AI Max on brand search: how it added spend with no efficiency gain, and how to test it safely

AI Max on brand search can inflate your brand spend significantly without improving efficiency. One team found it added roughly 400,000 rupees (about $5,000) in brand spend over a short window with no meaningful change in cost per funded lead, and rolled it back. The lesson is not that AI Max is broken. It is that brand and non-brand require different testing logic.

What AI Max actually does on brand search

AI Max is Google's expansion layer for Search campaigns. It uses broad match signals, query matching expansion, and landing page optimization to reach queries the original keywords would have missed. On non-brand terms, that expansion can surface real incremental volume. On brand terms, the dynamic is different.

When AI Max is active on a brand campaign, it can:

• Match to navigational queries your exact-match brand keywords already cover
• Bid on brand-adjacent or competitor-branded queries you did not intend to enter
• Pull budget toward brand queries that would have converted organically anyway

The result is more brand impressions, more brand clicks, and more brand spend, but not necessarily more conversions that would not have happened without the paid click.

The case: four lakh rupees in brand spend, zero efficiency gain

The team in this example ran AI Max on an existing brand Search campaign. Over the test period, brand spend increased by roughly 400,000 rupees (approximately $5,000). Conversion volume was up in absolute terms, which looked good on first read.

The problem appeared when they compared cost per funded lead against the control period and against the non-AI Max brand campaign running in parallel. The CPA had not improved. The volume increase was roughly proportional to the spend increase, meaning AI Max was buying the same quality of conversion at the same price, but in larger quantity, by capturing queries that likely would have converted through organic or existing exact-match brand coverage.

There was no efficiency gain. The incremental spend was not justified by incremental business impact. They rolled it back.

This is brand cannibalization in its most common form: paid spend absorbing demand that already existed rather than creating new demand.

Why brand cannibalization is hard to catch in the dashboard

The standard Google Ads dashboard will show you more conversions when brand spend goes up. That is true almost by definition. The question the dashboard does not answer automatically is: how many of those conversions would have happened anyway?

Three signals that suggest cannibalization rather than incrementality:

1. Organic brand clicks drop at roughly the same rate brand paid clicks rise. Check Search Console alongside Ads during the test.
2. Impression share on brand exact match stays flat or rises while total brand spend increases. You are not winning new auctions, you are paying more in the ones you already won.
3. CPA on brand holds steady or worsens even as volume grows. If the new volume were genuinely incremental, you would expect some efficiency improvement from the additional signal.

None of these are definitive on their own. Together, they build a case.

How to test AI Max on non-brand safely

Non-brand is where AI Max has a more defensible value proposition. Expansion into queries you did not think to bid on, better landing page matching, and broader signal use can surface volume that would not have come through a tightly managed exact-match non-brand campaign.

But non-brand tests still need structure. Here is a framework that gives you a real read without open-ended exposure.

Set a hard end date before you launch

AI Max tests have a way of running indefinitely because the results look directionally positive and nobody formally closes them. Set the end date in the campaign name and in your project management tool before you hit publish. Four weeks is usually enough to see CPA stabilize. Six weeks if your conversion window is longer than seven days.

Define your CPA threshold against a funded-lead benchmark

Before launch, agree on the number that would cause you to pause or roll back. Frame it as: if CPA on this AI Max non-brand campaign exceeds X by week two, we pause and review.

X should be your current non-brand CPA plus a tolerance for learning phase variance. A reasonable starting tolerance is 20 to 25 percent above your control CPA. If you are paying $80 per funded lead on non-brand today, set the pause threshold at $96 to $100. If the campaign is still above that at the midpoint of your test window, you have your answer early.

Use funded lead or qualified lead as your conversion event, not raw form fill. AI Max expansion will find volume. The question is whether that volume converts to something your business cares about.

Run a geographic or budget-isolated control

If you can split by geography, run AI Max in a subset of markets and keep your existing campaign structure in the rest. If geography is not clean, run both structures simultaneously with a fixed budget cap on the AI Max variant. The goal is a comparison point that was live at the same time under the same market conditions.

Avoid comparing AI Max performance to a prior period. Seasonality, competitor activity, and offer changes make period-over-period non-brand comparisons unreliable.

Keep brand and non-brand in separate campaigns

This sounds obvious but is frequently skipped. If AI Max is active on a campaign that contains both brand and non-brand keywords, you cannot isolate what is driving the results. Separate the campaigns before you start. Brand in one, non-brand in another, AI Max toggled only on the one you are testing.

Watch the query report weekly

AI Max expansion means queries you did not explicitly bid on will start converting. Pull the search terms report every seven days. Flag any brand queries appearing in the non-brand AI Max campaign. If brand queries are getting significant spend, you have a structural problem: the expansion is bleeding into brand territory and inflating your non-brand CPA with conversions that should be attributed to brand intent.

You can add negative brand keywords to the non-brand campaign to contain this, but do it before launch, not after you notice the bleed.

The decision tree at test close

At the end of your test window, you have three outcomes:

CPA is at or below threshold, volume is up. AI Max is adding incremental non-brand volume efficiently. Expand budget gradually and keep monitoring the query report.

CPA is above threshold but volume is significantly up. You have a scaling problem, not a structural one. Tighten landing page targets, add negative keywords, and retest at a lower budget ceiling.

CPA is above threshold and volume gain does not justify it. Roll back. Document what queries drove the overspend. Use that list to inform future keyword expansion manually.

Where Almond fits

The case above failed partly because the team did not have a clear pre-agreed exit rule, and partly because the context around the test, the funded-lead threshold, the brand CPA baseline, the decision to roll back, lived in a Slack thread that was hard to retrieve when someone asked why the campaign was paused three weeks later.

Almond's Business Context Engine captures that kind of test record automatically, versioned and linked to the account decision it relates to. When the next person asks whether AI Max was tried on brand before, the answer is in the system with the reasoning attached, not in someone's memory. That is the compounding value: not just running better tests, but keeping the learning from each one so it actually informs the next.