More great news for any marketer using paid search as a core channel, especially ahead of one of the busiest shopping periods of the year.
A new study by Lunio, ‘The Invalid Traffic Impact Report: Retail (2026)’, confirmed what I’d already suspected: fewer “real” users are clicking on ads.
Here’s the breakdown of the study, and what you can (and can’t) do about it.
Ad platforms have been pushing AI into everything
For anyone managing ad campaigns over the past few years, we’ve seen more AI features creeping in, and as a result, less transparency and control.
As I’ve said before, for advertisers with huge budgets and inventory, wasted spend due to click fraud and other forms of invalid traffic (IVT) can be offset more easily than for smaller advertisers, where every click counts.
But ad platforms don’t seem to care. Invalid traffic has been rising, and the fastest growth is in the automated campaign types that the same ad platforms are encouraging us to use.
Google began auto-upgrading legacy Search campaigns (DSA, auto-created assets, campaign-level broad match) to AI Max, just weeks before the busiest time for retailers. While the claim is that this type of campaign will find customers based on broader targeting, like intent, than a standard, exact or phrase match keyword-driven Search campaign would and that advertisers can see improved ROAS, the reality is a huge amount of spend being wasted, and bidding signals being impacted as a result.
What is invalid traffic?
Last month, I wrote about how AI bots are ruining marketing data, and paid search is no exception.
Invalid traffic is made up of bots, scrapers, accidental taps, competitor clicks and made-for-advertising (MFA) inventory: basically any click with no chance of converting.
This issue has been compounded by AI, which can scale convincing fake sessions mimicking ‘real’ user behaviour. And ad platforms don’t really seem to have an answer when it comes to stopping it.
The real cost is bigger than wasted clicks
The study looked at 414m+ clicks across 88 retail accounts, from Q4 2025 to Q2 2026. The findings are pretty dire, revealing that a $5m annual budget at a $3.70 CPC and a 5% IVT rate means about $250k of wasted spend, or about $750k of lost revenue at 3:1 ROAS.
But that’s not all! The bigger issue is signal pollution. Again, we’ve been forced to shift away from manual bid strategies to AI-powered, automated ones, but smart bidding now learns from fake sessions and inflated conversions, while remarketing and lookalike pools fill with bots, and micro-conversions and intent signals tracked through the funnel, such as button clicks, form fills (without submitting), and add-to-cart without checking out, all distorting the data.
I saw this happen last month, with 1,000s of key events firing for a click on a specific button, which is being tracked as a conversion, as it shows high intent.
Session data in GA4 showed that all of those events came from Google Ads. And that button click is a conversion (with a low conversion value attributed to it), so this month, I’ve seen fewer actual conversions, and signs that it’s now deciding to optimise campaigns for that instead, as the campaign is set to bid based on conversion value. Great.
Google is the ad platform most of my clients invest the big bucks in, so for me personally, I’ve noticed costs go up, and conversions go down for several campaigns I manage. Any time I’ve tested AI Max campaigns, they’ve been a complete disaster.
Although Google has an average IVT of 4.75% (which is lower than some other platforms), it carries 77% of volume and rose 37%, so it accounts for the most wasted money.
The IVT varies by campaign type, too. You might expect Shopping, which is powered by the Merchant Centre, to be a bit lower, but in fact, it’s the riskiest: 6.33% average and an 80% rise, reaching about 1 in 13 clicks by Q2. One example in the report showed a branded Shopping campaign at 19.85% IVT.
For the same period, Display more than tripled (2.45% to 8.13%), Demand Gen more than doubled, and even Search, the ‘safe’ channel, rose by 53%. Oof.
Performance Max, on the surface, looks like it fared better at 4.43%, until you realise it’s a blended average across advertising channels. As it makes up 47% of Google volume, it produces more invalid clicks than Shopping and Display combined.
I noticed a huge shift to Display on one of my Performance Max campaigns, so I was a bit baffled when I saw the CTR go up. That is, until I looked into the placement report and saw thousands of made-for-advertising (MFA) sites generating huge numbers of impressions (and likely, most of the bot clicks, not that Google shows you that data).
