The search index is evolving from ranking pages to supporting AI-generated answers. In a technical blog post “on the evolving technical characteristics of the index,” published today, Microsoft Bing explained why AI search needs a different indexing system than traditional web search.

Traditional search vs. grounding systems. Microsoft said traditional search can rely on users to self-correct, while AI systems need stronger evidence because they generate committed answers.

• Traditional search is built around documents. Users get ranked links, scan the results, and decide what to trust.
• Grounding systems are built around supportable facts with clear sourcing. The AI uses that information to generate a combined answer, where mistakes can compound across sources and reasoning steps.

They shared this table:

What’s different. Traditional ranking is optimized for relevance. Grounding must also assess whether information is accurate, up to date, clearly sourced, and sufficient to support an answer. That means AI indexes need to account for whether:

• A page’s meaning survives chunking and transformation.
• The source is clearly identified.
• The information is fresh enough to use.
• Important facts are actually retrievable and groundable.
• Grounding systems need to detect disagreements between sources before generating an answer.

Stale content. Stale content creates a different risk in AI answers, Microsoft said. In traditional search, it may hurt ranking quality. In grounding systems, it can directly generate a wrong answer.

Contradictions. A search engine can rank one source above another and let users decide. Grounding systems must recognize conflicting evidence before turning it into a single answer, according to Microsoft.

Retrieval is more complex. Search is usually a single interaction: query in, ranked results out. Microsoft said grounded AI systems may retrieve information repeatedly, refine based on earlier results, combine evidence, and reassess confidence before answering.

How indexing quality is measured. Search quality has traditionally focused on ranking performance and user behavior. Grounding systems also need to measure factual fidelity, source quality, freshness, evidence strength, and conflict detection. The industry is still learning how to rigorously measure grounding quality, Microsoft said.

Grounding doesn’t replace search. Grounding builds on existing search infrastructure while adding systems focused on evidence quality, attribution, and deciding when an AI system should avoid answering, Microsoft said.

Why we care. For decades, search indexes helped determine which pages users should visit. Today, AI grounding determines which information supports an AI-generated answer. Microsoft described grounding as a new layer on top of traditional search, built for AI systems that need higher confidence in the information they use. That shift could push brands and publishers to focus more on creating information AI systems can confidently use.

The blog post. Evolving role of the index: From ranking pages to supporting answers

AI may not see your brand the way you think it does, according to Scott Stouffer, co-founder and CTO at Market Brew.

Brands still publish content, optimize pages, build authority, and follow SEO best practices. But that may not be enough anymore.

Search has moved away from a simple battle over keywords, links, and page-level signals. It’s now shaped by meaning, intent, embeddings, and retrieval, Stouffer said during his SEO Week presentation.

In legacy SEO, a page could rank lower and still exist in the search results. In AI-driven systems, the first question isn’t whether you rank. It’s whether you’re ever retrieved.

“If you’re not retrieved, you do not exist to AI,” Stouffer said.

Your brand already exists inside AI systems as a mathematical object. You may call yourself one thing. Your homepage may say another. Your brand guidelines may promise a clear position. But AI systems build their own view of your brand from the content you have published.

That computed version of your brand may be different from the one you intended to build.

Retrieval now matters before ranking

AI visibility begins before ranking, Stouffer said.

In traditional SEO, marketers focus on positions — first, third, or tenth. But AI systems apply a filter earlier. Before anything is ranked, the system determines which content is eligible for consideration.

That is retrieval.

When a user asks a question, the system pulls a limited set of passages or chunks that best match the query. Those passages define the answer space.

If your content isn’t included, you get no impressions, no clicks, and no visibility at all, Stouffer said.

The real shift is moving from exclusion to inclusion.

“You don’t lose. You just never entered the game,” Stouffer said.

AI does not see pages the way SEOs do

AI systems don’t treat a webpage as one clean unit, Stouffer said. They don’t evaluate pages as whole objects or prioritize layout, structure, or formatting.

Content is broken apart. A page becomes chunks: passages, sections, and individual ideas.

