Reading Into Microsoft’s AEO & GEO Guide

On January 6, Microsoft published a playbook titled “From Discovery to Influence: A Guide to AEO and GEO” to explain how to influence brand mentions from AI. Microsoft’s guide is targeting eCommerce as it specifically mentioned

Interestingly, Google announced the Universal Commerce Protocol, which allows users to directly purchase inside AI Mode and Gemini, later in the same week.

Unrelated? Maybe, but putting them side-by-side – it is obvious to me that a website should start optimizing agentic experience to make sure content and product is discoverable, crawlable and understandable by AI agents.

Shifting from SERPs to AI-assisted buying journey

To most SEO professionals, Microsoft Playbook is just stating the obvious – users are moving away from searching directly or only on search engines. Instead, they may begin their buying journey in AI models such as ChatGPT or Gemini.

AI agents summarize, recommend, compare, and soon to transact on behalf of users. If your brand, content, or products are not legible to these systems, you risk neverentering the consideration set at all.

What makes it different from the traditional search is how personalized the recommendation could be. AI agents can infer and recommend products based on user info such as your preference, past history and location, to provide a more personalized result.

That’s why to those unfamiliar with AI search, it is important to understand how to influence AI models to interpret your website correctly.

How Microsoft explained the three data layers

One of the most interesting parts of Microsoft’s playbook is how it breaks down where AI systems get their understanding of brands and products.

It aggregates information from three different layers of data and translates it into personalized recommendations. What’s important is that these layers don’t replace one another. They stack.

Shout-out to Nano Banana for this nice infographic!

1. Crawled data

This is the foundation. It includes what AI models learned during training and what they can retrieve from indexed web pages in real time. This layer shapes the brand’s baseline perception, what category you belong to, how authoritative you seem, and how you’re positioned in the market.

In other words: your content, links, reputation, and traditional SEO signals still matter because they provide the grounding AI systems rely on when forming answers and recommendations.

2. Product feeds and APIs.

This is where brands gain real control. Feeds are structured data that you actively push to AI platforms, and they influence how your products appear in comparisons, summaries, and recommendations.

Unlike crawled content, feeds prioritize accuracy, completeness, and consistency.

Product information, especially pricing and availability, changes constantly. That’s why it’s important for AI agents to acquire the most up-to-date information about the product. You don’t want to click on a product just to discover that it’s sold out, right?

I would expect in the AI-driven shopping experience, maintaining control and audit over Merchant Center product feeds is essential for providing authoritative product truth to AI agents.

3. Live website data.

This is what AI agents see when they actually visit your site. Rich media, reviews, dynamic pricing, shipping policies, and even transaction capabilities all live here. This layer validates everything the AI thinks it knows about you.

If feeds say one thing and your site shows another, trust erodes quickly.

This layer also provides more context for the AI agents, which usually don’t exist in the crawled data or product feeds.

Here are a few examples that Microsoft give:

• Detailed reviews
• Video
• Current promotions
• Real-time delivery estimate

When optimizing, I would put my focus on surfacing real users’ reviews on the website. This often provides a trust signal and gives more context for AI agents to understand your product.

Microsoft’s recommended strategy to improve discoverability

Microsoft is very explicit about this: traditional SEO remains essential because AI systems perform real-time web searches throughout the shopping journey and not just at the moment of purchase.

Your site still needs to rank well to be discovered, evaluated, and ultimately recommended.

And apart from the traditional SEO tactics, Microsoft offers its own strategies for website owners.

1. Make your catalog machine-readable

Microsoft’s first recommendation is blunt: AI systems require structure and consistent data across all touchpoints.

At a practical level, this starts with making your catalog machine-readable.

That means treating your product data as something that needs to survive outside of your website’s visual layer. Prices, availability, variants, SKUs, GTINs, promotions should exist as explicit fields, not implied text.

I’m not surprised that Microsoft places a lot of focus on schema implementation:

• Implement relevant eCommerce schema, such as Product, Offer, AggregateRating, Brand, and FAQ schema.
• Dynamic fields like price, availability, and dateModified should be included in structured data.
• Promotions should have clear start and end dates.
• Product collections and category pages should use ItemList so AI understands groupings and relationships.
• For websites that serve customers in various countries, provide localized structured data explicitly via inLanguage and priceCurrency.

Another important thing to highlight is the alignment:

• Feed data
• On-site schema
• What users actually see

If those three disagree, AI systems lose confidence fast. And crucially: Microsoft explicitly warns against serving different HTML to bots. The rendered DOM needs to contain the same facts a human sees.

02. Reduce interpretation – content needs to answer intent and context

This is my favourite recommendation for AI search, because this highlights how different AI agents behave compared to the traditional search.

AI assistants interpret queries – not as keywords, but as intent. That’s why they push so hard on intent-driven, citable content.

In practice, this means:

• Pair product names with key differentiators in titles
• Write a clear descriptions with who it’s for, what problem it solves, and why it’s better
• Writing headings that mirror real-world questions and use cases

What I find interesting is how much emphasis they place on modular content to provide complementary context to the core product offering:

• Q&A blocks AI can reason over and cite
• Feature lists and key/value specs
• Comparison tables (“Model A vs Model B”) to highlight contextual difference
• “Goes well with” relationships between products
  

Apart from the textual content, Microsoft also explicitly calls out multi-modal signals:

• Video transcripts
• Detailed alt text and ImageObject schema

03. Build brand identity and trust via verifiable sources

The third pillar is trust, which. Microsoft treats it as something that must be provable. It’s about consistent, cross-verifiable signals.

AI systems prioritize sources that can demonstrate credibility through facts:

• Review volume
• Verified purchase ratios
• Aggregate sentiment that can be summarized (“highly rated for comfort and fit”)

This is why they stress proper use of Review and AggregateRating schema (so it’s machine-readable), not just testimonials buried in the UI.

Beyond reviews, Microsoft also recommends grounding brand authority in verifiable entities:

• Certifications (B Corp, sustainability standards)
• Partnerships with trusted brands
• Expert reviews and third-party articles
• Official brand identifiers and social links in structured data

My take

If I’m honest, parts of Microsoft’s playbook read a little…AI-generated.

I’m not talking about the em dash, but something in the playbook just doesn’t make sense to me.

For example,

I like the idea of adding video transcripts to parse out feature lists. But Voice assistants don’t consume structured data directly, so how structure data is put into play in the voice experience?

Another thing I don’t particularly like about the guide is the AEO vs GEO distinction. It reads a bit like taxonomy for taxonomy’s sake. It seems like they just try to capture both terminologies in the same playbook, but I don’t see any differences in the strategies when the same strategies ultimately power both.