Google AI Max for Search: The UK eCommerce Reality Check

If you judge Google's AI Max on cost of sale, it looks like a disaster. In the UK eCommerce accounts we manage, search terms matched by AI Max keywordless targeting returned a 61.5% cost of sale over the last 90 days. The keywords those same advertisers chose themselves returned 10.9%.

Six times worse. Switch it off, surely.

Except that is the wrong conclusion, drawn from the wrong measurement. And from 1 September 2026, for a lot of advertisers, it stops being a choice anyway.

What is Google AI Max, exactly?

AI Max is not a campaign type. It is a toggle inside a Search campaign — Google describes it as "a continuous optimisation layer for your search campaigns". You switch it on within a campaign you already have.

It bundles three things:

• Search term matching — matching your ads to queries using your landing pages and assets, not only your keywords. On by default when AI Max is on, controlled at ad group level.
• Text customisation — generative headlines and descriptions built from your existing ads and landing pages. Campaign level.
• Final URL expansion — Google choosing the landing page. Campaign level, and off by default in the September migration.

The mechanical detail that catches people out is in Google's own developer documentation: when AI Max is enabled, all keywords are treated as broad match by default, and the campaign-level match type setting stops applying. Turning AI Max on quietly converts your carefully-built exact and phrase keywords into broad-match behaviour, unless you disable search term matching at ad group level.

That is why the accounts most exposed are often the tightly-structured ones that assume their match-type discipline protects them.

Is AI Max opt-in or is it being switched on for me?

Both, and the balance has shifted decisively towards the second.

New Search campaigns have AI Max on by default — Google states this plainly in its help centre. And from 1 September 2026, Google begins automatically upgrading existing Search campaigns that use campaign-level broad match or standalone automatically created assets into AI Max. Search term matching is switched on for both migrated groups.

Worth sitting with: automatically created assets governed how your adverts were written. Search term matching governs which searches you appear for. Advertisers who agreed to the first are being migrated into the second.

There is no decline button. The only avoidance route was to switch off broad match or automatically created assets before the migration — and since 3 August 2026 you can no longer create either, so that exit is one-way.

How exposed are UK eCommerce accounts already?

We audited every enabled Search campaign across nine UK eCommerce accounts we manage. Client names and figures are anonymised and pooled throughout.

That third row is the one to understand. The Google Ads API exposes a field called bundling_required. Where it reads REQUIRED, that campaign has adopted AI Max at some point, and its text asset automation and brand lists now live inside AI Max. Switching AI Max off does not return you to the old setup — it turns those features off entirely, brand exclusions included.

Two of our ten currently have AI Max switched off. They cannot re-add brand list targeting without switching it back on. That is a door that only opens one way, and nothing in the interface says so.

What does AI Max traffic actually do?

Pooled across the five accounts with AI Max activity, 20 May to 18 August 2026. CoS is cost of sale — advertising cost as a percentage of the revenue it produced. Lower is better.

Two things jump out, and they pull in opposite directions.

AI Max keywordless converts at a higher rate than the advertisers' own keywords — 8.58% against 6.81%. But the average order value behind those conversions is roughly £12, against £84 for keyword traffic. It finds buyers. It finds them buying small things.

Two honest caveats. This is 57 conversions — a small sample, and not enough to call a trend. And AI Max is still tiny in these accounts: £905 of £38,354 in search-term spend, about 2.4%.

Why we treat AI Max as a top-of-funnel play

Here is the reframe. A 61.5% cost of sale is a catastrophe for a harvesting campaign. For a discovery layer, it is a cost of entry — provided something downstream is doing the harvesting.

In the same accounts, over the same 90 days, Performance Max spent £258,236 and returned £2,671,877 — a 9.7% cost of sale. That is where the money is made. Search sat at 15.9%.

So we run AI Max deliberately as a top and middle-of-funnel play: let it find query space our keyword lists never reached, and let Performance Max convert the interest into purchases. Judging the discovery layer on last-click return is like judging your shop window on till receipts.

We have to be straight about the limit of that argument. Platform data cannot prove that Performance Max converted demand AI Max created — last-click attribution will never show it, and Google publishes no incrementality tooling here. It is our operating model and our reading of the data, not a proven causal chain. If someone shows you a clean number for this, ask how they got it. Our wider view on why efficiency beats raw return sits in spend versus efficiency, and how the harvesting half works is in the new era of Performance Max.

What Google's own numbers do not cover

Google publishes headline claims for AI Max — 14% more conversions at a similar cost per acquisition, and 7% from using the full feature suite. Read the footnotes. Every AI Max for Search figure Google publishes is qualified "for non-Retail advertisers".

If you sell products online, Google has published no performance figure for this product that describes you.

The largest independent dataset — over 250 campaigns across 600+ accounts — found conversion value up 13% but cost per acquisition up 16%, a median return on ad spend change of zero, and a range from +42% to −35%. Roughly a fifth of campaigns landed near their original target. The volume half of Google's claim replicates. The "at similar efficiency" half does not.

What good looks like

Before 1 September, on every Search campaign:

• Build brand exclusion lists — your own brand, competitors, and common misspellings, which Google's filters miss.
• Pre-build negative keywords for informational and unprofitable intent — "free", "how to", "jobs", "login".
• Check your bidding. Search term matching does not function on manual CPC.
• Leave Final URL expansion off. It defaults off for migrated campaigns. Turning it on breaks asset pinning and can route paid traffic to your returns page.
• Consider disabling search term matching at ad group level on brand and high-intent ad groups, keeping it on for prospecting. This is the surgical control most advertisers never find.
• Baseline everything now. After migration you cannot reconstruct the before.

Do this one thing this week

Run this against every Search campaign you own, before 1 September:

SELECT campaign.name,
       campaign.ai_max_setting.bundling_required,
       campaign.ai_max_setting.enable_ai_max
FROM campaign
WHERE campaign.advertising_channel_type = 'SEARCH'
  AND campaign.status = 'ENABLED'

Anything returning REQUIRED has already crossed over, whether anyone told you or not. That single field tells you where you actually stand — and it is not shown anywhere in the Google Ads interface.

AI Max sits inside a much wider shift in how Google search works, which we track in the 2026 Google AI search shift. If you would rather someone ran this audit across your account before the deadline, that is part of what we do in paid search management.