Entity recognition before indexing: why Google can understand who you are before it trusts all your pages

Most people treat indexing as the first milestone.

That is too simple.

A search system can sometimes understand who an entity is before it decides to index, rank, or distribute every page around that entity.

This is why a site can have only a homepage indexed, but still trigger signs of entity understanding:

• a branded result that connects a handle to a person
• an AI answer that describes the person correctly
• a right-side entity card for a longer query
• a social profile ranking first for a short name query
• a person page that is crawlable but not yet fully trusted as the primary result

That does not mean the site is finished. It means the system is starting to separate two decisions that publishers often mix together:

1. Entity recognition: "Who or what is this?"
2. URL trust: "Which pages should we index, select, and show?"

When you understand that separation, a frustrating indexing problem becomes a useful diagnostic signal.

I am Mikhail Drozdov, also known as Casinokrisa. I study indexing-first visibility models: how search systems crawl, store, interpret, trust, and distribute information. The live identity layer for this site is here:

• canonical person page
• press and evidence layer
• research objects
• book page

This article explains the model behind the work.

The short version

If Google understands your entity but indexes only one or two pages, the problem is usually not "Google knows nothing."

It is usually one of these:

• the entity is recognizable, but the site is still young or narrow
• the homepage is the strongest source, so Google treats it as the safest canonical answer
• supporting pages are crawlable, but not yet trusted as separate destinations
• the external evidence graph is stronger than the internal page graph
• Google has enough confidence to describe you, but not enough confidence to expand the panel

That is a different problem from a technical block.

A technical block says:

"I cannot process this."

An entity trust bottleneck says:

"I can see this, but I am not ready to distribute it broadly."

The fix is not mass publishing. The fix is a cleaner graph.

Recognition is not the same as indexing

Indexing is a URL-level decision.

Entity recognition is a graph-level decision.

A URL can be ignored while the entity behind it becomes clearer. A person can be understood through:

• LinkedIn
• Instagram
• X
• Google Scholar
• ORCID
• Amazon
• SSRN
• press mentions
• a homepage
• a person page

The system does not need every page indexed before it starts connecting those signals.

That is why branded search is often the earliest visible indicator of entity formation.

If a query like who is casinokrisa produces an answer that connects the alias to a person, the system has already built a basic bridge:

Casinokrisa -> Mikhail Drozdov -> role -> website -> external profiles

If a query like Mikhail Drozdov casinokrisa triggers a small right-side entity block, the bridge is stronger:

name + alias + image + date/fact + profiles

The site may still have a coverage problem. But the entity model is no longer blank.

The three layers: identity, evidence, distribution

I use a simple model for diagnosing this stage:

1. Identity layer
2. Evidence layer
3. Distribution layer

Each layer has a different failure mode.

1. Identity layer

The identity layer answers:

• What is the canonical name?
• What aliases point to the same person?
• What is the primary role?
• What is the official site?
• Which image represents the person?
• Which external profiles are true identity matches?

For a person entity, this layer should be boring.

Use one name, one role line, one canonical URL, and one primary image. Variation feels natural to humans, but it raises cost for machines.

For this site, the intended identity line is:

Mikhail Drozdov (Casinokrisa) is an AI Search & Indexing Systems Researcher and Founder of Casinokrisa.

That line should appear consistently on the homepage, person page, LinkedIn, Instagram, X, and strong external references.

2. Evidence layer

The evidence layer answers:

• What independent sources mention the person?
• What formal profiles exist?
• Are there publications, papers, books, or research objects?
• Does the external web repeat the same name and role?
• Do third-party pages connect the person to the same site or organization?

This is where a press page becomes useful.

A press page is not a vanity page. It is an evidence router.

It should separate:

• official profiles
• external references
• quotes and mentions
• academic or research listings
• book or publication records

The page should not pretend every mention has equal weight. A topical quote in an SEO publication is stronger for an SEO entity than a generic business quote. Both can help, but they should not be treated the same.

3. Distribution layer

The distribution layer answers:

• Which pages get indexed?
• Which pages get selected for snippets or AI answers?
• Which sources appear in branded search?
• Which image is promoted?
• Which facts make it into the panel?

Distribution is where people get nervous because it is visible.

But distribution is downstream. If the identity and evidence layers are unstable, the distribution layer will keep changing.

That is why a small panel can appear, disappear, and reappear. The system is not necessarily confused. It may be recalculating confidence as new signals arrive.

Why only the homepage may index first

For a narrow or new expert site, the homepage is often the safest URL.

It has:

• the most internal links
• the clearest domain-level signal
• the strongest branded relevance
• the fewest duplicate intent conflicts
• the highest chance of matching navigational queries

If the homepage is indexed and the person page is only crawled, that does not automatically mean the person page is bad.

It can mean the homepage is currently doing the job of:

• site root
• entity summary
• brand page
• person summary
• topic doorway

That is heavy, but it is common.

The solution is not to force 50 pages into the index. The solution is to make the homepage a clean router that points to the few pages that matter:

• person page
• press page
• research page
• book page
• one strong topical hub

Then let the supporting URLs earn separate selection over time.

The wrong response: publishing more generic articles

When a site has one indexed page, the instinct is to publish more.

That can make the problem worse.

