Negative Keyword Workflow That Actually Improves ASA ROAS (Without Wrecking Delivery)

If your Apple Search Ads spend feels “leaky,” negative keywords are usually the fix—but only if you add them deliberately. Most teams either (a) negate too broadly and choke delivery, or (b) negate too late and keep paying for the same low-intent taps. Here’s a workflow you can run as a small studio to improve ROAS while keeping auction eligibility healthy.

Why negatives are tricky in Apple Search Ads

On Search Results placements, your bids and match types determine which queries trigger your ads. Even with exact and broad keywords, you’ll still see a range of “search terms” Apple matched to your keyword.

Negative keywords help by preventing ads from showing for specific queries. But if you add negatives based on noise (one day of data, one outlier query, or mismatched attribution), you can accidentally:

Block queries that are low-volume but high-converting
Reduce impressions enough that your CPT/auction dynamics change
Create a “data hole” that makes it hard to evaluate what actually improved

So the goal is to add negatives in a controlled, evidence-based way.

Step 1: Export the right search term report (and filter it hard)

Start with the Search Results keyword performance view that shows search terms and their outcomes (taps, installs, conversion rate, CPT/CPA, ROAS).

Export at least 14 days (or more if spend is low). Then immediately filter to the parts that can justify action:

Keep only rows with at least a small minimum activity (use a threshold that matches your traffic; for some indies it might be “≥ 5 taps”)
Separate Exact-match keyword performance vs Broad/Search Match behavior
- Negatives are often most valuable when broad or Discovery-style matching is pulling in the wrong intent
Focus on search terms that are clearly “intentive” in the wrong direction (more on how to spot those below)

Rule of thumb: Don’t build a negative list from zero-install data alone. First, look for patterns in query intent.

Step 2: Classify search terms into 3 groups

For each search term, assign one label:

A) “Strong exclude” (negate)

Negate queries that meet all (or most) of these:

Very low install conversion (e.g., near-zero install rate) and
No sign of downstream revenue (ROAS is dead, or CPI/CPA is wildly worse than your target) and
Query intent is clearly unrelated to your app’s value

Examples of intent mismatch patterns (illustrative):

Queries searching for a competitor name when you don’t want competitor traffic
“Free/Crack/Mod” style queries if you’re not capturing that audience reliably
Broad “wallpaper/image” searches when your app is a utility that doesn’t match that use case

B) “Hold” (don’t decide yet)

Keep queries that are:

Low volume (not enough taps to be confident)
Showing some installs, even if ROAS is noisy today
In a category where your product page or onboarding might be the bottleneck rather than matching

These are the queries you often regret negating too early.

C) “Maybe exclude” (watchlist)

Some queries are suspicious but not definitive. Put them on a watchlist and reevaluate after your next data window.

Good candidates for “maybe”:

Queries with a few installs but no revenue yet (install→purchase can settle later for subscriptions)
Queries where ROAS is low but conversion might be explainable by product page mismatch

Step 3: Add negatives in batches (and freeze variables)

This is where teams usually break their measurement.

Do this instead:

Freeze everything else for the test window (bids, budgets, campaign settings, custom product pages)
Apply negatives in one batch per iteration
Start with the strongest excludes (Group A) only
Give it time for attribution to resolve (Apple resolves ad attribution within ~24 hours, but your revenue events can take longer depending on your funnel)

A practical approach for many indies:

Day 0: add the first negative batch
Days 1–7: do not add more negatives (unless you see obvious runaway spend on new bad queries)
Day 7–10: reassess and plan the next batch

Step 4: Choose the right scope: Campaign vs ad group

Negative keywords can be applied at different levels depending on your setup. The key idea is to apply negatives where they prevent waste without breaking good match patterns.

Common strategy:

If a specific ad group is responsible for a keyword theme (e.g., your “productivity” cluster), apply negatives at the ad group level when possible so other ad groups that target other themes aren’t collateral damage.
If the wrong intent shows up across multiple themes, campaign-level negatives can be appropriate.

Avoid blanket negatives “because it’s bad for one keyword” unless you’re sure the query intent would never fit your app.

Step 5: Verify impact using the metrics that matter

After your negative batch, don’t only look at ROAS. Use a before/after sanity check:

What you want to see

Taps decrease for the excluded intent (good sign)
Install conversion rate improves or stays stable (you didn’t just reduce volume—you improved quality)
ROAS improves without a dramatic collapse in overall impression share
CPT and CPA don’t spike wildly

What to watch out for

ROAS improves, but installs collapse because you blocked borderline keywords
CPA improves but revenue doesn’t—could mean your attribution mapping / event timing needs attention (especially if subscriptions settle later)

Important note: Apple Search Ads doesn’t give you per-keyword revenue directly. Revenue is attributed through the install→purchase chain (often via RevenueCat mapping to App Store revenue). So evaluate performance at the campaign/ad group level where install sources are clearer.

Step 6: Don’t chase one-off failures—look for repeat offenders

Your search term stream will include odd queries that never repeat. Those are usually not worth negating.

A better test:

If the same or very similar bad intent appears in multiple days, it’s worth excluding
If it appears once with tiny volume, keep it in “hold”

This reduces the risk of overfitting your negatives to one week’s randomness.

Step 7: Maintain a “negative backlog” and a review cadence

Set a simple cadence:

Weekly: add the next batch of top offenders (based on Group A)
Monthly: review the watchlist (Group C) and decide whether to promote to Group A
Ongoing: remove negatives only if you find they’re blocking good queries (this requires caution; treat it like a controlled experiment)

Where negatives won’t fix the problem

If your main issue is that your ad is matching the right queries but users aren’t converting, negatives might not help.

Use these checkpoints:

Is your product page / custom product page aligned with the promise implied by your keywords?
Is your conversion rate (installs/taps) low compared to your historical norm?
Are you seeing taps but no installs (suggesting relevance/landing friction)

Negatives help query relevance. They don’t fix a mismatch between the app store page and what users expected.

One more practical tip: start with broad/Search Match waste

Exact-match often behaves. Broad and Search Match are more likely to pull in “almost-right” queries. If you’re going to add negatives, your highest ROI usually comes from stopping that waste at the query level.

Closing takeaway

A good negative keyword workflow isn’t “collect bad searches and nuke them.” It’s:

Export 14+ days of search terms
Classify into strong exclude / hold / watchlist
Apply negatives in controlled batches
Freeze other variables and verify with installs + conversion + ROAS

Do that consistently, and you’ll stop paying for low-intent taps without accidentally strangling delivery. If you want, AdsBuddy can turn your ASA + revenue data into a daily, prioritized negative plan you approve before changes—useful when you’re trying to move fast without breaking measurement.