Which Google Ads recommendations are worth applying, which to decline by default, how to disable auto-apply, and why optimization score belongs in no client report.
The call lasted eleven minutes. A friendly Google Ads strategist walked me through my account's "missed opportunities": enable AI Max, raise two budgets, move my exact match keywords to broad. The close was my optimization score, 74%, delivered in the tone of a doctor reading bad bloodwork.
I have taken versions of that call across every account I have managed, and the advice has one consistent property: each item either increases spend or hands Google more control over where the spend goes. Some of it is still good advice. The job is telling which part, and the Recommendations tab will not help you tell, because the tab itself is graded on acceptance.
The takeaways
Because your spend is Google's revenue, and the Recommendations engine is built by the seller. Its suggestions cluster around three moves: spend more (budget raises), give the algorithm more inventory (broad match, Display expansion, Search Partners), and reduce your ability to intervene (auto-apply, AI Max). No bad faith required. This is what any sales channel optimizes for when left to write its own advice column.
The human layer reinforces it. The strategists who call you work through quarterly adoption playbooks: two years ago the pitch was Performance Max, in 2026 it is AI Max, with its expanded query matching, auto-generated ad text, and final URL expansion pitched as one toggle. I treat those calls as a signal of what Google wants adopted this quarter. That signal has value. It just is not advice.
The repairs. When a recommendation describes something broken, acting on it costs nothing and unblocks delivery: fix disapproved ads and assets, resolve conflicting negative keywords that block your own queries, repair final URLs that lead to dead pages, restore conversion tracking that stopped firing. These have a property the rest of the tab lacks: a correct answer exists.
Even here I make the fix manually instead of clicking Apply. The diagnosis is often right while Google's one-click repair is wrong. "Conflicting negative keyword" is a real problem, but the right resolution is sometimes removing the negative, sometimes removing the keyword, and only you know which one encodes an actual decision. Treat the repairs section as a free monitoring service. Take the alert, do the surgery yourself.
Three recommendation types deserve a case-by-case look rather than a reflex.
New keyword suggestions are sometimes useful, often near-duplicates or category terms you excluded on purpose. Before adding any, check your own search-terms report; if the suggested keyword has real query volume there with decent conversion rates, add it with your match type and your structure.
Ad strength improvements for responsive search ads are a setup heuristic. Google's help pages frame Ad Strength as guidance on asset variety, and adding genuinely distinct headlines can help testing. Stuffing fifteen near-identical headlines to turn a label from "Good" to "Excellent" helps nobody.
Budget recommendations on capped winners are the one spend suggestion worth a look. If a campaign hits its daily cap with strong return and the Search Lost IS (budget) column confirms you are capped, raising it can be right. Raise it yourself, in steps of 10-20%, on your own schedule.
The growth bundle. "Raise your budgets" fires when a campaign hits its cap, with no profitability check attached; it triggers the same whether your ROAS is 4.2 or 0.4. When a cap is the real constraint, I want a human making that call. It is the same reason I built staged pacing alerts into Adscalr (overspend flagged at 110% of cap, runaway at 150%): budget moves stay a decision, never a default.
"Upgrade to broad match" is a migration sold as a tweak. You hand query selection to Smart Bidding, and whether that works depends on conversion volume and feed quality you can verify yourself. If you want to test it, do it deliberately in one campaign with its own budget. Never as a batch-applied recommendation.
"Remove redundant keywords" and similar consolidation erases structure you built on purpose, including the negative logic that lives in it. AI Max enrollment bundles broader matching with rewritten ad text and expanded landing URLs; if curious, pilot one campaign with URL exclusions set. And Display expansion plus Search Partners extend you onto inventory with no query report worth the name.
Open the Recommendations tab and click Auto-apply in the top corner. You will find the recommendation types grouped into two bundles, roughly "maintain" and "grow". Untick everything, including the maintenance items, because even the harmless-sounding ones (ad text changes among them) edit live creative without you. Then open the change history, filter for auto-applied changes, and read what the system already did to the account.
Do this audit on any account you inherit. Agencies find auto-apply enabled by a predecessor, by a rep "pilot" agreed to on a call months ago, or by someone who clicked through a blue banner. The change history tells you the real damage either way: keywords added, match types changed, ad text rewritten, all attributed and dated. I put a monthly calendar entry on re-checking it, because new recommendation types get added to the bundles over time.
No. Optimization score measures how much of Google's advice you have acted on. The mechanics give it away: dismissing a recommendation raises your score by the same amount as applying it, so a 100% score can mean a perfectly tuned account or an account whose manager clicked "dismiss" forty times. A score you can max out by declining everything measures engagement with the tab. Add that Google's Partners program requires a 70% score to hold the Partner badge, and you know whom the number serves.
Agencies put it in client reports at their peril. The score drops overnight when Google ships new recommendation types, and reporting it trains clients to grade you on Google's checklist. The first rival pitch deck that says "your current agency has you at 61%" will make you wish you had reported cost per acquisition and left the score out. Google's own help pages cite a 14% median conversion uplift for advertisers who raised their score by 10 points; that is Google measuring accounts that adopted Google's suggestions, published by Google. Weigh it accordingly.
Everything above concerns accepting the platform's judgment. Your own automated rules are the reverse arrangement: thresholds you wrote, derived from your margins, with your safeguards around them. I have covered where self-built rules go wrong in the automated rules mistakes that kill winners and the decision logic for ending an ad in when to kill an ad; both assume the human owns the thresholds.
That is also the design principle behind Adscalr's automation layer: the machine recommends by default, every evaluation is logged, full-auto is opt-in, and anything that moves money waits for approval. The Recommendations tab offers none of that posture, which is why auto-apply stays off in every account I touch. If you want to see what recommend-first automation looks like in practice, the automation page walks through it.
This is the thinking behind Adscalr.
See the product →