---
title: "Issue #171 | AI Max Is A Supplement, Not A Strategy"
date_published: "2026-06-07"
date_modified: "2026-06-08"
permalink: "https://www.samtomlinson.me/insights/issue-171-ai-max-is-a-supplement-not-a-strategy/"
taxonomies:
  Categories:
    - "Article"
  Tags:
    - "Advertising"
    - "Data"
    - "google ads"
seo:
  title: "Issue #171 | AI Max Is A Supplement, Not A Strategy"
  description: "In the last few months, AI Max has become the most-pushed campaign type by Google Reps even the smaller accounts we manage are getting the the full-court press: in-platform prompts, the rep"
  canonical_url: "https://www.samtomlinson.me/insights/issue-171-ai-max-is-a-supplement-not-a-strategy/"
---

In the last few months, AI Max has become the most-pushed campaign type by Google Reps ̵1; even the smaller accounts we manage are getting the the full-court press: in-platform prompts, the rep emails, the unwanted calls from Mountain View, the Optimization Score "recommendations" demanding you turn it on & the uplift projections ("get X more conversions with AI Max") that always seem to arrive alongside a recommended budget increase.

We've been testing AI Max across a portfolio of accounts for the past year (since the open beta), with significant pickup since April when the new controls rolled out. We've tested it across a number of verticals (lead gen, B2B, services, eComm, regulated/compliance-oriented industries, high-consideration purchases) ̵1; and the conclusion I've arrived at after reviewing all the data is this:

AI Max is actually pretty good ̵1; but it isn't a replacement for traditional search.

The better way to think about it is as a volume-enhancing supplement that can unlock incremental volume at a comparable CPA/ROAS, when (and only when) the conditions are right.

If you operate in a compliance-heavy industry, the conditions (at least right now) are NOT right. Despite Google's updates that were rolled out in late April (including adding text customization controls a/k/a AI Brief, disclaimer & URL controls), I've consistently found that AI Max doesn't adhere to your guidelines ̵1; it finds ways around them. I'll go into more details below, but suffice to say that if you're in a regulated or compliance-oriented company/industry, proceed with caution.

## What AI Max Actually Is (And What It Isn't)

Before we dive into the details, here's the deal behind AI Max: it's 3 features pseudo-merged onto a standard Search campaign:

Search term matching ̵1; keywordless query expansion. Google looks at your keywords, your creative + your landing pages, then matches you to queries you never bid on. The best equivalent is a slightly smarter broad match + DSA.

Text customization ̵1; AI-generated headlines + descriptions, built in real time from your landing page, your existing ads and the query (this is the feature formerly known as automatically created assets, with better PR).

Final URL expansion ̵1; Google picks the landing page it believes best matches the query ̵1; which is (quite often) not the one you actually put into the ad. Sometimes (shocker) it's the Terms & Conditions or Careers page, so exclusions are needed.

The other major advantage of AI Max (which is also true of PMAX) is that it is given higher priority for AI Mode / AI Overview ads. In other words, it's a natural expansion of existing search campaigns.

The best way to think about AI Max is as Google's first attempt to bridge the gap between the reality of broad match keywords (which appear random, competitor-heavy and/or adjacent-but-not-exactly-relevant), evolving search behavior (thanks, AI) and fixed RSAs (which have a set of headlines and a single LP) to create ads that searchers will actually find relevant.

All of that goes to why "replacement" is the wrong mental model. Google didn't build AI Max to replace your Search campaigns; they built it to mine the queries your Search campaigns structurally can't reach. And with DSA getting folded into AI Max later this year, this is where Google is consolidating ALL of its "find the queries you're missing" machinery.

## The Case For Giving AI Max A Try

The honest case for AI Max is an incrementality case based on one statistical truth: a meaningful (and growing) share of searches are queries nobody has ever bid on. We've all seen the data on the number of searches containing more than 8 words: ~17% of all searches in Google are 8+ words, up from ~2% just a few years ago.

The reality is that since the launch of AI tools, queries have become longer, more conversational, more complex and (increasingly) hyper-specific. As more people search in more natural language, keyword lists have become less effective at capturing the universe of people seeking the solutions your product/service provides.

AI Max is interpolation: it fills the gaps between your keywords. It covers the stuff you'd never think to include.

When it works, the result tends to look like this: a chunk of incremental conversions come from net-new queries (searches that have never appeared in your search terms report) at a CPA/ROAS in the same neighborhood as your existing search campaigns. THAT is the win. If you go in hoping for a 30% reduction in CPA (i.e. what PMAX promised a few years ago), you'll be disappointed ̵1; that's not what AI Max does. Its win condition is transforming the stuff you'd never buy into leads/sales you never would have gotten otherwise at a cost that's on-par with what you're paying for standard search.

