AI seems to be infiltrating everything — advertising, publishing, content consumption — across the digital ecosystem.

‍

So it’s right to ask: Do adblockers use AI? And if they do, is AI going to revolutionize adblocking?

‍

AI is already part of the adblocking landscape. Major players like AdGuard, eyeo (makers of Adblock Plus), and Brave have all been testing AI-based approaches. It’s no longer theoretical.

‍

In this article, we’ll look at how adblockers traditionally work, how AI is starting to enhance them, and where the technology is heading next.

‍

💡 Definition note. In this article, “adblocker” refers to tools primarily designed to detect and block advertising — whether delivered as browser extensions, built-in browser features, or network-level filters. It doesn’t include LLM bots or AI assistants that bypass ads as a side effect of how they operate.

‍

How Adblockers Work Today

At their core, adblockers do one thing: identify what counts as an ad and stop it from reaching the user. That process depends on filter lists — human‑curated rule sets that tell the blocker what to intercept or hide.

‍

These rules fall into three main categories:

1. Network rules block ad requests before they hit the browser, stopping calls to known ad servers
2. Cosmetic rules work by telling the browser which parts of a webpage to hide, so ads never appear on screen
3. Scriptlets are small JavaScript snippets that disable specific behaviors like anti-adblock detection scripts (e.g. “turn off your adblocker” walls)

‍

Browser‑based adblockers — including Brave, Comet, and extensions like Adblock Plus, AdGuard, and Ghostery — can use all three types. Network‑level adblockers, such as those deployed in IT‑managed environments, typically rely only on network rules.

‍

Importantly, adblockers don’t need AI to be effective. Most web advertising still runs through third‑party domains, making it relatively easy to block. In fact, adblockers deliver near‑flawless results for everyday users. The exceptions are niche cases — like first‑party ads served from a publisher’s own domain on lower‑traffic sites — which may slip through simply because they go unreported.

‍

AI Experiments So Far

For years, the adblocking community has explored whether AI could automate ad detection in real time — the “holy grail” of adblocking. In theory, this would improve accuracy and cut down the endless manual work of maintaining filter lists.

‍

Early efforts used machine learning models trained to recognize patterns typical of ad content. eyeo's Project Moonshot (2021-2022), for instance, analyzed page structure and network behavior using existing filter lists as training data. Results were promising, but the performance overhead — CPU, memory, battery — made it impractical outside of a lab setting.

‍

Brave pursued a different strategy: mapping relationships between page elements — how scripts called resources, how those resources linked to others, and which interactions triggered network requests. The browser could then classify whether a resource was likely an ad based on that “relationship map.” It was technically elegant but fragile. Each major browser update risked breaking the model, requiring constant engineering support. In time, the maintenance load outweighed the benefits.

‍

They weren't alone — other companies and research groups ran similar experiments, arriving at the same conclusion: these systems didn't eliminate human maintenance, they shifted it from list curation to model tuning.

‍

The Latest Wave: LLM‑Driven Detection

Large language models (LLMs) opened a new avenue: understanding what content means, not just recognizing patterns in code or layout.

‍

AdGuard has prototyped three LLM-based approaches:

• Embedding-based classification turns page content into numeric representations that can be compared to known ad patterns. It’s fast and cacheable, but not precise — accuracy varies with language, layout, and ad format, making it hard to set a consistent threshold.
• Prompt-based analysis asks an LLM directly whether a piece of content is an ad, much like you'd ask a person. It's more accurate, but slower and costlier.
• Vision-based detection sends a screenshot of the page to an AI model and lets it visually identify ads the way a human would. It achieves the highest accuracy, but at 10–15 seconds of latency — far too slow for production.

‍

All three share the same obstacle: they’re resource‑intensive. Running these models inside browser extensions consumes too much CPU, memory, and battery. And browser vendors are tightening the rules — Chrome’s Manifest V3 update now limits extensions to lightweight, declarative filtering, leaving little room for real‑time AI inference.

‍

Chrome’s experimental Prompt API hints at future browser‑native LLM features. Still, even advanced methods depend on filter lists to pre‑identify which elements to analyze.

‍

What about network‑level adblockers? They aren’t bound by browser API restrictions, but they hit a different limit: visibility. They typically only see domains and metadata, not content or layout. That makes embedding, prompt, and vision models ineffective in this environment — keeping network‑level blocking largely filter‑driven.

‍

Plus, at the scale adblockers need to operate, large‑scale inference simply isn’t economical. Adblock Plus alone claims to have blocked 200 trillion ads. If each prompt cost half a cent, that works out to $1 trillion in compute expenditure.

‍

Until costs come down, it’s simply not viable.

‍

Where AI Is Making An Impact

‍Here’s the real insight: AI won’t replace filter lists anytime soon — they’re too efficient, lightweight, and deeply integrated into browser architecture.

‍

The real progress lies in using AI to generate and maintain those filter lists faster and more precisely than humans can. It’s not about replacing human curators — it’s about augmenting them.

‍

Today, filter updates are still community‑driven: volunteers spot new ad patterns, write blocking rules, and submit them for review. Increasingly, AI is taking on parts of this workflow — identifying emerging ad styles, drafting new filters, and deploying updates automatically.

‍

This shift is already visible in three key areas:

1. Speed: Adblockers reacting to new ad placements within hours instead of days
2. Accuracy: AI filtering ads that human curators have overlooked
3. Breadth: Coverage extending deeper into the long tail of smaller, less‑visited sites

‍

What This Means For Publishers

As AI accelerates filter generation and maintenance, browser-based adblockers won't just get better at blocking traditionally served ads — they'll push more aggressively into areas they've been only semi-effective in to date.

‍

Expect these not to survive the AI era:

• First-party ads
• “turn off your adblocker” walls
• Cookie consent pop ups
• Email sign up forms

‍

AI dramatically shortens the lifecycle from detection to blocking. Patterns that were once missed entirely — or took days or weeks to filter — can now be identified, classified, and disabled in hours.