Performance Max: How Google PMax Works and How to Run It With AI

Performance Max (PMax) is Google Ads' goal-based campaign type that runs one campaign across every Google channel at once: Search, Shopping, YouTube, Display, Discover, Gmail, and Maps. You give it a conversion goal, a budget, creative assets, and (for stores) a Merchant Center feed; Google's AI decides where your ads show, what to bid, and which assets to serve. Since 2021 it has gone from a "black box" to a campaign you can actually steer: negative keywords, search terms reports, channel reporting, brand and audience exclusions all arrived between 2025 and 2026. This guide covers how Google Performance Max works, when it beats Standard Shopping, how to set it up step by step, and how to use Claude or ChatGPT to build and optimize PMax campaigns faster.

Key takeaways

• Performance Max is one campaign, every Google channel, driven by your conversion data. Send revenue as the value and it chases revenue; send profit and it chases profit.
• You now have real controls: 10,000 negative keywords per campaign, a full search terms report, channel performance reporting, brand exclusions, device and demographic exclusions, "Your data" audience exclusions, and asset experiments.
• Structure follows margin and creative. One campaign per bidding target, a few asset groups per genuine product theme, listing groups that separate heroes from low-margin items.
• Give it a floor of data. About 30 conversions a month per campaign is the practical minimum for stable Smart Bidding; below that, Standard Shopping or Search is your workhorse.
• AI shortens the grind. Writing 25 text assets per asset group, mining search terms for negatives, and reading channel reports are jobs Claude and ChatGPT do well with the prompts below.

What Performance Max actually does

A PMax campaign is three systems working together:

• Smart Bidding sets a bid for every auction based on your goal (Maximize conversions, optionally with a target CPA, or Maximize conversion value, optionally with a target ROAS).
• Automated placement decides whether a user sees a Shopping ad, a text ad, a YouTube pre-roll, a Discover card, or a Gmail promotion. There is no setting to pick channels.
• Asset assembly mixes your headlines, descriptions, images, videos, and logos into ads per placement. No video uploaded? Google generates one from your images.

For stores, PMax is the successor to Smart Shopping: attach a Merchant Center feed and it uses product data for Shopping ads and your assets for everything else. In most retail accounts the bulk of spend lands on Shopping and Search; the channel performance report (below) shows your split.

Where PMax campaigns fit next to other campaign types

• Search: you pick keywords and write ads. Use it for brand terms and query-level control.
• Standard Shopping: Shopping ads only, with product groups and manual or automated bidding. More work, more control, works with less volume.
• Demand Gen: YouTube, Discover, and Gmail with full creative control.
• Performance Max: all of the above, automated.

If a query exactly matches an eligible keyword in a Search campaign, that Search campaign is prioritized; otherwise the higher Ad Rank wins, which is why brand exclusions in PMax matter.

When to use Performance Max (and when not to)

Google PMax campaigns are the right call when:

• You run a Shopify or other ecommerce store with a healthy Merchant Center feed and want Shopping plus everything else from one budget.
• You have at least ~30 conversions per month per campaign and reliable conversion values.
• You're capped on Search and want incremental reach without building separate YouTube and Display campaigns.
• Margins are consistent enough within each campaign that a single ROAS target makes sense.

Stick with Standard Shopping or Search when:

• Volume is low (new store, niche catalog, under 30 conversions a month).
• You need product-level bid control for a mixed-margin catalog.
• You sell services and lead quality matters more than lead count. PMax will fill your form with junk unless you send offline conversion values (qualified lead, closed deal) back to Google.
• You have no real creative yet. PMax will still spend, on autogenerated creative you never approved.

A pragmatic 2026 hybrid: Standard Shopping with a manual CPC cap for low-margin or clearance products, PMax for hero products with real margin, and a Search campaign for brand terms, excluded from PMax.

How to set up a Google Performance Max campaign step by step

Here's the manual build, so you understand each setting before AI helps you fill it in.

Before you build: fix tracking and the feed

• Conversion tracking: your Purchase conversion must be primary, with a dynamic value. Shopify's Google & YouTube app sets this up; add Enhanced Conversions or server-side tagging so iOS and ad-blocker traffic doesn't disappear.
• Merchant Center: fix every disapproval, fill in GTIN, brand, color, size, and product_type, and add custom labels (for example margin_band: high/mid/low). You'll use them to split listing groups.
• Customer Match: upload your buyer list (last 180 days) to use as a signal and, later, as an exclusion.

Step 1: Objective, campaign type, and feed

Go to Campaigns → + New campaign → Sales, choose Performance Max, and select your Merchant Center account and country of sale. Name the campaign after its job: "PMax – Hero products – tROAS 400".

