Lunio finds 5.28% invalid traffic on AI Max retail search

Ad fraud vendor Lunio published an analysis on 12 August 2026, reported by PPC Land, showing that retail search campaigns with Google's AI Max enabled recorded an invalid traffic rate of 5.28% in the second quarter of 2026, against 3.07% for standard search campaigns. That gap works out at roughly 72% more invalid traffic on the AI Max side. The two rates moved in opposite directions over the study period: AI Max invalid traffic climbed from 2.46% in the fourth quarter of 2025, while standard search fell from 3.72%.

What Lunio measured, and how

Lunio recorded 414 million paid clicks between October 2025 and June 2026 across Google Ads, Bing, LinkedIn, Meta and a group of native and social platforms. The sample ran in monitor-only mode, so nothing was blocked and no traffic was filtered out during the window. The rates describe what landed on advertiser sites after Google's own invalid click filtering had already run, not what a blocking product would have caught.

Invalid traffic in this context means clicks with no chance of converting: bots, scrapers, automated tooling, click farms, and repeat clicks from a single source. Google filters some of this itself and credits it back, which is why a third party figure and a Google figure count different populations and will never reconcile against each other. The Lunio number sits on top of whatever Google had already removed.

One structural point about the sample is worth holding on to. Monitor-only means the advertisers in the dataset paid for every click counted, including the invalid ones. A blocking deployment would have changed the traffic mix mid-study and made the quarterly figures incomparable, so the choice keeps the trend readable. It also means the analysis cannot say how much of that 5.28% would survive a filtering layer, or how much of it Google would have credited back on review.

The channel numbers

Lunio broke the retail sample down by platform and by campaign type. These are the figures carried in the PPC Land report of 12 August 2026.

The average across everything Lunio measured in retail was 5%. Google Shopping was the worst single channel in the set, and its second quarter peak of 7.51% sat above any quarterly AI Max figure. Search is not the only place a retail budget leaks and it is not even the leakiest. What makes the AI Max line stand out is direction rather than absolute level: it is the one line in the set that got worse while its closest comparison got better.

The money Lunio attached to the rates

Lunio modelled a retailer spending 10 million US dollars a year at an average cost per click of 3.70 dollars. At the 5% retail average, that advertiser hands over roughly 500,000 dollars a year for clicks that were never going to convert. Applying a 3 to 1 return on ad spend assumption, Lunio put the revenue that never arrived at about 1.25 million dollars. Nick Morley, the company's chief executive, tied the exposure to automated media buying and said retailers are particularly vulnerable to the risks and costs associated with invalid traffic.

The model is arithmetic on an average rather than a measurement of any real retailer, and Lunio presented it that way. Its value is as a scaling rule. At a 5% rate, invalid traffic costs 5% of media spend, and the revenue consequence is that figure multiplied by whatever return on ad spend the account actually earns. An account earning 6 to 1 loses twice the revenue of an account earning 3 to 1 from the same wasted clicks.

Where AI Max came from, and why 1 September matters

AI Max for Search entered beta on 6 May 2025 as a bundle of controls layered onto standard search campaigns, widening how Google matches queries and generating assets on the advertiser's behalf. From 1 September 2026, Google auto-upgrades campaign-level broad match and automatically created assets to AI Max. Advertisers who never chose the product end up inside it unless they change those settings first.

That is the reason to read the Lunio figures in August rather than in December. The dataset covers a period when AI Max adoption was voluntary, so every account in it opted in. After 1 September the population changes. Accounts running broad match for their own reasons get moved across without ever having asked for the matching behaviour that comes with the upgrade, and the invalid traffic in retail search that Lunio measured on volunteers becomes a question for everyone else.

What the analysis does not establish

Lunio reported a correlation and did not claim a cause. Four gaps are worth stating plainly, because a headline number of 72% invites conclusions the evidence does not carry.

• The analysis does not establish that AI Max causes higher invalid traffic. Campaigns that adopted AI Max may differ from campaigns that did not in budget, category, geography or account maturity, and none of those variables were controlled for in the reporting.
• It does not explain the divergence. Standard search rates improved across the same quarters, and nothing in the coverage accounts for why one line fell while the other rose.
• The data stops in June 2026, so fourth quarter peak trading, the exact window Morley points at, is not in the sample at all.
• It does not test whether the pattern holds after June 2026. Three quarters is a short series for a product that was still changing throughout the measurement period.