When the report dug into the top offenders for IVT, casual game apps and free VPNs (worst: Rosa VPN at 90.8% IVT), plus a templated MFA network of SaaS/CRM-themed domains in a 52% to 57% band.
If you look at your placement report, I guarantee that you will see a load of casual game app sites that look like dog eggs that no human in their right mind would ever visit.
The AI Max effect
Now, Google says advertisers using the full AI Max suite see about 7% more conversions at a similar cost per acquisition, with an important caveat – this data is based on its internal data for ‘non-retail advertisers’.
The study focuses on retail, and doesn’t provide data on conversions to see if there was any uplift from shifting budgets to AI Max. However, AI Max grew to 55% of Search clicks but 68% of invalid Search clicks. That’s a lot of wasted spend.
This perfectly demonstrates why broad and keywordless matching plus final URL expansion doesn’t work for most advertisers.
It’s not just Google, either
Think it’s worth shifting budget away from Google to other ad platforms, then? Wrong. Meta has the highest average at 5.99%. Advantage+ offers the same trade-off as PMax: more reach, less visibility. TikTok is the fastest-rising for IVT, up 68% (3.94% to 6.61%).
Microsoft Ads is the only platform improving (4.70% to 4.18%). But who even uses Bing as a search engine anymore lol.
Can’t smart bidding just sort this out?
Well, it can, sort of. Maximise Conversion Value is the best built-in defence for e-commerce businesses, or B2B/lead gen businesses that have lots of data on customer lifetime value, at least. It helps lower bids on traffic that never generates revenue, and it works better in retail than in lead gen because completing the checkout process, and the revenue generated by that sale is much harder for a bot to fake. But while smart bidding reduces the cost of invalid traffic, it doesn’t remove it completely.
Why it isn’t enough:
- It learns after you’ve paid. There’s always new territory it hasn’t priced in yet: new made-for-advertising (MFA) domains, new apps, new AI Max queries.
- It’s only as clean as your conversion setup. If add-to-carts or progress through a multi-step sign-up form count as primary conversions, bots are training your bidding.
- It works on segments, not individual clicks. It lowers bids but doesn’t block a known bad user.
- The data shows it. Most retail Shopping and PMax campaigns already run on value bidding, and Shopping IVT still rose 80%.
- Peak season loosens the guardrails. Bigger budgets and relaxed ROAS targets mean more exploration just as bot activity peaks, and AI Max is designed to widen reach.
- It protects spend, not data. Fake visitors still end up in remarketing lists, GA4 and audience signals.
So, what can you do about it?
Ultimately, ad platforms won’t fix a problem they profit from, least of all in Q4 when it’s bumper revenue time, baybay. And to be honest, unless people start significantly cutting paid search budgets (which they won’t because both social media is algorithm-driven and organic search is basically all bloody AI Mode and AIOs), they can get away with it.
A few things I would do before we get too far into Q4, if this is a busy time for your business:
- Check whether campaigns are being auto-upgraded to AI Max. Review search terms and final URL expansion settings, and decide deliberately rather than letting Google do it for you.
- Audit Performance Max channel-level reporting. Add account-level placement exclusions for apps and MFA domains, and check the search term report to add negative keywords. Do it as often as you can!
- Segment IVT by campaign, not just account.
- Make purchases or form fills the only primary conversion. If you can, move micro-conversions to secondary, even if using a bid strategy based on conversion value. Unfortunately, this isn’t always realistic for accounts with low volume, high-cost conversions, as limited bidding signals can mean campaigns perform worse. Fun!
- Use enhanced conversions, ideally with server-side tracking, so conversion data is harder to spoof.
- Import refunds, cancellations and fraud-flagged orders so the model learns from real value.
- Review ROAS targets before peak rather than loosening them by default.
- Sanity-check conversion data feeding smart bidding: add-to-cart vs purchase ratios, bounce rates, device, and geography.
- Set peak-season monitoring thresholds before budgets increase.
- Consider independent click verification. Lunio (the author of the study) is offering a free audit.