Step 2: Bidding

• Choose Conversion value. Leave target ROAS off for the first 2 to 4 weeks, then set it at or slightly below the ROAS actually achieved.
• For lead gen without values, choose Conversions with an optional target CPA.
• Decide on customer acquisition: "Bid higher for new customers" to tilt toward acquisition, or "Bid only for new customers" for strictly net-new buyers. "High value" mode uses your Customer Match data to bid more for new buyers likely to have high lifetime value.

Step 3: Campaign settings that keep you out of trouble

• Locations: "Presence," not "Presence or interest," unless you ship internationally.
• Final URL expansion: on by default; Google can send clicks to any page and rewrite headlines to match. Turn it off for your first campaign, or keep it on with URL exclusions for blog, careers, returns, and account pages. Feed campaigns can also use "URL contains" rules to keep expansion inside a category.
• Brand exclusions: add your own brand list so branded searches stay in Search. Feed campaigns can apply brand exclusions to Search text ads only while keeping Shopping ads eligible.
• Your data exclusions: exclude your buyer list and recent converters if the campaign's job is acquisition. New in 2026.
• Device and demographic exclusions: generally available since 2025. Use sparingly, based on data.
• Automatically created assets: leave on if you'll review the output; off if you have legal copy constraints.

Step 4: Asset groups and listing groups

An asset group is one creative theme plus one audience signal. The ceiling is 100 per campaign, but most stores need two to five: one per category or offer that genuinely needs different images and headlines.

For each asset group, fill the full spec:

• Headlines: up to 15, max 30 characters. Mix product, benefit, offer, and brand.
• Long headlines: up to 5, max 90 characters.
• Descriptions: up to 5, max 90 characters (one at 60 or less).
• Images: up to 20 across landscape, square, and portrait. Lifestyle plus product-on-white.
• Videos: up to 15 as of 2026 (it was 5). Add at least one 10 to 15-second vertical; otherwise Google autogenerates.
• Logos, business name, call to action, final URL.

Then build the listing group: subdivide by custom label, brand, category, or product ID so the asset group only promotes products that match its creative. Exclude here anything PMax shouldn't advertise (out of stock, loss leaders, negative margin after shipping).

Step 5: Audience signals and search themes

• Audience signals are suggestions, not targeting. Add your Customer Match list, website visitors, and a custom segment built from converting search terms and competitor URLs.
• Search themes tell Google which queries you consider relevant, up to 50 per asset group. Use real converting queries from Search, not generic category words. The usefulness indicator shows which themes generated incremental traffic.

Step 6: Negative keywords and budget

• Add campaign-level negative keywords (shared lists work too): free, cheap, jobs, manual, repair, wholesale, and competitor brands you don't want to pay for. Negatives in PMax only apply to Search and Shopping inventory.
• Set the daily budget at roughly 3× your target CPA, enough for about one conversion a day. Splitting $30 a day across four campaigns guarantees four under-fed campaigns.

Launch, then leave bidding alone for two weeks while you review search terms, placements, and assets.

How to do it faster with AI: Claude, ChatGPT, and Claude Code

The slow parts are writing assets, mining search terms, and reading reports. These prompts cut each to minutes.

Write a complete asset group from a product page

Paste the product page text (or the URL if your model can browse) into Claude or ChatGPT:

```

You are a Google Ads copywriter. Using the product page below, write a Performance Max asset group.

Constraints:

- 15 headlines, max 30 characters each, no exclamation marks, at least 3 containing the product name, 3 with a benefit, 2 with an offer or guarantee, 2 with the brand.

- 5 long headlines, max 90 characters.

- 5 descriptions, max 90 characters, one of them 60 or fewer.

- Comply with Google Ads policies: no superlatives we can't prove, no "best", no medical claims.

- US English. Brand voice: [direct, warm, no hype].

- Output as a table with a character count column. Then list 10 search themes (max 80 chars each) based on how a buyer would actually search for this.

Product page:

[paste text]

```

Ask for a second pass ("rewrite any line over the limit and add 5 headlines that lead with price"), then paste into the asset group. The same prompt works for collection pages, one asset group per category.

Turn the search terms report into negatives and search themes

Export the PMax search terms report (last 30 days, with conversions and value) as CSV and paste it:

```

Here is a Performance Max search terms export. Columns: search term, impressions, clicks, cost, conversions, conv. value.

1. Group the terms into intent clusters (brand, competitor, product, informational, job/support, irrelevant).

2. Propose a negative keyword list: every term with 0 conversions and cost above $15, plus every informational or job/support term, in phrase match. Explain any term you're unsure about instead of guessing.

3. Propose up to 20 search themes from terms with ROAS above 300% that are not already brand.

4. Flag any competitor brand names so I can decide on brand exclusions.

Output three code blocks I can paste directly into Google Ads.

```

Repeat every two weeks. Claude spots the "repair" and "manual" terms a human skims past at row 400.

