How I Lost My Knowledge Panel (And What I Learnt About The Google Entity System)

What I Learnt From Losing My Panel

I learnt some important things:

You can have a Knowledge Graph entry without a visible panel – You can have an entry and a really good knowledge panel in Google’s Knowledge Graph (the big database of all the entities it knows about), but that doesn’t mean that your awesome knowledge panel will show up when someone searches your name in a normal Google search.

Google regularly purges uncertain entities – Google regularly updates its Knowledge Graph and gets rid of any entities it’s not 100% sure on the facts about. Jason Barnard has a great Knowledge Graph sensor tool which allows you to see the current volatility – https://kalicube.pro/tools/knowledge-graph-sensor

Multiple entities with the same name create competition – It takes real dedication to build a knowledge panel when there are multiple people with the same name as you.

At the time, there were several Kelly Sheppards in Google’s Knowledge Graph. Now, as of 2026, there are even more.

My biggest knowledge panel “rival” was an employee of Skidmore College. He always showed up with his knowledge panel, even from UK searches, because he had a much stronger entity than me. I could see this in Google’s Knowledge Graph by the relevancy score. His was 223, mine was only 24 (the “default” score for person entities). All I had was this Knowledge Panel Sprout.

The Entity Migration Problem

As I track my brand and person entities daily, I noticed something strange happening after I’d left. There were now multiple knowledge panel sprouts for my name, all associated with the “real” me and my LinkedIn.

It was clear that Google was really confused, and I’m not surprised. I’d built up a lot of authority at Sleeping Giant Media, and all of a sudden, Google realised that the Kelly Sheppard who started The Structured Data Company wasn’t necessarily the “same” Kelly Sheppard as the one who worked at Sleeping Giant Media, so it deleted my previous knowledge panel entry. It started creating new ones based on various different sources that it could find, as they were all pointing to a slightly different “Kelly Sheppard”. This continued until I worked to make sure that one of them became stronger than the others. Eventually, Google will decide that the other sprouts it created lack verifiable sources, and will deletes them.

The same thing happens during a brand migration, as I’d recently carried out a brand and website migration for a client and noticed that the same thing happened to them.

I started to dig deeper to understand how Google really finds entities and how it works.

How Google Stores Entity Data

I found out that Google uses something called data-maindata to store the structured facts it has about an entity. When it renders a Knowledge Panel, it doesn’t do a fresh search every time. It looks up the kgmid which has a pre-loaded set of data from the Knowledge Graph.

Here’s what it looks like for a person:

data-maindata='[null,"/g/11x61tzq2y","Kelly Sheppard","vise:/g/11x61tzq2y",["PEOPLE"],null,null,null,"PEOPLE","en-GB",null,null,null,[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,[133]],null,null,null,null,null,132,null,null,null,null,"CKsV"]

And For a book:

data-maindata='[null,"/g/11x5tljt0v","The Structured Data Guide for Beginners","vise:/g/11x5tljt0v",["BOOKS"],null,null,null,"BOOKS","en-GB",null,null,null,[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,[133]],null,null,null,null,null,132,null,null,null,null,"CKsV"]

This basically says: here’s the kgmid, here’s the name of that entity, here’s the type of entity it is (PEOPLE / BOOKS) and then lots of other data points. Sometimes these are filled in if it’s a big knowledge panel, like this one for Nike:

 

data-maindata='[null,"/m/0lwkh","Nike","vise:/m/0lwkh",["COMPANIES","CHAINS"],null,null,null,"SHOPPING_MERCHANT","en-GB",null,null,null,[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,[132,133]],null,null,null,null,null,104,null,null,null,null,"CKsV"]

That’s a shopping merchant panel. Then you have the actual “real” full knowledge panels for companies like Google:

 

data-maindata='[null,"/m/045c7b","Google","vise:/m/045c7b",["COMPANIES","CHAINS"],null,null,null,"COMPANY","en-GB",null,null,null,[null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,null,[133]],null,null,null,null,null,132,null,null,null,null,"CKsV"]

If a company has multiple locations (over 10) then Google can group them into a single kgmid and use “CHAINS” to resolve them into a single company entity.

The Knowledge Graph Relationship Map

Inside the Knowledge Graph, Google stores relationships to other kgmids (entities). For example, the kgmid for Microsoft would have a relationship to the kgmid for Bill Gates. These relationship maps (aka Knowledge Graphs) for companies are really, really important. We can help influence these using structured data.

Google builds these entities using something called Entity Reconciliation. It looks for a consensus of facts:

1. Wikipedia, Wikidata & LinkedIn Information

All three sources should agree on the facts, e.g. when a company was founded. If they do, Google believes it to be true and it gets entered into the Knowledge Graph.

2. Structured Data

Using `sameAs` we can help Google understand what social media profiles belong to the company. But there’s more we can do – based on testing, using actual kgmids within properties like `knowsAbout` and `memberOf` appears to create stronger entity connections than just using text strings.

3. Entity Home

Google looks for your entity home, which is nearest website it thinks is about your company. If it can confirm this link between the domain and the kgmid, then you get a little world symbol on your knowledge panel and Google will look there as the source of truth for future updates.

