Performance Max runs across almost Google’s entire inventory — Search, YouTube, Display, Discover, Gmail, and Maps — from a single campaign. When it works, it’s the most efficient prospecting engine in the account. When it doesn’t, it feels like a black box with your credit card attached.

Here’s the good news: PMax is much less of a black box than it used to be. The 2025 update cycle gave advertisers real controls: campaign-level negative keywords (limit raised to 10,000), search themes (now 50 per asset group), full search term reporting, and channel-level performance reports that finally show where your budget actually went.

Most underperforming PMax campaigns aren’t victims of bad AI. They’re victims of sloppy inputs. These are the 9 fixes I apply in every PMax audit, in the order I’d apply them.

1. Fix your conversion goals before touching anything

Performance Max optimizes aggressively toward the primary conversion goals you select. If “page views” or a soft micro-conversion sits next to “purchase” as a primary goal, the AI will happily chase the cheap, easy conversions and call it a win. Garbage in, garbage out — at machine speed.

Audit your goals first:

• E-commerce: your primary conversion should be the purchase event — accurately tracked, deduplicated between browser and server events, with correct revenue values. If your tracking is off, everything downstream is off. Run through our server-side tracking setup guide before scaling PMax spend.
• Lead gen: don’t let PMax optimize on raw form fills if lead quality varies. Set up Enhanced Conversions for leads and import offline conversions from your CRM, so Google learns which queries, geos, and audiences produce leads that actually close — not just leads that fill out forms.

Secondary conversions are for observation. Primary conversions are the steering wheel. Get this wrong and nothing else in this list matters.

2. Stop brand traffic from inflating your ROAS

This is the number one PMax illusion I see in audits. PMax loves brand terms: cheap clicks, sky-high conversion rates, and a ROAS number that makes the campaign look like a hero. That brand ROAS gets blended into the campaign average, quietly hiding the fact that prospecting — the thing you’re actually paying for — is failing.

The fix is campaign-level brand exclusions (Settings → Content → Brand exclusions). Google’s own guidance calls these the most precise tool for brand-term control — more complete than negative keywords — because they block your brand and its common variations across both Search and Shopping in one move. Retail advertisers get a granular toggle most people miss: you can exclude brand terms from Search text ads while letting Shopping keep serving on them.

Keep a separate branded Search campaign running to capture that demand with full query visibility and bid control. Roles stay clean: brand Search captures existing demand, PMax hunts new demand.

3. Steer the AI with search themes

Search themes are the closest thing PMax has to keywords — but they’re not keywords. Google’s own wording: themes “provide a broader signal,” ads still serve based on several factors, and the algorithm serving beyond your themes is by design. Treat them as steering signals, not targeting.

Practical rules:

• 3–7 themes per asset group is plenty. Use broad, intent-level terms (“running shoes”), not hyper-specific ones (“red running shoes size 10”) — the AI handles the long tail itself.
• The limit was doubled to 50 per asset group in 2025, but more themes is not better. Keep them focused.
• Don’t duplicate what your product feed and website already teach the algorithm. Add event-specific themes (seasonal, promotional) to speed up ramp when timing matters.
• Check the Source and usefulness columns in the search themes report — they show whether a theme actually drove incremental traffic or just sat there. A widely cited 2025 study found 71% of advertisers saw mixed or flat results from search themes, largely because they couldn’t see which themes worked. Now you can.

Pull theme ideas from your Search terms report, not from brainstorming.

4. Use negative keywords like a scalpel, not a machete

Campaign-level negative keywords finally arrived in PMax in January 2025, and the limit was raised from 100 to 10,000 during the year. But Google’s warning is worth reading twice: negatives in PMax are “a highly restrictive control that can harm performance by preventing the AI from finding valuable traffic.”

The practitioner’s rule:

• Use negatives for brand-safety essentials and clearly irrelevant terms only — competitor conflicts, job seekers, free-seekers, products you don’t sell.
• Prefer brand exclusions over negatives for your own brand. They’re more comprehensive.
• Go easy on negatives in the first weeks after launch. Heavy negative lists during the learning window have been measured as a performance drag.
• Account-level negative keyword lists (up to 1,000 keywords) also apply to PMax — audit them too, since an old shared list can silently throttle a new campaign.

Negatives are guardrails, not strategy.

5. Structure asset groups around real business themes

One asset group should connect one product or service theme, one audience signal, one landing page, one message, and one set of images and videos. A generic “all products / all services” asset group mixes signals and weakens both learning and creative relevance.

When to split PMax campaigns: when the business genuinely needs it — separate budgets, different tCPA/tROAS targets, meaningfully different margins, different geos or languages, new-customer acquisition vs. existing customers. When not to split: to get prettier reports. Fragmenting campaigns to see cleaner channel splits starves each campaign of conversion volume and makes everything worse.

Consolidation is a feature, not laziness. Every split should have a business reason attached to it.

6. Feed it real creative — especially video

If you don’t upload a video, Google auto-generates one from your other assets. Auto-generated video is better than nothing, but it’s a placeholder — a real 15-second video (horizontal, vertical, and square if you can) gives the algorithm something genuinely competitive on YouTube and Discover, which are often where PMax’s incremental reach comes from.