Analyze channel and product performance with Claude Code

Claude Code reads files on your computer and writes scripts, so it pairs well with Google Ads exports. A 15-minute flow:

1. Download the channel performance report and the product report (Listing groups → Products, with cost, conversions, and conversion value) for the last 30 days.

2. Export a CSV from Shopify with product ID, price, product cost, and average shipping cost.

3. In that folder, run Claude Code and prompt:

```

Read pmax_channels.csv, pmax_products.csv and product_costs.csv.

Join products on item ID. For each product compute gross profit per order (price excl. tax minus product cost, shipping, and 3% payment fees), POAS (profit / ad cost) and ROAS.

Give me: (a) products with ad cost > $50 and POAS < 1 that I should exclude from listing groups, (b) products with POAS > 2 and impression share < 50% that deserve their own asset group with a higher tROAS budget, (c) the channel breakdown with cost share vs conversion share so I can see if Display or YouTube is eating budget without converting.

Write the results to pmax_review.md with a short action list.

```

It writes a small script, runs it, and hands you a readable review you can rerun monthly.

Creative: generate, then test properly

Google Ads' built-in asset generation produces lifestyle backgrounds for product shots inside the asset group. For video, ask Claude or ChatGPT for a 15-second script and on-screen text, then assemble it in Canva or CapCut. Test AI sets against your own photos with asset experiments, added in 2026, instead of guessing.

Reading Performance Max reports: channels, search terms, assets

The 2025 and 2026 reporting updates are why PMax is now manageable. Check these weekly:

• Channel performance report: cost, conversions, value, and ROI by channel, with format breakdowns and diagnostics, also at MCC level. If Display takes 20% of spend and brings 2% of value, add account-level placement exclusions and strengthen your feed; you can't "turn off" a channel.
• Search terms report: the same granularity as Search and Standard Shopping, with a source column showing whether a term came from search themes, your feed, or Google's matching. Feed the negatives back in.
• Asset reports: segmentable and downloadable by asset group; ratings (Low / Good / Best) tell you what to replace. Swap "Low" assets monthly; never touch "Best" ones on a hunch.
• New customer reporting: new versus returning buyers and how many were high-value, so you can see whether the acquisition goal does anything.
• Where ads showed: now segmented by network, down to sites, apps, and YouTube channels.

Make Google PMax bid on profit, not revenue

This is the biggest lever most stores leave untouched. PMax optimizes toward the conversion value you send. If that value is the order total, two $100 orders look identical even when one carries a 45% margin and the other, after free shipping and a discount code, loses money.

Three ways to fix that:

• Cost of goods in Merchant Center + conversions with cart data. Add the cost_of_goods_sold attribute to your feed and enable cart data in your conversion tag; Google Ads then reports gross profit per campaign and product. Pros: free and native. Cons: reporting only (tROAS still optimizes revenue), ignores shipping and payment fees, and Shopify setup needs custom tag work.
• Conversion value rules. Multiply value for certain audiences, locations, or devices. Pros: quick. Cons: coarse; it can't see the margin of each cart.
• Send profit as the conversion value. Calculate each order's real margin (price without tax minus product cost, shipping, payment fees, and other costs) and send it as the Purchase value, or as a secondary conversion you promote once it has volume. Your tROAS target then becomes a POAS (profit on ad spend) target. Tools that do this include ProfitMetrics, Triple Whale, and savvy.so. Pros: bidding actually chases margin. Cons: needs accurate product costs and a relearning period.

Whichever path you pick, judge campaigns by POAS, not ROAS. A 400% ROAS campaign on 20% margin products loses money after ad cost; a 250% ROAS campaign on 60% margin products is your best performer.

Common Performance Max mistakes

• One campaign, one tROAS, whole catalog. PMax funds whatever converts cheapest, usually your lowest-margin items. Split by margin band with custom labels.
• Brand traffic left in PMax. It inflates ROAS. Exclude brand and measure PMax on non-brand only.
• Final URL expansion with no exclusions. Clicks land on your returns page.
• Moving the target every few days. Each change restarts learning. Move tROAS 10 to 15% at a time, every two weeks at most.
• No real video. Autogenerated slideshows run on YouTube regardless.
• Judging it in week one. Give it two to four weeks and 30+ conversions.
• Treating audience signals as targeting. Signals are hints; exclusions are limits.

How savvy.so does it

savvy.so connects to your Shopify store and Google Ads and calculates the real profit of every order: price without tax minus product cost, shipping, payment fees, and other costs. It shows profit and POAS per campaign and channel for Google Ads, Meta Ads, and Idealo, so you see which Performance Max campaigns actually make money. It sends each order's profit as a conversion value to Google Ads, Meta, and GA4, so your tROAS targets bid for margin instead of revenue. It also monitors competitor prices in Google Shopping and writes catalog texts and attributes for your feed. There's a free trial.

FAQ