Your Search campaigns are on AI Max now. Here is how to tell what it actually changed

The September auto-upgrade has run. If your Search campaigns used campaign-level broad match or automatically created assets, they are on AI Max whether you chose it or not, and your October numbers look different. The hard part is not finding a change. It is working out which part of the change AI Max caused, and the comparison most people reach for cannot tell them.

The two headline numbers describe different advertisers

Start here, because almost every argument about AI Max is two people quoting statistics that do not overlap.

Google's figure: advertisers who turn AI Max on typically see 14% more conversions or conversion value at a similar CPA or ROAS. That comes from Google's internal 2025 data, and it is for advertisers outside retail. A narrower comparison, full feature suite against search term matching alone, gives 7%.

Lunio's figure, published 12 August 2026 from a dataset of more than 414 million clicks recorded between October 2025 and June 2026: invalid traffic on retail search campaigns with AI Max enabled rose from 2.46% in Q4 2025 to 5.28% in Q2 2026, while standard search campaigns moved the other way over the same period, from 3.72% down to 3.07%. That is 72% more invalid traffic on AI Max than without it, and retail AI Max campaigns accounted for 68% of all invalid clicks in the search dataset examined.

Neither number is dishonest and neither answers the other. The uplift claim excludes retail; the invalid traffic finding is specifically retail. If you sell products online, the number that was measured on advertisers like you is the second one. If you sell services or B2B, the first one is the more relevant evidence and the invalid-traffic concern is less likely to be your problem.

Stop comparing September to August

The instinct after a forced migration is a 30-day pre/post comparison on the whole campaign. It is the wrong measurement, for a reason that has nothing to do with AI Max: September to October is a seasonality cliff in most verticals, Q4 auction pressure is building, and your own budgets and promotions moved. A blended before-and-after confounds all of it with the migration.

There is a cleaner cut available. AI Max surfaces an "AI optimized" match type label in your search terms and keyword reports. That label marks the queries AI Max matched that your keywords would not have. It splits your traffic into the baseline you already had and the increment AI Max introduced.

Judge the increment. Specifically:

1. Pull the search terms report for the period since your upgrade, segmented by match type, and separate the "AI optimized" rows from the rest.
2. Calculate the increment's own economics. Cost, conversions, CPA or ROAS, for the AI-optimized rows alone. Not blended with the baseline.
3. Compare that against your baseline's economics in the same window — the same weeks, the same auction conditions, the same promotions. This is the comparison that controls for seasonality, because both segments experienced it.
4. Check whether the baseline itself got more expensive. If exact and phrase match CPCs rose while the incremental segment looks fine, you have a budget reallocation problem rather than a matching problem: the incremental traffic is competing with your own keywords for the same budget.
5. Read the increment's query text, not just its metrics. Fifty rows is enough. You are looking for intent drift: informational queries, competitor names, job-seeker language, queries about a product you do not sell.

The decision rule that follows is simple. If the incremental segment converts at an acceptable CPA on its own, AI Max is working and your blended number moved for other reasons. If the increment is expensive and the baseline got more expensive too, the feature is taking budget from traffic that was already working.

The traps in the settings you now have

The migration did not just add matching. It changed settings you may have deliberately chosen, and two of the changes are easy to miss.

The manual CPC override is the one to check first. An advertiser who chose manual bidding usually chose it for a reason — thin conversion volume, a long sales cycle, a distrust of smart bidding on sparse data — and all of those reasons survive the migration even though the setting did not.

Keep or kill, feature by feature

AI Max is not one switch. Treat it as three, and decide each separately against the increment you measured.

The common right answer for lead-generation accounts with purpose-built landing pages is: keep search term matching under close query review, turn off Final URL expansion. Sending paid traffic to a page you did not pick undoes most of the work that makes a landing page for paid traffic convert in the first place.

If you are retail, run the invalid traffic check

Given where the Lunio finding is concentrated, retail advertisers should look specifically for invalid clicks rather than assuming poor performance is a matching problem. On the rates in that study, a retailer spending $10m a year is losing around $500,000 to invalid traffic. Scale that to your spend and decide whether it is worth an afternoon.

1. Compare sessions to clicks. Analytics sessions materially below reported ad clicks, concentrated in the AI-optimized segment, is the first signal.
2. Look at engagement depth on paid landing pages. Single-page sessions at near-zero duration, clustered by time of day or geography that does not match your customer base.
3. Segment by device and network. Invalid traffic tends to concentrate. If one device-and-placement combination carries most of the spend and none of the revenue, you have found it.
4. Use placement and IP exclusions on what you find, and keep the exclusions under review rather than setting them once.

