AI in Google Ads 2026: What Actually Works and What Doesn't

Google has integrated AI into nearly every aspect of Google Ads. Some features genuinely improve performance. Others are designed to increase Google's revenue, not yours.

The AI Landscape

AI features fall into three categories:

1. Helpful AI: Features that genuinely improve performance
2. Neutral AI: Features that may help depending on your situation
3. Harmful AI: Features that typically hurt performance or hide data

Why This Matters

Blindly accepting all AI recommendations:

• Often increases spend without proportional returns
• Removes control and visibility
• Optimizes for Google's interests, not yours

Selectively using AI:

• Leverages what works
• Maintains control where needed
• Produces better overall results

The Evaluation Framework

For each AI feature, ask:

1. Does this improve my metrics or Google's?
2. Can I measure the impact clearly?
3. Does this reduce my control over important levers?
4. What's the downside if it fails?

These AI features consistently improve performance when used correctly.

Smart Bidding (When You Have Data)

What it does: Optimizes bids in real-time based on conversion likelihood.

When it works:

• 30+ conversions per month per campaign
• Stable conversion tracking
• Consistent campaign structure

Best options:

• Target ROAS for ecommerce
• Target CPA for lead gen
• Maximize Conversion Value for high-value focus

Results: 15-30% efficiency improvement vs. manual bidding when conditions are met.

Responsive Search Ads (RSAs)

What it does: Tests combinations of headlines and descriptions automatically.

When it works:

• Provide 10+ quality headlines
• Provide 3-4 quality descriptions
• Don't over-pin (limits testing)

Best practices:

• Write diverse headlines (different angles, not variations)
• Include keywords, benefits, CTAs, differentiators
• Pin only essential positions (1-2 max)

Results: 10-15% CTR improvement vs. limited ad variations.

Dynamic Remarketing

What it does: Shows personalized ads featuring products users viewed.

When it works:

• Product feed is complete and accurate
• Sufficient remarketing audience (1,000+)
• Products have good images

Results: 2-3x higher CTR and 50%+ better conversion rates than generic remarketing.

Audience Signals in Performance Max

What it does: Guides PMax's targeting toward your ideal customers.

When it works:

• You provide quality first-party data
• Audience signals are specific and relevant
• You have conversion data for learning

Results: Faster learning period, better initial targeting, improved overall performance.

Recommendations (Selective)

Some recommendations are helpful:

• "Add negative keywords" (usually valid)
• "Fix disapproved ads" (always do)
• "Improve ad strength" (sometimes valid)

Review each recommendation individually—don't bulk apply.

These features can help or hurt depending on your situation. Use with caution.

Broad Match + Smart Bidding

The theory: Broad match finds new opportunities; Smart Bidding controls costs.

When it works:

• High conversion volume (50+/month)
• Strong negative keyword lists
• Budget for discovery
• Smart Bidding is well-calibrated

When it fails:

• Low conversion volume
• Weak negatives
• Limited budget
• New or unstable campaigns

Verdict: Test cautiously with 20% of budget before expanding.

Performance Max

The theory: All-in-one campaign reaching all Google inventory.

When it works:

• Strong creative assets
• Good product feed
• Brand exclusions enabled
• Run alongside Standard Shopping for comparison

When it fails:

• Weak assets
• No brand exclusions (cannibalizes brand)
• Only PMax running (no benchmark)
• Limited conversion data

Verdict: Use as part of campaign mix, not as replacement for everything.

Optimized Targeting

The theory: AI expands beyond your targeting to find converters.

When it works:

• Display/Discovery campaigns seeking reach
• Strong conversion tracking
• You're okay with less control

When it fails:

• You need precise audience control
• Brand safety is important
• You have limited budget

Verdict: Test with caution; monitor placements closely.

Auto-Apply Recommendations

The theory: Google automatically implements recommendations.

When it works:

• For simple, low-risk recommendations (fixing typos)
• When you have no time for management

When it fails:

• Most cases—auto-applying budget increases, bid changes, or targeting expansions

Verdict: Keep most auto-applies off. Review manually.

AI-Generated Assets

The theory: AI creates ad headlines, descriptions, images.

When it works:

• You need quick starting points
• You'll edit and improve the outputs
• Testing volume is high

When it fails:

• You use outputs without review
• Brand voice is important
• Quality matters more than quantity

Verdict: Use as starting point, not final product.

These features often reduce performance or hide important data. Approach with skepticism.

Budget Recommendations (Almost Always)

What Google suggests: "Increase budget by X to get Y more conversions."

The problem:

• Assumes linear scaling (rarely accurate)
• Ignores diminishing returns
• Pushes you toward Google's revenue goals

Reality: Budget increases typically see efficiency drops. A 50% budget increase rarely delivers 50% more conversions.

What to do instead: Scale gradually (20-30% max) and monitor efficiency.

Search Themes in Performance Max

What it does: Lets you add keyword themes to PMax.

The problem:

• Gives false sense of control
• PMax may or may not use them
• Can't see query-level data anyway

What to do instead: Accept PMax's black-box nature or run Search campaigns for control.

Data-Driven Attribution (Sometimes)

What it does: Credits conversions across touchpoints using AI.

The problem:

• Can over-credit awareness channels
• May under-credit bottom-funnel
• Black box—can't verify accuracy

When to use: Large accounts with complex funnels.

When to avoid: Simple funnels, low conversion volume.

