# Build a Performance Max Campaign With AI

Build PMax by prompt: brand assets, a paused campaign, then a complete asset group — headlines, images, logos — reviewed in the UI before enabling.

Tutorials · Published 2026-08-10, updated 2026-10-26.

> To build Performance Max with an AI agent: create the business-name and logo assets first (PMax requires both at campaign level), then create the campaign paused with a budget and a maximise-conversions strategy (optionally with a target CPA), then build an asset group in one batch — final URL, headlines, long headlines, descriptions, landscape and square images, logos. AdCopilot proposes each write first, and your AI client asks before it runs if it is set to ask before AdCopilot's write tools. Review everything in the Google Ads UI, then enable with one status change.

A Performance Max build with an AI agent runs in a strict order: brand
assets first, then a paused campaign, then an asset group assembled in one
batch, then review, then enable. The order is not style — PMax requires a
business name and logo before the campaign exists, and asset groups must
arrive complete. Every step below is a prompt, and each write in it is one AdCopilot proposes first, and your AI client asks before it runs if it is set to ask before AdCopilot's write tools; a tool set to always allow runs without asking.

This is the doing page. What PMax shows you, what it hides, and whether
you should run it at all is
[the Performance Max pillar](https://adcopilot.cloud/performance-max-ai) — read it first if PMax
is new to you. The search-campaign equivalent of this walkthrough is
[launch a campaign with AI](https://adcopilot.cloud/tutorials/launch-a-campaign-with-ai).

## Gather the inputs the agent cannot invent

Text, the agent drafts. Everything else is yours to bring:

- **Hosted image URLs** — landscape 1.91:1 (1200×628), square 1:1
  (1200×1200), and a square logo. Real product or brand images, publicly
  fetchable.
- **The final URL** — where PMax sends everyone this asset group wins.
- **Budget and bidding** — maximise conversions is the sane default;
  target CPA or target ROAS only if the account has the conversion
  history to feed them.
- **The brand facts** — exact business name spelling, and two sentences
  on the offer for copy drafting.

## Step 1 — brand assets

```text
In account <account>, create the brand assets for a new Performance Max
campaign:
1. a text asset with our business name: "<exact name>"
2. an image asset from this logo: <logo URL>, named "<name> logo square"
Give me both resource names when done.
```

Two quick writes — one text asset, one image asset — and
the agent holds the two resource names the campaign creation call
requires. Assets are inert on their own; nothing here can serve or spend.

## Step 2 — the campaign, paused

```text
Create a Performance Max campaign in account <account>:
- name: "PMax - <theme> - <month>"
- daily budget: $25
- bidding: maximise conversions
- business name and logo: the assets you just created
- status: paused, and keep it paused through the whole build
Confirm the campaign ID.
```

One campaign-creation write. Paused is the default for
campaigns created through the connector, and on PMax that default earns
its keep twice over — a live PMax campaign starts exploring immediately,
so you want the entire asset group reviewed before the first auction.

If you have a working target, say it in the same prompt — "target CPA of
$12" — because bidding is set at creation. And if you do not want Google
expanding beyond your final URL, say so now too: URL expansion is a
campaign-creation setting, and "opt out of final URL expansion" belongs
in this prompt, not in a later regret.

## Step 3 — the asset group, in one batch

Asset groups are created whole — the group and its assets land together.
Draft the text in chat first, then let the agent assemble:

```text
Draft the text for asset group "<theme>" (final URL: <URL>):
- 5 headlines under 30 characters
- 4 long headlines under 90 characters
- 4 descriptions under 90 characters, one under 60
Value proposition: <two sentences>. Show me everything for edits before
creating anything.
```

Edit the copy in the conversation — this is where your time should go —
then:

```text
Copy is final. Create the text assets, create image assets from these
URLs: <landscape URL(s)>, <square URL(s)>, and build asset group
"<theme>" on the new campaign with all of it plus the logo. Confirm what
was linked.
```

Expect a chain of writes: text assets, image assets, then one
asset-group call carrying the final URL, every headline,
description and image. If Google rejects an image for size or ratio, the
agent gets the error back verbatim and can tell you exactly which file to
re-export — ask it to quote the error rather than summarise.

## Step 4 — review where Google shows you the preview

Open the campaign in the Google Ads UI. Each prompt from your client showed
one call, and only if it was set to ask before AdCopilot's write tools; the UI review verifies the whole:

- Asset group preview renders sensibly across formats
- Ad strength verdict and whatever Google says is missing
- Final URL is right — and URL expansion is set the way you chose
- Budget, bidding and target match the brief
- Status: paused, everywhere

Ask the agent for its own inventory — `Summarise the PMax build: campaign
settings, every asset, the asset group` — and check the two accounts of
reality agree. Then, if signals allow it in your account, add audience
signals in the UI while you are there; they steer early exploration and
are worth the manual minute.

## Step 5 — enable, then watch the right things

```text
Everything is reviewed. Enable campaign "PMax - <theme> - <month>".
```

One status-change write — the only spend-affecting call of the build, and your AI client asks before it runs if it is set to ask before AdCopilot's write tools.

Then set the monitoring expectation honestly: PMax reports
less than search campaigns do, so your weekly check focuses on what is
actually queryable:

```text
It's been a week. For the PMax campaign: spend, conversions, cost per
conversion, and asset group performance. What can you see about where
spend went — and tell me plainly what you cannot see.
```

An agent that states its blind spots is reporting correctly — channel
split is [mostly hidden by design](https://adcopilot.cloud/performance-max-ai). Fold the
campaign into the
[weekly routine](https://adcopilot.cloud/tutorials/weekly-optimization-routine-with-ai) and give
it two to three weeks of learning before judging it.

The whole build is a forty-minute session on a connected account:
[Start free](https://app.adcopilot.cloud/signup), bring a logo
URL and two product images, and run Step 1. By the end you will have a
complete PMax draft — reviewed, paused, and spending nothing until you
say the word.

## Questions

### Why does the build start with a business name and logo?

Because Google requires them. A Performance Max campaign carries brand guidelines — a business-name text asset and a logo image asset linked at campaign level — so those two assets must exist before the campaign that references them. The agent creates both in seconds; the only thing it needs from you is the exact brand spelling and a hosted logo image.

### What images does the asset group actually need?

At minimum: landscape marketing images at 1.91:1 (1200×628 works), square images at 1:1 (1200×1200), and a square logo. Provide them as hosted URLs and the agent creates image assets from them. Text it can draft itself — headlines under 30 characters, long headlines and descriptions under 90 — but images are the input only you can supply.

### Will I be able to see where PMax spends once it is live?

Partially, and it is worth knowing exactly where the boundary sits. Asset-group level performance and overall campaign metrics are queryable, and the agent will report them honestly. Channel-level distribution — how much went to Search versus YouTube versus Display — is largely not exposed by Google to anyone, agent or human. Distrust any tool that claims otherwise.

---

This is the markdown copy of https://adcopilot.cloud/tutorials/performance-max-campaign-with-ai. The index of the site's pages is https://adcopilot.cloud/llms.txt.
