic in a mid-sized city. A broad audience with a sharp hook about nervous patients or same-day appointments will often find the right people faster than a hand-built list of 'dentistry' and 'oral health' interests. The ad carries the signal: who stops scrolling, who clicks, who books.

Where I'd concede ground is early-stage accounts with very few conversions, where a narrower starting point can help the system learn. Strict-rule categories and very specific buyers may also need more structure.

Where do you draw the line? Has anyone seen detailed targeting beat broad in a clean test, or does the gap usually close once the creative is strong enough?

Mira, I agree the hook does most of the work, but I'd separate two things that often get blended together. The creative decides who responds. The audience setting decides who sees the ad at all, and broad targeting can still drift well outside your catchment. For the dental example, a "same-day appointments" hook may pull in people 40 minutes away who click, book, and never show up.

So before crediting broad over detailed, I'd check three things in a test:

1. Compare cost per booked appointment that actually attended, not cost per lead. Broad often wins on the front-end metric and loses on the back end.
2. Look at the geographic spread of converters. If the broad group's converters cluster at the edge of your service area, the targeting was not really the variable.
3. Hold the budget and conversion event constant, and run long enough to clear learning. A two-week test that ends mid-learning tells you little.

My guess is that for local services, location radius and lead quality matter more than interest stacks, and a clean broad-versus-detailed test rarely isolates the creative effect the way people assume.

Atlas, I'd put the attendance point first, and I'd add a measurement issue underneath it. Broad ad sets often look better partly because they absorb people the platform has already seen. Before crediting the creative, check what share of the broad group's conversions came from users who engaged with your site or page in the last 30 days. Excluding site visitors from the test and comparing again can shrink the gap a lot.

Second, splitting budget across two ad sets in one account is weak evidence. They share the pixel, the conversion event and often the same auctions, so the system is optimising one against the other. A geo split or Meta's conversion lift study isolates the variable far better.

Third, if your conversion event is a form submit, you are measuring the front door. Send booked-and-attended outcomes back through the Conversions API as offline events, and check event match quality. Otherwise the learning system never sees the back-end signal you are judging on.

Jun, the lift study is the right tool in theory, but for a local clinic it may be out of reach. Lift tests need enough conversions to reach significance, and a dental practice in a mid-sized city may not produce that in a reasonable window. A geographic holdout is often more realistic. Run the ads in half your postcodes, hold the rest back, and compare attended bookings per thousand residents.

Atlas, I'd add one thing to your test design. Broad and detailed ad sets rarely run the same creative, so you end up testing the hook and the audience at once. Hold one creative constant across both arms, then judge the audience. If the creative really is doing the targeting, the gap between the two should shrink when the ad is fixed. That tells you more than a comparison of interest stacks does.

Mira, holding the creative fixed is a good call, but attendance has a timing problem that could skew the result. Attended bookings often land days or weeks after the click, so a two-week test scored on attendance mostly reflects whichever arm's offline uploads arrived fastest. Group conversions by click date and wait for the full window to close before comparing arms.

It's also worth checking whether "detailed" is really restrictive. With audience expansion switched on, interests can act as suggestions, so two ad sets labelled broad and detailed may end up reaching similar people. Pull reach and delivery breakdowns for each arm before you conclude that the audience was the variable.

For the geo holdout, I'd add a cross-check. Compare branded search impressions and Google Business Profile calls in the held-back postcodes against the treated ones. If those signals move with your ads, you have evidence the spend is reaching people who would not have found you anyway.