Two framing points. First, invalid traffic is not the same as fraud; it includes bots, crawlers and non-human traffic that no one monetises deliberately. Second, this is the same measurement discipline that catches bot traffic contaminating A/B tests — if you have already built that instrumentation, you can reuse it here.

What the blended numbers will not tell you

One last caution about reporting. Agencies have reported AI Max driving search budget increases of around 15% a year after launch, with CPC rising. Whether that is AI Max causing cost inflation or AI Max finding volume you then chose to fund is not answerable from an account-level chart, and it matters which one it is.

The same problem applies to the platform's conversion figures. If Final URL expansion is sending traffic to pages whose tracking you never checked, your conversion counts and your actual orders will diverge, and the direction of the error flatters the platform. Reconcile platform-reported conversions against your own backend numbers before you conclude anything about performance, for the reasons set out in blended versus platform ROAS. A diagnosis built on the platform's own count of its own success is not a diagnosis.

And if you also run Microsoft Ads, the same feature set arrived there too, so the decisions above are worth making once and applying in both places rather than discovering them twice. We covered that rollout in AI Max on Microsoft Ads, and we run this segmentation as the first step of any performance marketing account audit.

Frequently asked questions

How can I tell what AI Max actually changed in my account?

Use the 'AI optimized' match type label in your search terms and keyword reports. It marks the queries AI Max matched that your keywords would not have, which splits your traffic into the baseline you already had and the increment AI Max added. Calculate the increment's own cost, conversions and CPA or ROAS, then compare it against your baseline's economics over the same weeks. Because both segments experienced the same seasonality and auction conditions, that comparison isolates the feature in a way a 30-day pre/post comparison cannot.

Why shouldn't I just compare the 30 days before and after the upgrade?

Because a blended pre/post comparison confounds the migration with everything else that changed. September to October is a seasonality shift in most verticals, Q4 auction pressure builds through the period, and your own budgets and promotions moved. Any of those can produce the swing you are attributing to AI Max. Segmenting by the 'AI optimized' match type label compares two segments inside the same window instead.

Does AI Max override manual CPC bidding?

Yes. Search term matching does not work with manual CPC, so if a migrated campaign was running manual or enhanced CPC those settings are overridden. This is the first thing to check after an upgrade, because advertisers who chose manual bidding usually did so for reasons that survived the migration even though the setting did not, such as thin conversion volume or a long sales cycle.

How do I stop AI Max sending traffic to pages I didn't choose?

Disable Final URL expansion explicitly. Note that the feature is linked to text customization: turning off text customization automatically disables Final URL expansion, but reviewing the text setting and leaving it on means URL expansion stays on too. Also be aware that while expansion is active, pinned assets are not used when a more relevant URL is chosen, so pinning is not a reliable way to guarantee required copy.

Is AI Max bringing more invalid traffic?

In retail, measured evidence says yes. Lunio's study, published on 12 August 2026 across more than 414 million clicks recorded between October 2025 and June 2026, found invalid traffic on retail search campaigns with AI Max enabled rose from 2.46% in Q4 2025 to 5.28% in Q2 2026, while standard search campaigns fell from 3.72% to 3.07% over the same period. That is 72% more invalid traffic with AI Max than without. The finding is specific to retail, so service and B2B advertisers should not assume it describes their accounts.

Does Google's claimed 14% uplift apply to e-commerce?

No. Google's figure that advertisers typically see 14% more conversions or conversion value at a similar CPA or ROAS comes from internal 2025 data for advertisers outside retail. The invalid traffic finding that cuts the other way is specifically retail. The two headline statistics in this debate describe non-overlapping populations, which is why quoting one at the other settles nothing.

Which AI Max features should I keep?

Decide the three separately. Keep search term matching if the incremental segment converts at an acceptable CPA on its own and the query text shows relevant intent. Turn off text customization if your copy is regulated, legally reviewed or brand-locked. Turn off Final URL expansion if you run dedicated landing pages whose tracking, offer logic or form routing only work on the chosen URL. For lead generation accounts with purpose-built landing pages, keeping matching and disabling URL expansion is usually the right combination.

The takeaway

Measure the increment, not the account. The "AI optimized" match type label is the only clean cut available between the traffic you already had and the traffic AI Max added, and judging that segment against your baseline over the same weeks is the one comparison that controls for a Q4 you cannot hold still. Before you draw any conclusion, check three settings the migration may have changed under you: bidding, because manual CPC is overridden; Final URL expansion, because it is linked to text customization and overrides pinned assets; and brand controls, because they exist and are probably unset. And reconcile platform conversions against your own orders first, because a diagnosis built on the platform's count of its own success is not a diagnosis.