Built for Google's AI.
Won on the feed.
The Alpaca Ads methodology for structuring an ecommerce Google Ads account in 2026. A consolidated, four-campaign system that feeds Google's AI clean signal and concentrates budget on what's proven to drive revenue.
The old search playbook now starves the machine.
Single-keyword precision and manual bids fought for control over an auction Google has now automated. The old tactics don't just underperform. They cut the AI off from the signal it runs on.
Single-keyword ad groups
One keyword per ad group splits your conversion data into pools far too small for Smart Bidding to learn from.
Exact match & manual bids
Hand-set bids and exact-match-everything fight for a precision the auction stopped rewarding years ago.
Mirrored, fragmented campaigns
Many near-identical campaigns divide the conversion signal, and every bid strategy ends up under-fed.
Consolidate the account, feed the AI clean signal, and concentrate budget on what is proven to drive revenue. Every decision in this playbook traces back to that one line.
Google Ads in 2026 is an AI-native auction. Three systems now run the account: Performance Max matches your products to demand across every Google surface; AI Max extends that keywordless matching into Search campaigns; and Smart Bidding sets every individual bid from your conversion data. Together they have removed the levers media buyers spent fifteen years pulling: keyword, placement, audience, bid. What you steer instead are the inputs: the feed, the conversion signal, audience signals, bid targets and exclusions. Two consequences drive everything below: the feed and the conversion signal are the new targeting, and fragmentation starves the engine (Smart Bidding learns per campaign, so consolidation concentrates signal and shortens learning).
Five principles behind every decision.
Steer with inputs, not levers
You no longer set keywords, placements and bids directly. You set the feed, the conversion signal, audience signals, bid targets and exclusions. Win by making those five inputs excellent, not by micromanaging an auction you can't see.
Concentrate on proven revenue drivers
Roughly 80% of budget goes to campaigns and products with a measured, profitable return; roughly 20% funds discovery. Pareto is brutal in ecommerce. Honour it at both campaign and product level.
Ring-fence brand
Brand searches are demand you already created, captured, not acquired. Separate them completely so brand performance never flatters Performance Max or non-brand Search.
Protect the signal
One clean conversion setup: Enhanced Conversions on, deduplicated, one primary purchase event, real revenue passed at the product level. Smart Bidding is only ever as good as the data you feed it.
Feed the learning, then leave it
Smart Bidding needs conversion volume and stable conditions to learn. Don't restructure, don't swing budgets, don't move bid targets daily. Patience is a performance setting.
Get the signal right first.
Google's AI is only as good as the feed and the conversion data it learns from. None of the structure below performs if the plumbing is wrong.
One Merchant Center, one healthy feed
All required and recommended attributes populated, valid GTINs, no disapprovals. Front-load titles with the terms buyers actually search.
Custom labels for segmentation
Tag every product by margin tier, bestseller status, price band, season and stock. These labels are how you concentrate budget on hero products later.
Enhanced Conversions, deduplicated
Enhanced Conversions for Web on and verified, first-party hashed data flowing, one primary purchase event, deduplication confirmed.
Value-based, not volume-based
Pass actual revenue dynamically with every purchase, ideally net of COGS, so the model optimises to profit, not turnover.
Customer Match + new-customer goal
Upload purchaser and high-LTV lists, then configure the new-customer acquisition goal so PMax bids up for first-time buyers.
Brand list + naming locked
Define the brand term list. It drives both the Brand campaign and the PMax brand exclusion. Lock the naming convention before launch.
Four campaigns. That's the whole account.
A consolidated full-funnel build. Each campaign has a single job, and brand is ring-fenced so every read stays honest.
Performance Max, Shopping-led
Where the majority of spend and most new-customer revenue happens. One campaign matching your products across Search, Shopping, YouTube, Display, Gmail, Discover and Maps.
Brand Search, exact / phrase
Cheap, defensive, and the highest-ROAS line in the account, but only useful if you keep it honest and ring-fenced from everything else.
Non-brand Search, AI Max
High-intent generic and category queries. The transparent complement to PMax. Here you see the actual search terms and keep a hand on the wheel.
Demand Gen, YouTube / Discover / Gmail
The other three campaigns capture demand. This one creates it. Visually-led ads across YouTube, Shorts, Discover, Gmail and Maps. This is the ~20% discovery allocation.
There is deliberately no standalone retargeting campaign. PMax already serves returning visitors automatically, Demand Gen carries dedicated retargeting audiences, and repeat-customer value is handled through Customer Match signals and the new-customer goal. A separate retargeting campaign would just fragment signal and double-count conversions the other campaigns already earn.
Naming conventions.
Consistent names make reporting legible and stop accidental overlap between campaigns. Lock this before launch.
One structure. Every budget.
The four-campaign architecture is the target state. At lower spend you collapse toward it; at higher spend you extend it, but you never fragment it.
Lean
Performance Max + Brand Search. There isn't enough conversion volume to feed four bid strategies.