Each chunk is evaluated independently. A paragraph deep in a guide can compete on its own. A single sentence can be selected if it aligns closely with the query.

This shifts competition from page versus page to passage versus passage.

Most of a page may never be considered. Only the most aligned chunks are evaluated.

Meaning becomes math

Each chunk is converted into a vector, Stouffer explained.

This vector represents meaning as a position in a high-dimensional space. It captures context and intent rather than exact wording.

Two pieces of content can use different words but sit close together if they express the same idea. Others can share keywords, but sit far apart if they represent different meanings.

“It’s comparing meaning, not wording, measuring distance, not keyword overlap,” Stouffer said.

Relevance is determined by proximity. The closer a chunk is to a query in this space, the more likely it is to be retrieved.

Your content forms clusters

As chunks are mapped into this space, they group together.

Content with similar meaning forms clusters, even across different pages. These clusters reflect how AI systems understand topics.

This understanding comes from how content naturally groups by meaning, not by site structure or labels, Stouffer said.

If content is consistent, clusters become dense and clear. If content is scattered, clusters become fragmented.

What matters is not what a brand intends to say, but what its content actually communicates.

The centroid is your brand to AI

Within these clusters, there is a center point — the centroid, Stouffer said.

The centroid represents the average position of all related content. It reflects the site’s core meaning.

Every page and paragraph influences that position. Consistent content creates a clear, stable centroid. Inconsistent content dilutes it.

That centroid is how AI understands your brand.

Not your homepage. Not your messaging. Not your brand guidelines.

Your centroid is the combined signal of everything you have published, Stouffer said.

“Your centroid doesn’t care about intent. It reflects the math of everything you’ve ever published,” Stouffer said.

Alignment beats isolated optimization

This changes how content should be evaluated.

The key question isn’t whether a page is optimized in isolation. It’s whether it aligns with the rest of the site.

Each page either strengthens the centroid or pulls it in a different direction.

“Optimization without alignment creates drift, and drift is what breaks consistency,” Stouffer said.

As drift increases, the site becomes harder for AI systems to interpret and retrieve.

“You don’t write pages, you project meaning,” Stouffer said.

Retrieval starts with proximity

When a query is entered, the system converts it into a vector, Stouffer said.

It then searches for the closest matches in meaning space.

This includes both individual chunks and the centroids that represent broader content clusters.

If your content is close enough, it enters the candidate set. If it is too far away, it is excluded.

Only after this stage do traditional ranking signals apply.

Content quality, links, and structure matter — but only if the content is first retrieved.

If not, those signals are never evaluated, he said.

Most brands look too similar to AI

Many brands follow similar strategies, use the same sources, and produce similar content.

As a result, their centroids converge in the same region, Stouffer said.

He described this as cluster collision.

When multiple brands occupy the same space, AI systems don’t select all of them. They choose a few and ignore the rest.

“They’re not failing best practices. They’re colliding with everyone else using them,” Stouffer said.

Distinct meaning is the new advantage

Producing more content or improving existing content isn’t enough. If content remains similar in meaning, it remains in the same space.

“You need a distinct centroid,” Stouffer said.

A clear, separate position in meaning space reduces competition and increases the likelihood of retrieval.

SEO becomes a control loop

This is not a one-time adjustment.

Every piece of content shifts the centroid.

That requires an ongoing process of measurement and adjustment, Stouffer said.

Teams need to monitor alignment continuously and correct drift as it occurs.

Over time, this creates a more stable system where new content reinforces the existing structure.

The visibility problem is really an observability problem

Most teams can’t see how their content exists in this system.

They can’t see clusters, centroids, or distances — or why content is excluded.

So they rely on trial and error, Stouffer said.

They publish, optimize, and wait for results. When nothing changes, they try something else.

Without visibility into the system, they react to outcomes rather than understanding causes.

Is AI seeing the brand you think you’ve built?

Your brand already exists as a mathematical object inside AI systems, Stouffer said.

You do not get to choose that.

You only choose whether to measure and control it or let it drift.

AI does not see your brand the way you describe it. It sees the aggregate meaning of your content.

“If you control your centroid, you control your visibility,” Stouffer said.