← all editions

Readiness Radar #10 · week of 2026-09-07

GEO Wars: Manipulation, Defense, and the Race for AI Search Visibility

This week's research zeroes in on Generative Engine Optimization (GEO), the practice of shaping content so AI search tools cite it. New studies show both how GEO can be weaponized and how defenders are racing to catch up, alongside fresh data on what buyers actually ask AI engines and how AI agents cope with real websites.

New defense system aims to stop AI search manipulation via rewritten content

Researchers describe malicious Generative Engine Optimization (GEO), where documents are rewritten to trick AI search engines into citing them, even though the rewritten pages still contain the same facts as the originals. They introduce GEO Defender, a two stage system designed to catch these manipulative rewrites without needing to retrain the underlying AI models.

What this means for your website: Expect AI search platforms to add new filters that could affect how any GEO-optimized content, not just malicious content, gets treated.

Source: arxiv.org (arXiv) →

Study finds current AI safety filters barely catch GEO-based misinformation

Researchers created Counter-GEO-Bench, a benchmark of over 200 verified queries paired with normal and information-distorting GEO rewrites, to test how well existing safety tools catch manipulated content fed into AI search systems. According to the study, three widely used defense tools reduced successful attacks by only a modest amount.

What this means for your website: Don't assume AI platforms can reliably filter out manipulated or misleading rewritten content that competitors might publish about your industry.

Source: arxiv.org (arXiv) →

New AI system auto-tests nine content tactics to win AI search citations

Researchers built Agent2UCB, a system that automatically tests nine different content optimization strategies for each piece of content and learns, using online feedback, which one actually gets it cited or summarized by generative search engines like Google AI Overviews and Perplexity. The system also checks that these changes don't hurt traditional SEO factors such as readability, topic coverage, and credibility signals, according to the researchers.

What this means for your website: Businesses may soon compete against automated, self-improving optimization tools, so ongoing content monitoring will matter more than one-time SEO fixes.

Source: arxiv.org (arXiv) →

New million-persona dataset maps how buyers ask AI engines

Researchers built PersonaGen-1M, a synthetic dataset modeling more than a million buyer personas across 511 industries, capturing the kinds of search queries people might ask AI engines like ChatGPT, Gemini, and Perplexity while researching purchases. According to the study, 78.3% of these queries are informational, while a much smaller share are commercial or transactional.

What this means for your website: Prioritize answering broad informational questions in your content, since that's reportedly how most AI search buyer queries are framed, not just direct sales pitches.

Source: arxiv.org (arXiv) →

Study: repeated AI optimization makes online content more alike

Researchers built a simulation called CHASE to study what happens when many content creators repeatedly rewrite pages to rank better with an AI ranking signal. Across 20 rounds of simulated optimization in six different topic areas, the study found that the link between content quality and ranking position consistently weakened.

What this means for your website: Chasing AI ranking signals too aggressively could make your content blend in with competitors instead of standing out on quality.

Source: arxiv.org (arXiv) →

AI web agents can now learn to navigate messy real-world websites on the fly

Researchers found that AI agents trained on simple, clean test environments often struggle when they encounter complex real websites. They propose a method allowing agents to break down complicated pages into smaller parts and adapt their behavior during actual use, without needing extra labeled training data.

What this means for your website: Keep your website's structure and page layout as clear and modular as possible, since AI agents reportedly perform better when they can break pages into simple, understandable parts.

Source: arxiv.org (arXiv) →

Weekly · 100% free

AI search: all you need to know, every Monday

One short email a week on the trends shaping AI search and what they mean for your website. Plain language, no jargon, always free. Read every edition here on the site, follow along on LinkedIn, or get it straight to your inbox.

One short email a week. No spam, unsubscribe anytime.

All sources this week

Is your site ready for AI agents?

Run the free Siteiz scan to see exactly what AI crawlers can (and can't) read on your site.

Run the free scan

What changed this week in how AI agents read the web, and what it means for your site.