The coverage also carried no country breakdown for the retail sample, so the figures read as a global retail aggregate and cannot be assumed to describe any single market.

What this means for Thai marketers

This section is analysis, not something Lunio or PPC Land said. The 1 September auto-upgrade is a global change with no separate Thai opt-out, so Thai retail and e-commerce accounts running campaign-level broad match are on the same clock as accounts everywhere else. Click quality monitoring is uncommon in Thai accounts, which tend to be judged on cost per conversion and blended return, so a shift of one or two percentage points in invalid traffic would surface as a slightly worse month rather than as a traffic quality problem with a name attached to it.

The arithmetic scales down without drama. An account spending 500,000 baht a month at a 5% rate is paying around 25,000 baht a month for clicks that cannot convert, and at a 3 to 1 return that is roughly 75,000 baht of revenue that does not appear. Those are not catastrophic numbers, and that is the difficulty: they sit comfortably inside normal monthly variance and never trigger a review. The only way to separate a quality problem from a demand problem is to have measured quality before the change.

Thai retail also skews toward Shopping and toward Meta, both of which carried higher average rates than search in the Lunio set. An advertiser who reads this story as a search story alone is looking at the smaller of the two exposures. The Google Ads side is where the 1 September deadline bites, but the e-commerce feed channels are where the higher rates were measured.

What to check in your own account before 1 September

1. Find the switches. AI Max is enabled inside the settings screen of an individual search campaign, alongside the search term matching controls, and automatically created assets sit with the campaign asset settings. Those two settings are what Google says it will upgrade on 1 September, so write down which campaigns currently have either of them on.
2. Baseline performance first. Export clicks, cost, conversions, conversion rate and cost per conversion by campaign and by month for June, July and August 2026. After 1 September you want a comparison that predates the change rather than one built from memory.
3. Baseline quality separately. Performance and quality are different measurements, and cost per conversion hides the difference. In GA4, pull sessions, engaged sessions, engagement rate and average engagement time for paid search traffic by campaign for the same three months. A click quality benchmark is a before and after on engagement rate and conversions per thousand clicks, not a fraud score.
4. Save the search terms report. Export search terms with clicks, cost and conversions for August before the upgrade date. Query expansion is the mechanism most likely to change on 1 September, and the only way to see the change is to hold the old list.
5. Decide in advance what counts as a problem. Pick a threshold now, for example engagement rate falling more than five points or conversions per thousand clicks falling more than fifteen percent, and agree it before the data exists. Thresholds chosen after the fact get argued about.
6. Annotate 1 September in whatever report the client or the management team reads, so a September step change is read as a product change first and a performance change second.

Questions marketers are asking

Should I turn AI Max off because of this report?

No, not on the strength of this report alone. Lunio measured a correlation between AI Max and higher invalid traffic in retail search and explicitly did not establish a cause, so switching off a matching product on that basis trades a measured risk for an unmeasured one. The defensible response is to baseline your own account before 1 September and decide from your own numbers.

Is the AI Max invalid traffic rate higher in Thailand?

The source did not say. Neither Lunio's analysis as reported nor the PPC Land coverage broke the retail sample down by country, so there is no published Thai figure to compare against. Treat 5.28% as a global retail aggregate and measure your own account rather than assuming the average applies locally.

Can I opt out of the 1 September auto-upgrade?

Google's stated mechanism is that campaign-level broad match and automatically created assets are upgraded to AI Max on 1 September 2026, which means the settings themselves are the lever. Nothing in the reporting described a separate opt-out toggle or a regional exemption, so an advertiser who does not want the upgrade needs to look at those two settings before the date.

Does Google refund invalid clicks?

Google runs its own invalid click detection and credits filtered clicks back, which is why Google's reported figures and a third party vendor's figures never match. The Lunio rates are what its detection flagged after that filtering, so the two systems are counting different things and neither one is a check on the other.

Do these numbers apply to lead generation accounts too?

The sample discussed here is the retail cut, so the 5% average and the 5.28% AI Max figure describe retail advertisers. The wider dataset covered Google Ads, Bing, LinkedIn, Meta and native platforms, but the coverage did not publish an equivalent breakdown for other verticals, so a lead generation account has no benchmark here and should measure its own.

If you are running retail search or Shopping in Thailand and have never taken a click quality baseline, the three weeks before 1 September are the last quiet window to take one. Our team is happy to walk through what to export and what to watch after the upgrade lands.