All that to say: anyone promising transformational results from AI Max is (very likely) lying to you. It's 10% ̵1; 25% incremental volume at a comparable CPA/ROAS ̵1; which is downright awesome for accounts where diminishing returns have hit your core search campaigns, or you're trying to maintain volume while mitigating CPA inflation / ROAS deflation.

It's an incremental win, not a transformation.

That all being said, 3 caveats:

The results are wildly variable. Data from dozens of campaigns shows a range of outcomes ̵1; a majority of accounts have recorded significant-but-not-eye-popping lifts. Some have gotten a few points of lift. A small-but-meaningful share has gotten nothing (which is actually an efficiency degrade).

If you're already running broad match, DSAs and/or PMax, expect less from AI Max. AI Max mines the gaps in your current query coverage for cost-effective volume. All three of those campaigns/tactics effectively strip mine the long tail, which means there's less gold to find in the gaps.

AI Max works best in industries/verticals where being a little wrong is OK. All the test data converges around this point: AI Max is wonderful for industries where it has some room to be wrong. Think: a law firm that accepts personal injury cases, medical malpractice cases and the like. A home services company that does windows, doors, gutters & roofs. An eComm brand that sells face serums, overnight creams, lip plumper, face toner and lip gloss. A marketplace that sells multiple different types of a given product. The more hyper-specific/restrictive your niche, the less AI Max can be helpful.

## The Math That Matters

The single biggest mistake I see advertisers make with AI Max: evaluating it on blended performance. Blended numbers will lie to you in BOTH directions. AI Max can cannibalize queries your exact match keywords would've caught anyway (making it look better than it is), or it can drag in low-intent long-tail volume (making blended CPA look worse even while the incremental conversions are profitable).

The only question that matters: of the conversions attributed to AI Max, how many would not have happened without it? What did I pay for those?

Practically, that means:

Don't run AI Max if your impression share on the high-converting, high-quality keywords/themes in your account is miniscule. The goal of AI Max is to mine the long tail ̵1; so if you aren't even making in a dent in the obvious stuff, you're (almost always) going to see better returns there.

Run AI Max either as (a) a proper campaign experiment using Google's built-in experiment feature or (b) a parallel campaign with high-intent / high-value keywords excluded as Exact Match. Google will push hard for immediate adoption; resist that. I've had the most success with option (b) for accounts with sizable spend (>$50k/mo) and option (a) for smaller accounts (under $50k/mo).

Watch the share of conversions coming from net-new queries. That's the single-most-reliable proxy for incrementality. If AI Max is mostly "winning" queries your keywords already covered, you haven't gained anything. You've just rerouted attribution.

Hold it to your existing efficiency target. "Comparable CPA/ROAS" is the standard. If incremental volume only comes at a 40% CPA premium, that's not expansion ̵1; it's paying more for lower-quality traffic. Hard pass.

Finally: give the algorithm a fair fight. AI Max uses Google's machine learning system, which is only as good as the signal it is given. Sporadic/low-volume conversion data, insufficient budget (aim for 1x tCPA per day to start) and/or a convoluted, difficult-to-understand landers will make it more difficult to get AI Max to work.

## A Few Other Tips

In addition to everything above, here are a few other points I've found to be helpful in working with AI Max:

Exclude your brand. AI Max (like PMAX) will hoover up branded queries if you give it the option. Don't.

Be Clear + Specific on the Guidelines. Invest a significant amount of time crafting real guidelines for automatically created assets. Give examples where possible (i.e. "Do NOT use competitor names in ad copy. Examples of competitors are X, Y and Z.")

Use Page Feeds. You can include AND exclude page feeds/specific URLs. Use it. Create page feeds for categorical exclusions (privacy policy, terms & conditions, careers page, help documents, etc.), strategic exclusions (blogs, PDF guides, webinar pages, etc.) and individual product types/departments. The last thing you want is AI Max routing a ton of traffic to a lower-value offer/service/product, but using the efficiency target from the higher-margin product/service.

Consider Excluding Competitors. AI Max will go HEAVY on competitors if you allow it. For some brands, that might be OK. For others, it's a definite no-go. The same is true for commonly-associated brands/partners (i.e. Shopify or Klaviyo for eComm B2B SaaS).

## The Compliance Problem

If you only read one section of this newsletter, read this one.

If you operate in a compliance-heavy category (financial services, healthcare/medical, legal, insurance, anything with mandated disclosures or restricted claims), you need to understand how an optimization system actually behaves. AI Max has ONE objective: maximize conversions/conversion value within its constraints. It does not understand why your constraints exist. It doesn't care. To the algorithm, your guardrails aren't principles to honor ̵1; they're obstacles between it and its objective. And, if the last few years have taught us anything, it's that optimization systems are incredibly adept at finding paths around obstacles.

