Auto-apply Google Ads recommendations — when Google's AI works against you
What Google promises vs what it actually does
You open the Google Ads panel. Tools & settings → Recommendations. Optimization Score 84%. Below: "Accept all auto-apply, gain +14% conversions in 30 days."
It sounds reasonable. Google has more data than you, algorithms learn 24/7, why not benefit. Especially since it is free, and Optimization Score starts shining green.
The problem: Google's recommendations are driven by aggregated averages, internal adoption goals, and products Google is actively promoting — not by the specifics of your business, your acquisition model, or your customer-cost tolerance. As Search Engine Land put it in 2026: auditors often see business owners ask why their campaigns "are not working" — and the answer is usually that they applied Google's bad recommendations that killed the campaign.
91% of accounts with PMax/Search overlap
Overlap — in short — means your own campaigns are bidding for the same queries against each other. You pay twice for the same click and one ad format "eats" the other.
The strongest empirical evidence comes from an Optmyzr analysis of over 500 accounts:
- 91% of 503 accounts show keyword overlap between Search campaigns and Performance Max
- 56% of 5,768 Search campaigns have documented overlap with PMax
What this means in practice: on a clear majority of ad accounts, Performance Max was competing with your own Search campaigns. PMax is aggressive in the auction by design — Google prefers it because it is a newer format. Your Search campaigns with brand and protective keywords were either invisible or paying 2–3× more for the same queries.
Auto-apply removes "inactive or low-performing" keywords — often those very same ones that worked as a buffer between Search and PMax (let us call them protective keywords: low-volume brand-related phrases that block PMax from walking onto your brand queries). After removal, PMax walks fully onto your brand — in my experience brand CPCs jump 2–3× after this happens, sometimes more.
7 auto-apply types to TURN OFF
1. Remove redundant keywords
Google decides that a duplicate between group A and B is "redundant" and removes the weaker one. The problem: in your strategy that duplicate existed on purpose (e.g. testing match types, allocating budget to a specific group, protecting against PMax). After the removal, one group remains — and it is the worse one that starts steering traffic.
2. Remove non-serving keywords
The most dangerous on the list. "Non-serving" is often your low-volume brand keywords with a critical function: they protect you from PMax, which is happy to walk onto your brand-name queries. After they are removed — PMax walks in, and in my experience brand cost jumps 2–3× (sometimes more, depending on competition).
3. Add new keywords
Auto-apply adds keywords in broad match (or close variants), with no intent verification on your side. In practice: a campaign on "accounting Warsaw" gets "accounting software Warsaw" added — completely different intent, but Google sees semantic similarity. The campaign budget flows on unattainable keywords for a week before you notice.
4. "Accept Performance Max upgrade" / "Use Performance Max" — watch out for "Accept all"
Small technical note: this recommendation works a bit differently than the others — auto-apply does not create PMax campaigns automatically, you have to click "Apply" or "Accept all." The trick is that Google designs the interface so that "Accept all" is the first, most obvious click when you enter the panel. Converting Search or Shopping into PMax is a one-to-one swap — you lose transparency (where the money goes, where the conversions come from) and control over placements. The mechanism is not fully "auto," but the UX leads you there with one move. Practical takeaway: assess every "upgrade to PMax" recommendation manually, NEVER click "Accept all" in bulk.
5. Upgrade to AI Max (from DSA or Search)
Google is actively promoting the DSA → AI Max migration — the suggestion appears as a recommendation on DSA-enabled accounts, and the volume of these recommendations keeps growing. AI Max decides keywords on its own, matches destination URLs, and generates copy. If your conversion data is weak, AI Max guesses. Expensively. From my observations during audits and industry discussions (PPC Hub, Google Ads forums), most advertisers I have spoken with report neutral or worse results after switching to AI Max on a weak conversion data base. A concrete Google timeline for the DSA sunset has not been officially confirmed as of May 2026 — but the pressure to migrate is real and growing month by month.
6. Expand audiences / targeting
Google widens targeting to "similar groups" or "affinity audiences." Sounds good — in practice: a campaign with carefully selected affinity for a small local business gets expanded to general segments, CTR drops 30–50%, cost rises.
7. Adjust budgets
Auto-apply can raise campaign budgets if "performance shows potential." The algorithm decides based on 7-day data — you often see a 30–50% spend bump in the morning panel before you even checked the previous night.
2 safe types + 1 to assess manually
Not all auto-apply is harmful. Two categories work in your favor in most scenarios, one requires careful judgment:
A. Add negative keywords from search terms (SAFE)
Google identifies search terms that generate spend without conversions and adds them as negatives. This is safe optimization — you remove wasted traffic without losing anything valuable. Condition: your conversion tracking is set up correctly (otherwise Google sees "no conversions" on phrases that actually work but you do not measure).
B. Fix disapproved ads (SAFE)
Google automatically fixes rejected ads (e.g. capitalization, health policies). This is safe because the alternative is: the ad does not show at all. Auto-apply restores it in a policy-compliant version.
C. Update keyword match types (ASSESS MANUALLY)
Google suggests switching from exact to phrase or broad. Here conditionally — safe only if you have solid conversion data per match type (roughly 30+ conversions per month per match type). In practice: do not enable auto-apply, review the recommendations manually once a month. Auto mode for these changes is too aggressive at low conversion volume.
How to turn them off — step by step
Panel procedure, about 5 minutes (the exact location may vary by UI version, but the names of sections are stable):
- In the left menu open Recommendations (in older interface variants it may be under Tools & settings)
- Select the Auto-apply tab at the top of the Recommendations screen
- Maintain your ads section: uncheck everything. I only leave "Fix disapproved ads" on.
- Grow your business section: uncheck everything. This is where Google usually pushes PMax / AI Max / Expand audiences — those are the 5 most dangerous from the list above.
- Save. After 24 hours check the panel — you should see recommendations (Google still suggests them) but without automatic application. You make the decision manually.
Exact section names may vary by account type or UI version; the auto-apply checkboxes are typically grouped under the same two categories.
The payoff: over 30–60 days you typically see CPC stabilize on brand campaigns and the end of unexpected budget jumps.
What is next
Turning off auto-apply is the first step. The next step is understanding how AI Max enters your account from another direction (via DSA migration) — I will describe that in a separate article. For full context — the manifesto on AI safety in advertising 2026 ties the threats together.
About the sources: Primary data analysis — Optmyzr: "Is Performance Max Cannibalizing Your Search Campaigns?" — study of 500+ accounts, 5,768 Search campaigns. Plus Search Engine Land — "Why Google's Performance Max advice often fails new advertisers"; Savvy Revenue — "6 Performance Max Claims That Have No Basis"; Marlin SEM — "7 Reasons Not to Use Performance Max"; Google Ads Help — Manage auto-apply recommendations. Plus observations from accounts run by JP Ads — the list of 7 to turn off is based on specific client cases (anonymized).