Automatically Created Assets

What it does: Creates headlines and descriptions without your input.

The problem:

• Often generic or off-brand
• May include inaccurate claims
• Reduces your control over messaging

What to do: Disable in campaign settings unless you've reviewed and approved.

Network Expansion (Display/Search Partners)

What Google says: "Expand reach across partner sites."

The problem:

• Quality varies dramatically
• Often includes low-quality placements
• Conversion rates typically much lower

What to do: Disable Search Partners and Display expansion by default. Test separately if interested.

Location Expansion ("People interested in")

What it does: Shows ads to people "interested in" your target locations.

The problem:

• "Interest" is loosely defined
• Includes people who will never convert (tourists, researchers)
• Wastes budget on irrelevant users

What to do: Set to "Presence" only unless you're a travel/tourism business.

Here's how to configure AI features for optimal performance.

Campaign-Level Settings

Bid Strategy

• ✅ Use Smart Bidding when 30+ conversions/month
• ✅ Start with Maximize Conversions, graduate to Target ROAS
• ❌ Don't use Smart Bidding with <15 conversions/month

Networks

• ✅ Keep Search (core)
• ❌ Disable Search Partners (low quality)
• ❌ Disable Display Network expansion (off-target)

Locations

• ✅ Set to "Presence: People in your targeted locations"
• ❌ Don't use "Presence or interest"

Ad-Level Settings

RSA Optimization

• ✅ Provide 10-15 diverse headlines
• ✅ Provide 4 descriptions
• ✅ Pin 1-2 positions strategically
• ❌ Don't over-pin (limits learning)

Auto-Created Assets

• ❌ Disable "Automatically created assets" unless reviewed
• ✅ Review and approve any AI-generated content

Audience Settings

Optimized Targeting

• ✅ Use in Display for prospecting
• ❌ Disable when audience precision matters
• ⚠️ Monitor placements closely

Audience Expansion

• ❌ Disable in most cases
• ✅ Only enable with large budgets for testing

Account-Level Settings

Auto-Apply Recommendations

• ✅ Enable: "Remove redundant keywords," "Fix broken URLs"
• ❌ Disable: All budget, bid, and targeting auto-applies
• ❌ Disable: "Add new keywords," "Add new audiences"

Google Reps Suggestions

• ⚠️ Treat with skepticism
• ✅ Evaluate each suggestion on your metrics, not Google's

The Override Philosophy

Google's defaults favor:

• Broader targeting (more impressions, more revenue for Google)
• Higher budgets (more spend for Google)
• Less visibility (harder to identify waste)

Override to favor:

• Precise targeting (better ROAS for you)
• Controlled budgets (profitability)
• Full visibility (optimization opportunities)

When testing AI features, follow this framework to minimize risk.

Testing Principles

1. Isolate the variable: Test one AI feature at a time
2. Limit exposure: Use 20% of budget for tests
3. Define success: Set clear metrics before starting
4. Set duration: Run for 2-4 weeks minimum
5. Have a control: Keep non-AI version running

Testing Broad Match + Smart Bidding

Setup:

• Create duplicate ad group
• Convert keywords to broad match
• Apply same Smart Bidding strategy
• Set 20% budget allocation

Measure:

• ROAS vs. phrase/exact
• Search query quality
• Conversion rate
• CPA

Decision criteria:

• If ROAS within 10% and volume up 20%+: Expand
• If ROAS down 20%+: Revert

Testing Performance Max

Setup:

• Run PMax alongside Standard Shopping
• Use same products, same budget split
• Enable brand exclusions in PMax
• Create strong asset groups

Measure:

• ROAS comparison
• New customer rate
• Incremental volume
• Query transparency (via scripts if possible)

Decision criteria:

• If PMax beats Shopping by 15%+: Shift more budget
• If PMax underperforms: Reduce or pause

Testing New AI Features

When Google launches new AI features:

1. Wait 2-3 months for bugs to be fixed
2. Read case studies from similar businesses
3. Test with small budget first
4. Measure rigorously
5. Scale only if proven

Documentation

Record every AI test:

• Feature tested
• Hypothesis
• Setup details
• Results
• Decision
• Learnings

Build institutional knowledge about what works for your specific business.

Use this checklist to audit and optimize your AI settings.

Immediate Actions

Campaign Settings Audit

• Check location targeting (switch to "Presence" only)
• Disable Search Partners
• Disable Display Network on Search campaigns
• Review bid strategies vs. conversion volume

Ad Settings Audit

• Disable automatically created assets
• Review RSA headline diversity
• Check pinning strategy

Account Settings Audit

• Review auto-apply recommendations
• Disable budget and bid auto-applies
• Keep only safe auto-applies enabled

Performance Max Audit

• Ensure brand exclusions are enabled
• Review asset group quality
• Check audience signals
• Compare to Standard Shopping performance

Ongoing Monitoring

Weekly

• Review recommendations (don't bulk apply)
• Check for new auto-applies turned on
• Monitor broad match query quality

Monthly

• Audit AI feature performance
• Compare AI vs. non-AI campaign performance
• Review PMax vs. Shopping metrics

Quarterly

• Reassess AI feature adoption
• Test new features with controlled budget
• Update settings based on learnings

When Google Releases New Features

• Wait 2-3 months for stabilization
• Research case studies
• Test with 20% budget
• Measure rigorously
• Scale only if proven