If the new pages are generic, overlapping, or AI-shaped, they dilute the graph. Google sees a site that is trying to expand surface area before it has proven a stable center.

For an entity-driven site, the better sequence is:

1. stabilize the homepage
2. stabilize the person page
3. build an evidence page
4. create durable work objects, such as research pages or a book page
5. publish supporting articles only when they clarify the model

The question is not:

"Can we publish another post?"

The better question is:

"Does this post make the entity easier to understand, verify, or select?"

If the answer is no, do not publish it.

The right response: make the graph legible

There are five moves that usually help.

1. Use the homepage as the current entity anchor

If the homepage is the only reliably indexed URL, do not fight that reality.

Make the homepage say the essential facts cleanly:

• name
• alias
• role
• founder relationship
• canonical person page
• book or research object
• evidence page
• primary image

This does not mean turning the homepage into a biography. It means making the identity layer unavoidable.

2. Keep the person page as the canonical identity page

The homepage can be the strongest indexed node, but the person page should still be the canonical identity page.

That page should contain:

• full name
• aliases
• role
• birth facts if publicly used
• official image
• official profiles
• evidence links
• author-of relationships
• structured data

The homepage points to it. The person page explains it.

3. Treat the book as a work entity

A book listing on Amazon is useful, but it is not fully controlled by the site.

An on-site book page helps create a work object:

Person -> authorOf -> Book -> sameAs -> Amazon

This gives the site a cleaner way to explain that the person has authored a specific work.

That is why the Indexing-First Search Systems book page matters.

It is not there to sell harder. It is there to make the work entity explicit.

4. Separate sameAs from evidence

This is one of the most common structured data mistakes.

sameAs should point to profiles or pages that represent the same entity:

• LinkedIn profile
• X profile
• Instagram profile
• Google Scholar profile
• ORCID profile
• Crunchbase profile

It should not point to generic domains like amazon.com or ssrn.com.

Articles, quotes, press mentions, and book listings are better modeled as evidence:

• subjectOf
• authorOf
• mentions
• visible links in a press page

This distinction matters because it tells the system which URLs are identity equivalents and which URLs are supporting proof.

5. Keep the image boring and consistent

A person image is not just a design asset.

It is a candidate for entity resolution.

The best image is not always the most flattering. It is the one that is:

• stable
• repeated across strong profiles
• high resolution
• face-forward or clearly recognizable
• not competing with many other versions

If Google keeps using a small or awkward image, the answer is usually not to upload ten new photos. The answer is to make one photo the obvious winner.

For the deeper playbook, read Photo authority for Person entities.

How to diagnose the stage you are in

Use queries, not feelings.

Track the same set every week:

• Mikhail Drozdov
• Mikhail Drozdov casinokrisa
• who is casinokrisa
• Casinokrisa
• Mikhail Drozdov AI Search
• Mikhail Drozdov indexing

Then classify what you see.

Stage 1: No entity resolution

Symptoms:

• unrelated people dominate
• no branded understanding
• no consistent image
• your site does not appear for your name

Fix:

• strengthen identity profiles
• use one role line
• add a canonical person page
• connect official profiles

Stage 2: Alias recognition

Symptoms:

• Google connects the handle to the person
• a social profile ranks first
• the site appears for branded queries
• AI answers may describe the entity

Fix:

• strengthen homepage and person page
• add evidence page
• reduce conflicting internal pages
• avoid mass content expansion

Stage 3: Partial panel

Symptoms:

• right-side card appears for longer queries
• image and basic facts show
• panel is unstable by query or locale
• Google still prefers social profiles for short queries

Fix:

• keep signals stable
• add stronger independent evidence
• improve image consensus
• send only core URLs for recrawl

Stage 4: Stable panel

Symptoms:

• panel appears for short name or brand queries
• image is stable
• role is stable
• site and profiles are consistently connected

Fix:

• stop changing core facts
• add new evidence slowly
• maintain clean pages
• keep external profiles aligned

What to send for recrawl

Do not request indexing for every support page.

Send the core graph:

• homepage
• person page
• press page
• research page
• book page

If Google indexes only the homepage first, that is still useful. The homepage can carry the strongest summary while the rest of the graph becomes more trusted.

What this means for AI answers

AI answers are often the first place where weak entity understanding becomes visible.

If an AI Overview correctly says:

"Casinokrisa is the pseudonym of Mikhail Drozdov..."

then the entity relationship is already being used in answer generation.

That does not guarantee a Knowledge Panel. But it is a strong intermediate signal.

It means the system can summarize the relationship.

The next challenge is not comprehension. It is confidence.

To move from comprehension to confidence, the web needs to repeat the same facts across independent surfaces.

The practical rule

When only the homepage is indexed, treat it as the current entity anchor.

When the person page starts getting indexed, treat it as the canonical identity source.

When the press, research, and book pages start getting indexed, treat them as proof objects.

Do not panic if those steps happen out of order.

Search systems do not build trust in the same order publishers build websites.

They often understand the entity first, test the homepage second, and distribute the supporting URLs later.

That is why the right goal is not "index everything."

The right goal is:

Make the entity impossible to misunderstand, then make each supporting URL worth selecting.

That is how a small site moves from one indexed page to branded search confidence, then to entity cards, then to a more stable Knowledge Panel.