The full asset mix per asset group:

• Images: landscape (1200×628 minimum), square (1200×1200), and portrait (960×1200)
• Up to 5 short headlines (30 chars) + 5 long headlines (90 chars)
• Up to 5 descriptions (90 chars)
• Business name and logo
• At least one real video asset

Then review asset performance monthly. Cut assets that sit at “Low” ratings for 30+ days, and build new variations around your winners. Creative testing inside PMax is slower than in Meta, but it’s still the highest-leverage activity most accounts aren’t doing.

7. Tame Final URL Expansion

Final URL Expansion is on by default. That means Google can replace your asset group’s final URL with a page it judges more relevant and generate matching assets on the fly. On a clean e-commerce catalog, that’s free extra reach. On a site with a blog, support center, careers pages, or policy pages, it’s ads sending paid traffic to pages that can’t convert.

Three tiers of control, in order of preference:

1. URL exclusions — block individual non-commercial pages.

2. Page feeds — curated include/exclude lists at the asset-group level via custom labels.

3. URL-contains rules — exclude whole sections like /blog or /careers. Google warns these can significantly limit the AI’s ability to find relevant traffic, so use this tier last.

After 4–6 weeks, check the landing page report: identify the highest and lowest-converting pages, exclude the losers, and consider building dedicated asset groups around your winners.

8. Read the channel report before you touch anything

Since 2025, PMax’s channel performance report breaks spend and results down by channel — Search, YouTube, Display, Discover, Gmail, Maps. This changed the entire diagnostic workflow. Before you restructure anything, diagnose:

• A PMax campaign that’s “failing at prospecting” is often a campaign spending 90% of budget on brand Search. That’s a brand-exclusion fix, not a rebuild.
• A campaign drowning in Display clicks with no conversions might need asset or audience-signal work, not a budget cut.
• Full search term reporting (same granularity as Search campaigns, available since March 2025) tells you exactly which queries you’re matching. Feed the winners into search themes and the junk into negatives.

You still can’t directly control budget allocation across channels — the report tells you where budget went, not where to force it. But knowing is the difference between an informed decision and a guess.

9. Stop micromanaging: consolidate and give it time

Smart Bidding needs volume to stabilize — Google’s own guidance points to roughly 30 conversions in 30 days for Target CPA and about 50 for Target ROAS as the zone where performance stops bouncing around. Below that, the algorithm spends most of its time relearning. Every major change (goal edits, big budget shifts, target changes) restarts that process.

The boring fundamentals that make PMax work:

• One consolidated campaign per business objective, not five starved fragments.
• Realistic budgets: PMax rejects shared budgets, and as a practitioner rule of thumb you want at least $50–100/day to give the algorithm room to learn. Under that, standard Search usually wins.
• No major changes during the learning window — add assets and refine signals if you must, but don’t restructure.
• Accurate, deduplicated conversion tracking feeding real business value (see fix #1, and our GA4 conversion tracking audit checklist if your measurement is suspect).

If your PMax strategy is “launch and panic-tweak weekly,” the algorithm never gets the stable data it needs. Patience is a tactic.

The bottom line

Performance Max rewards better inputs, not more levers. Define what a valuable conversion is, keep brand traffic out of your prospecting numbers, give the AI focused signals and real creative, control where it can send people — and then let it run. If your PMax campaigns feel like a black box with a credit card attached, the problem is almost always on this list.

Frequently Asked Questions

How many asset groups should a Performance Max campaign have?

As many as you have genuinely distinct product or service themes — each asset group needs a coherent theme, audience signal, landing page, and message. There’s no magic number. Avoid the single catch-all asset group; it mixes signals and weakens learning. Most e-commerce accounts land on 3–8 asset groups per campaign.

Can I exclude keywords in Performance Max?

Yes. Campaign-level negative keywords have been supported since January 2025 (limit raised to 10,000), and account-level negative keyword lists also apply to PMax. Use them sparingly — Google warns that heavy negatives can prevent the AI from finding valuable traffic. For your own brand terms, brand exclusions are the more precise tool.

Why is my PMax ROAS high but sales aren’t growing?

Almost always brand cannibalization. PMax captures cheap branded clicks, and that inflated ROAS blends into the campaign average while true prospecting quietly underperforms. Apply campaign-level brand exclusions, keep a separate branded Search campaign, and judge PMax on non-brand growth and incrementality — not blended ROAS. For the broader troubleshooting framework, see our Google Ads not converting fixes.

How long does Performance Max take to learn?

Allow about 1–2 weeks of stable learning after launch or any major change, and aim for at least 30 conversions in 30 days (Target CPA) or 50 (Target ROAS) before judging stability. Making major changes faster than the data supports just restarts the clock.

Should I run Performance Max alongside standard Search and Shopping campaigns?

Yes, with clear roles. A common setup: a dedicated branded Search campaign (full visibility, full control), PMax with brand exclusions handling non-brand prospecting and Shopping. Note that PMax takes priority over standard Shopping for the same products and queries, so don’t run both on the same inventory without a reason — you’ll just bid against yourself.

Israr Ads is a performance marketing agency managing Google & Meta ad spend for e-commerce and lead-gen brands. If your Performance Max campaigns aren’t pulling their weight, get a free audit and we’ll tell you exactly what’s wrong — no fluff.