AI Ad Fraud and Fake Leads: The Attacks On Your Marketing

AI cuts both ways: your bidder can't spot a toaster

The framing that carries everything else is simple: marketers get good results using AI, and "there are some fraudsters around there using AI also to speed up their work." The sharpest line in the whole argument is also its most quotable — the AI buying your media "cannot distinguish between a human and a toaster." Automated bidding optimises toward whatever looks like engagement, and a sophisticated bot looks exactly like engagement.

The AI buying your media cannot distinguish between a human and a toaster.

The 2026 data has caught up with — and overtaken — that warning. On 3 June 2026, Cloudflare leadership said bots had, for the first time, passed humans in share of web-page requests on its network: roughly 57% bot vs 43% human, a crossover Cloudflare had not expected until late 2027. And the bots are not crude anymore. Akamai's research puts 63.1% of bad-bot traffic at "advanced" evasion — randomised device fingerprints, headless browsers, human-mimicking mouse movement and keystroke timing. The toaster has learned to hesitate.

Ad fraud: the silent tax

Ad fraud is a term most marketers have heard but few have measured. Lunio's Global Invalid Traffic Report 2026 — built from more than 2.7 billion paid ad clicks across Google, Meta, TikTok, LinkedIn and Bing — put invalid traffic at 8.51% of all paid ad traffic, roughly one wasted click in twelve, translating to an estimated $63 billion burned in 2025 alone. The rate is wildly uneven by platform: TikTok at 24.2%, LinkedIn at 19.88%, Meta at 8.2% and Google at 7.57%. Fraudlogix's independent analysis of 105.7 billion impressions found a global invalid rate of 20.64%. The two reports measure different inventory, which is why the honest read is a range — from ~8.5% of clean paid-search clicks to north of 20% of open-programmatic impressions — not a single number.

Performance Max is the clearest mechanism to walk through. Hand Google a budget and it splits spend across Search, Display and YouTube so "you don't have to worry" about allocation. Over time the optimiser notices one channel looks like it's performing and shifts more money there — until "you don't see your budget anymore... but you also don't see the conversions." The reason: "the bots cannot afford your product. They don't have credit cards." The black box chased cheap, bot-friendly clicks, hit its engagement signals, and reported success while real revenue flatlined.

The fix is unglamorous: download your impression and placement report and check whether the domains match the brand you sell. Google's 2026 account-level placement exclusions — now unified across PMax, Demand Gen, Display and YouTube — finally give you a lever to pull on what this walkthrough used to describe as a sealed box. The payoff has been proven before: in 2017, JPMorgan Chase found that of roughly 400,000 sites carrying its ads, only about 5,000 drove activity beyond an impression — and cutting to that vetted list produced no deterioration in reach or cost. Most inventory your budget touches is doing nothing, and cutting it returns performance rather than costing it.

Fake leads: the war inside the company

Fake leads hit three different people, and treating them as one problem is the mistake. Sales wastes hours qualifying junk — hand a rep 10,000 leads of which two are real, and you've weaponised the funnel against the people meant to work it. Imperva's 2026 Bad Bot Report puts automated traffic at a record 53% of all web traffic, bad bots alone at 40%, with a 12.5-fold rise in AI-driven attacks in a single year. Then comes the political layer: marketing and sales blame each other over lead quality. And the cost the boss never sees is opportunity cost — while reps sift fakes, they aren't preparing the calls that actually close business.

The fix most tools get wrong is blocking. Blocking bots from submitting forms helps sales, but it destroys the diagnostic data marketing needs — you never learn which form is under attack or which channel is leaking. The alternative is to filter, not block: let every submission into the CRM, but tag each valid or invalid, so sales works the real leads while marketing keeps the full picture.

If you put a bucket there, it's kind of blocking the leads… but you don't have any data. So you don't know how to fix it.

The payoff compounds downstream. Guidable, a Japanese job platform, was bleeding budget because fraudulent form leads were poisoning its bidding algorithm's training data — the optimiser was learning from fakes. After filtering invalid submissions out of the data feeding the campaigns, the models renormalised on real humans: CPC fell ~85% and ROI rose ~152% within roughly two months. The cost of fake leads isn't only wasted qualification hours — it's the silent corruption of every automated decision downstream, because the black box that can't tell a human from a toaster is also the one being trained on the toasters.

The audit nobody has time to run

The sharpest argument here challenges the industry's own premise. Everyone is busy learning what AI can do for marketing — so if AI is so great at optimisation, surely that freed up five hours a week to think? "It's on top. It's not reducing. It's putting more effort on you." Marketing Week's 2026 Career & Salary Survey found 65.3% of marketers felt overwhelmed over the prior 12 months and 55.1% felt emotionally exhausted. Gartner's 2025 survey found marketers actively use just 49% of their stack's capability — a recovery from a 33% low in 2023, not a solved problem.

Because nobody has time, the basics rot. Roughly 20% of the websites one practitioner scans lack a sitemap XML — teams insist they have one until the report says otherwise. Security, in this expansive sense, is "the silent things in the back end of our marketing operations" that nobody owns: ad fraud, fake leads, site performance, header structure, GEO signals. Free scanners now exist for exactly this gap — the point isn't the vendor, it's that almost nobody is checking the leaks AI is quietly making worse, faster and cheaper.

The EcomExpo playbook

Six moves that convert this into an operating discipline for any store or agency:

• Treat fraud as a line item, not a footnote. Build a fraud-loss assumption into every media plan, weighted by channel — TikTok and open programmatic run dirtiest.
• Run the placement audit on every PMax and Display account, then act on it. Download the report, check the domains, use the 2026 exclusion controls to cut junk sites.
• Filter, don't block, on lead forms. Tag validity instead of blocking submissions, so sales gets clean leads and marketing keeps the data that trains your bidding algorithm.
• Sell fake-lead protection as an alignment fix, not just a tech fix — it resolves the sales-versus-marketing feud, not only the CRM.
• Audit the basics you've delegated. Sitemap, header structure, SSL/DNS, Core Web Vitals — the leaks nobody has time to check are exactly where AI is making things worse.
• Reset expectations on AI and time. It raises output and pressure before it saves anyone an hour — plan the audit into the calendar rather than waiting for free time that never arrives.

The thread tying it together is a temperament as much as a tactic: AI is genuinely powering your marketing, and it is powering everything that preys on it with the same step-change. The marketers who win the next phase will be the ones who put a guard on the door, a filter on the form and a doctor on the calendar — while everyone else automates their way into a bigger, faster, more expensive blind spot.