- PMax does the acquisition work, broad asset groups
- Brand Search stays ring-fenced
- Add Search and Demand Gen once volume supports them
Mid
The full architecture exactly as specified: PMax, Brand, Non-brand Search and Demand Gen all running.
- PMax brand-excluded, tiered asset groups
- Brand ring-fenced, non-brand Search on AI Max
- Demand Gen carrying the discovery layer
Scaling
Extend, never multiply. Options in order of preference:
- Segment PMax by margin / category tier
- Add Standard Shopping as a control layer
- Split Demand Gen into prospecting + retargeting
- Move to portfolio bid strategies
The mistake at every tier: duplicating campaigns to "scale." Raise budgets on what works in 20–30% steps and improve the feed. Duplication just splits the signal you worked to concentrate.
Under Google's AI, the feed is the media plan.
The structure only performs if it's fed. Product titles and attributes decide which searches you show for, and custom labels are how you steer budget to the products that pay.
Product titles are the single highest-leverage asset you control. Front-load them with the terms buyers search and fit the most important attributes into the first ~70 characters. Custom labels (margin tier, bestseller, price band, season, stock) are what let asset groups and tROAS targets follow the products that actually make money. And every PMax asset group needs video you produce yourself: leave it blank and Google auto-generates a weaker one in your place.
The AI Max rollout
Check the prerequisites: Enhanced Conversions, full negative lists, quality landing pages, 100+ conv/mo.
Pick your single strongest Search campaign. Never roll AI Max out account-wide.
Launch the built-in 50/50 experiment against the unchanged control.
Run a full 4–6 weeks, through the learning period, to significance.
Scale only on a clear positive, then repeat on the next campaign.
Where the money goes.
A mid-tier reference split. Adjust to LTV and margin. The principle underneath: ~80% on proven revenue drivers, ~20% on discovery.
Bidding, value first
Default to Maximize Conversion Value. Launch with no tROAS so Smart Bidding can gather conversions, then add a tROAS set to profit reality. A target too high just throttles delivery.
Volume, clear the bar
Smart Bidding wants roughly 30+ conversions per 30 days per campaign to optimise well. Budget each campaign to clear that bar. If it can't, consolidate further.
Scaling, steady steps
Raise budget in 20–30% increments, then hold 1–2 weeks. Move one lever at a time. Demand-led pacing (GML 2026) flexes daily spend, but doesn't replace deliberate decisions.
Read it right. Touch it less.
Read at campaign, asset-group & product level
Use PMax channel reporting. If spend drifts to Display instead of Shopping, the feed or asset groups need attention. Keep concentrating on the products that win.
Strip brand out before you judge
Brand demand inflates blended ROAS. Judge PMax and non-brand Search on their non-brand numbers only, and read MER and new-customer CAC as the true north stars.
Respect the learning period
After launching or editing, leave it 1–2 weeks before judging or adjusting. Never judge mid-learning or on too few conversions.
Defensible in 2020. Costing you now.
Every one of these once made sense under manual bidding. Against Google's AI they all actively work against the engine.
Letting Performance Max absorb brand traffic and counting it as acquisition. It flatters ROAS and hides true new-customer cost.
One giant PMax with every product in a single asset group. Hero products subsidise dead stock and you can't steer.
Single-keyword ad groups and hyper-segmented Search. They starve Smart Bidding of the volume it needs to learn.
Exact-match-everything with manual bids. Fights an auction that no longer rewards that precision.
Setting tROAS too high, too early. Throttles delivery and stalls learning before the model has a chance.
No negative keyword management. The cardinal sin of the AI era. Broad match and AI Max will find queries you never wanted.
Enabling AI Max account-wide overnight instead of proving it with a 50/50 experiment.
Restructuring, swinging budgets over 30%, or moving bid targets weekly. Each one resets learning.
Treating PMax as a creative exercise and ignoring the feed. The feed is the campaign.
Optimising to last-click or front-end conversions instead of value. You get volume, not profit.
Running PMax with no video assets. Google auto-generates weak ones in your place.
Duplicating campaigns to "scale." Splits the signal you worked to concentrate.
Judging PMax without channel, asset-group and product reporting. That's flying blind.
From zero to rhythm in four weeks.
A clean launch sequence. Foundations first, core campaigns next, the funnel once the engine is out of learning.
Foundations
- Merchant Center feed health + custom labels
- Enhanced Conversions verified + deduplicated
- Value-based tracking, real revenue
- Customer Match + new-customer goal
- Lock the naming convention
Launch core
- Campaign 01, PMax, brand-excluded, tiered groups
- Maximize Conversion Value, no tROAS yet
- Campaign 02, Brand Search
- Let learning settle untouched
Add the funnel
- Campaign 03, Non-brand Search, broad + Smart Bidding
- Campaign 04, Demand Gen
- Introduce a tROAS on PMax
Establish rhythm
- Run the weekly loop
- Start the AI Max 50/50 experiment
- First feed + asset refresh
- Tier tROAS by product margin
This is how Alpaca Ads
runs Google.
Use it as the default structure for every ecommerce account. Adapt only with a reason, and write the reason down.
Paid Media Methodologies · Google Ads · v1.0