Don't take my word for it. Take Google's: the decks their reps share about AI Max + all their new, shiny "controls / safeguards" carry a recurring footnote warning that using controls can limit your performance uplift.

Put another way, Google hands you guardrails with one hand and tells you they're costing you conversions with the other. That's the entire incentive structure in one sentence. Even the "fact-checking" Google touts is circular: generated copy gets validated against your own landing page. If your site says it somewhere, in some context, the system considers it fair game. Most LPs I've reviewed have bite-sized phrases that, taken completely out of context and used as a headline, could be problematic. Hell, I think if you excerpted a handful of words from this newsletter and used it in a headline, it could be problematic.

Concretely, here's what "finding a way around" looks like in practice:

Text customization can write claims you never approved. If a page mentions clinical outcomes, performance stats or testimonial results, the AI can happily promote those into a headline, stripped of the context + qualifiers that made them defensible on the page. The ad run now says something materially and substantially different, even though some of the words are identical.

Query expansion will match you into intent you're prohibited from targeting. The system doesn't see a bright line between "energy support supplement" and "ADHD treatment," or between "retirement planning" and "guaranteed returns." It sees semantic adjacency + conversion probability. Adjacent intent is often both a fertile source of potential conversions AND a legal/regulatory/compliance minefield.

Final URL expansion will route users to pages missing required elements. Most brands have designed their PPC landers with the right disclosures, risk language, approved case studies, up-to-date stats, etc. URL expansion can decide your blog from 2021 is "more relevant" to the query. The end result? Potentially expensive, adjacent-intent traffic is landing on a page that was never designed or reviewed for it. That might not be a problem. Google might genuinely be right that the blog is the best place for some of that traffic. But it's also impossible to defend against infinite options.

And ̵1; to be clear ̵1; none of this is "blaming" Google. Nothing I've described above is Google "breaking the rules" or "violating guidelines" ̵1; it's the system working exactly as intended. All of your ads/LPs are still reviewed and pass Google's policy checks ̵1; but that's the trap: platform compliance and regulatory compliance are NOT the same thing. Google's policies are designed to protect them. Your compliance/legal/regulatory policies are designed to protect you ̵1; but Google doesn't know those.

To Google's credit, the control levers available are improving ̵1; text disclaimers can now lock required language into your ads even with URL expansion enabled, and text guidelines let you write natural-language restrictions on what the AI can + can't say. Specific exclusions are helpful in preventing universal "bad words" from ever appearing in your ads. My recommendation is to use all of them.

A few other things to keep in mind:

Pinning doesn't work with Final URL Expansion. The dedicated disclaimer asset is the only text guaranteed to serve.

Disclaimers are capped at 90 characters, apply at the campaign level & override whatever you had pinned to description line 1.

If your disclaimer is disapproved, the campaign doesn't pause (surprise!). It keeps serving WITHOUT it, falling back to a standard description line. For most advertisers, that isn't the end of the world ̵1; but having a backup Description 1 that covers your bases is not a bad idea.

These are probabilistic constraints on a probabilistic system, with a failure mode that defaults to "spend." If you operate in an industry where ad copy issues can result in bigger problems (i.e. lawyers, doctors, healthcare, etc.), here are some practices we've found to be helpful:

Either keep URL expansion off, or aggressively exclude pages that aren't intended for paid traffic. URL feeds are the best way to do this (unless you want to spend hours copy-and-pasting).

Lock disclaimers in with the dedicated disclaimer asset, pin anything that absolutely must appear in non-FUE ads. Check the status of the disclaimer regularly (daily is ideal; weekly if not). This is a 2-minute task that can save you hours of aggrivation. It's worth it.

Check the Asset report regularly ̵1; this is the best way to review what assets Google is generating on your behalf.

Review the Search Terms Report constantly for both (a) terms/themes that have significant potential + should be moved to standard search and (b) any terms/themes that are NOT relevant and should be excluded.

## The Bottom Line

AI Max can be a fantastic supplement to your existing search campaigns that delivers significant incremental volume at comparable efficiency. Evaluate it on incrementality, not blended metrics. Give it real budget + a high-quality, consistent signal.

And if you're in a regulated category, proceed with caution. You're handing creative + targeting authority to a system that optimizes around constraints and has no incentive to understand the "why" behind a restriction that a person would.

If you go into it with that frame of mind + you're willing to stay on top of it (especially at the beginning), it can be a great addition to your account.

That's it for this week ̵1; thanks for reading!

Cheers,

Sam