When Mondelez discovered that Oreo was appearing in about 10% of AI chatbot cookie recommendations, the company faced a problem most digital marketers haven't put a name to yet. The brand was strong. The product was everywhere. But AI couldn't cite what it couldn't read — and Mondelez had blocked the crawlers.

After a year-long rebuild, Oreo's citation rate for cookie-specific queries climbed to around 70%. Andrew Lederman, Mondelez's VP of Global Digital Commerce, laid out the strategy on a recent Digiday Podcast. The story is worth understanding in detail — not as a case study about Mondelez, but as a blueprint for every brand whose customers now use AI to decide what to buy. This is the discipline called Generative Engine Optimization, and it applies at every scale.

What is Generative Engine Optimization (GEO)?

Generative Engine Optimization (GEO) is the practice of structuring website content to be cited by AI answer engines including ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews. It is distinct from SEO. Search engine optimization produces pages designed to rank in results. GEO structures content so AI retrieval systems can extract it, verify the source, and place it directly into a generated answer.

The distinction matters because AI systems don't rank pages. They retrieve passages. A page can hold the top organic position on Google and still be nearly uncitable by AI — because the requirements are different. AI retrieval favors short, declarative sentences, consistent entity information across the web, and content structured so each section is independently extractable as a clean answer. Most brand content is written for human readers and fails these requirements entirely.

When Mondelez's retail partners flagged that they expected 30% of site traffic to be agentic by 2028, Lederman described the response: "We actually had to fully take a different look at our brand.com experiences and make sure that they're optimized towards AI. This is not something we had previously done." That statement describes GEO. It's a discipline every brand with an online presence needs now.

Why was Oreo only cited in 10% of AI chatbot recommendations?

Mondelez had blocked AI bot crawlers from its brand sites — a decision made to protect intellectual property while the company developed its AI strategy. It was reasonable at the time. AI couldn't read their content, so it didn't cite their brand. AI doesn't match keywords — it matches meaning. And with crawlers blocked, there was no meaning to match.

This failure mode is more common than most marketing teams realize. Default security configurations, Cloudflare bot protection settings, and cautious robots.txt files regularly block AI crawlers without anyone making a deliberate decision to do so. The crawlers that need access — GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and others — are not always on the default allowlist.

For Oreo, the fix was the starting point, not the solution. Unblocking the crawlers was Lederman's "only the first step in a three-part rebuild." Citation requires more than crawlability. It requires content structured for extraction and a brand identity AI systems can verify and trust.

What did Mondelez do to improve Oreo's AI citation rate from 10% to 70%?

Mondelez unblocked AI crawlers, structured consistent brand information across all online sources, and rewrote content for AI extraction rather than search ranking. Lederman described this as three pillars that had to work together — and each one maps directly to a core GEO requirement.

Consistent, structured product knowledge everywhere AI looks. AI systems cross-reference brand information across dozens of external sources before deciding how confidently to cite a brand. A company that describes itself differently on its homepage, its review platform profiles, and its press releases isn't a coherent entity to an AI system. It's a reliability signal pointing away from citation. Research analyzing 250M+ AI responses found that entity richness — specific, consistent brand information spread across the web — produces a 267% lift in citation rates.

AI-native content at scale. The second pillar was content written to be cited, not just ranked. Sentence-level research on 42,971 AI citations found that 92% of cited sentences fall between 6 and 20 words. Short, declarative, specific — that's the extraction pattern. Content full of hedged qualifiers and compound sentences doesn't extract cleanly into AI answers regardless of how well it ranks.

Measurement built around citation, not just traffic. Lederman called this the most important piece — treating AI visibility as a performance channel with KPIs built around citation rate, share of voice, and sentiment. Most brands can't see this clearly yet because over 70% of AI-referred traffic appears as Direct in standard analytics. The data infrastructure has to be fixed before any of the other work is measurable.

Which AI crawlers should brands allow access to their websites?

Brands should explicitly allow GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Bingbot, OAI-SearchBot, ChatGPT-User, and Googlebot in their robots.txt configuration. These are the primary user agents used by the AI platforms that answer consumer and business queries at scale. Blocking any of them removes your content from the AI knowledge base for that platform.

Beyond robots.txt, brands should check HTTP response headers for X-Robots-Tag directives and verify that WAF configurations and CDN bot protection rules aren't blocking these agents at the network layer. A clean robots.txt file doesn't help if Cloudflare's bot fight mode is blocking the request upstream.

Lederman put the baseline plainly: "if they can't crawl our site, nothing else matters." All GEO work — content optimization, entity consistency, schema markup — depends on AI systems being able to read the content first.

How do brands measure AI search visibility?

AI search visibility is measured through citation rate, share of voice across AI platforms, and analytics corrected to properly capture AI-referred traffic. Citation rate tracks how often your brand appears when AI systems answer queries in your category. Share of voice compares your citation frequency against competitors across specific query types. Corrected analytics separate AI-referred sessions from Direct traffic, which is where they currently appear by default.

The measurement correction is the prerequisite. Research by Seer Interactive found that over 70% of AI-referred traffic is misclassified as Direct in standard Google Analytics configurations. Google AI Mode strips referrer data entirely. ChatGPT does the same for paid accounts. Without correcting for this, brands are optimizing a channel they can't see.

Lederman framed Mondelez's measurement approach as treating AI citation like a performance marketing channel: "Our first move was to make sure that we had a measurement system in place, so that we can treat it like a performance marketing channel." That sequencing is right. Measurement first, then optimization. MeetGEO's Citation Authority agent tracks citation rate and share of voice across ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews on a weekly basis.

What is agentic commerce and why does it matter for digital marketing?

Agentic commerce is AI-driven purchasing where autonomous agents research, compare, and recommend products on behalf of consumers without traditional search. Instead of a consumer typing a query and clicking through results, an AI agent handles the research — reading reviews, comparing products, and returning a recommendation directly.

Mondelez's retail partners are planning for 30% of their site traffic to be agentic by 2028. Lederman's response: "There's not too much question around 'will it happen?' The question just comes up of when it will happen." For brands whose customers are already using AI for purchase research, that shift is already underway. A brand that isn't cited in AI answers isn't considered in agentic purchase decisions — regardless of how well its pages rank.

AI citation presence compounds over time. Brands that build it now become more associated with their categories in AI model understanding, making them progressively more likely to be cited. Brands that wait face a compounding disadvantage that grows harder to close as the gap accumulates.

What every digital marketer should do next

Mondelez spent a year on their GEO rebuild with a dedicated team and a $36 billion consumer goods operation behind the investment. Most digital marketing teams don't have that runway. The three things Lederman described — entity consistency, AI-native content, and citation measurement — aren't complicated in concept, but executing them systematically across hundreds of pages requires a different approach than most teams currently have.

The starting point is the same as Mondelez's: confirm your AI crawlers are unblocked, check your brand information for consistency across every platform where AI looks, and fix your analytics so you can see AI-referred traffic. None of that requires a year. It requires knowing what to look for.

The brands that treat AI citation as a performance channel now are the ones that will be in the answer when someone asks an AI what to buy in their category. See how MeetGEO automates GEO for WordPress and Shopify brands →