 

Tracking Entities

When you have an entry in Google’s Knowledge Graph, it gives each entity a relevancy score based on how relevant they are to a search term, or another entity.

So for example, if I search for Kelly Sheppard, it will return all of the Kelly Sheppard’s in its Knowledge Graph, each with a relevancy score. The higher the score, the more related it thinks that the entity is to what you are searching for, and the more likely you are to see a knowledge panel in search results.

You can track your entity scores using a brand entity tracking tool like Entity Scout which tracks brand, person or product entities on a daily basis to tell you how they are performing over time. This can be really handy to know, as you can instantly spot if your brand entity relevancy score drops for your key brand names, meaning you can take action to try to fix the issue.

2026 – The Age of AI

However, in the age of AI, traditional entity reconciliation isn’t enough. There’s now something I call the entity trident:

1. Your search presence
2. LLM presence
3. Knowledge Graph presence

Google is increasingly using what I call fragmented reconciliation which means if your website isn’t the strongest signal for Google, then it might pick your LinkedIn, a Wikipedia page or even a website or social media profile that isn’t even yours as your entity home if it thinks it belongs to your company. This is also particularly common for knowledge panel descriptions.

This is why at The Structured Data Company, we use kgmids in our structured data. This creates a loop that tells crawlers this page is about this exact entity. This means it can’t get it wrong as to which entity we are referring to, as each kgmid is a unique and distinguishable entity. There is a down side to this though – as we know from earlier, kgmid’s can, and do, change on a regular basis.

When looking for kgmids to add to things like sameAs, you should use the /m/ prefix identifiers rather than the /g/ ones wherever possible, as the /m/ prefixes are from Freebase which is a huge database that Google acquired in 2010 and used it to build the Knowledge Graph in 2014. The /g/ kgmids are new Google Knowledge Graph ids which can change more regularly.

To make this even more foolproof, you need the links that you are referring to in your sameAs, such as your social media profiles, to also link back to the entity home. If the entity home doesn’t have at least five to ten high-profile sources linked, then Google may choose a third-party source.

What Google Really Trusts

Behind the scenes, Google has a reconciliation engine which looks for sources of facts. Here’s what actually matters:

Wikipedia Still Matters

Despite some of the recent news, Google still loves Wikipedia. Google’s confidence score for a fact jumps if the Wikidata statement has a “stated in” reference that points to a high-authority news site (New York Times, Bloomberg, Reuters) rather than just “imported from Wikipedia.”

WikiData Quality Matters More Than Existence

Don’t just create a WikiData page and leave it blank. It needs statements and idenfiers to connect the data together. A well-structured WikiData entry with well-researched properties and cited sources carries significantly more weight than a bare-bones entry.

The Freshness Algorithm

Google has a freshness algorithm which scrapes news and Google News in particular. This is why syndicated press releases can be really good for changing information in the Knowledge Graph quickly. A press release through a high-tier wire (BusinessWire/PR Newswire) with your brand name in the headline acts as a “trigger” for the Knowledge Graph to re-verify the entity.

This is particularly important for brand migrations. When a company goes through a re-branding process, or a website migration, it’s normal for Google to delete their knowledge panel and make a new one. By adding structured data to the new website which references both the old and new name and company information, and pointing press releases at it using the new name, Google can be more certain that the new company it has found is the same as the old one, albeit under a different name.

Building Knowledge Panels for “Non-Notable” People

Google claims it requires notability, but really it requires confidence.

Here’s what actually works based on testing:

The Google Books/Scholar Credibility Boost

Google trusts its own vertical indexes (Books and Scholar) more than the open web. If a person is an “Author” in Google Books (even of a small technical manual) or has a profile in Google Scholar with 1-2 citations, the threshold for a Knowledge Panel drops significantly.

The Alternative Wiki Strategy

In 2026, Google is indexing “Entity Databases” as alternatives to Wikipedia and Wikidata like:

• Crunchbase
• Grokipedia
• Golden.com
• Wikitia (paid service)

If you can get an entity into several of these, and they all reference the same information about the company or person, the Knowledge Graph will often generate a panel even without a Wikipedia page. There’s no guaranteed threshold, but appearing in multiple credible databases with consistent information significantly increases your confidence score.

Advanced Knowledge Panel Techniques That Actually Work

Based on extensive testing, here’s what appears to work:

memberOf

If you belong to any professional bodies or even notable groups (like an alumni association of a major university), reference them using memberOf when possible. This uses what’s called transitive authority – if Google trusts the organisation, and you are a verified member, some of that trust flows to your brand entity.

The Entity Co-Occurrence Strategy

Google doesn’t just read your site; it scans the web for co-occurrence patterns. This means that it almost “expects” to see Entity A and Entity B together in a sentence, for example Elon Musk and Tesla.

If “Entity A” (you) and “Entity B” (your company) appear together in a single sentence across multiple third-party high-authority domains, Google’s entity link prediction algorithms increase the confidence score of that relationship.

This is why consistent semantic triples are important. Focus on digital PR where your name and the company/product title appear in the same heading or first paragraph of external articles (interviews, reviews, podcasts, news mentions). It’s important to have this same semantic triple everywhere – your social media, directories, your about us page, third party mentions. That way Google understands it’s the same company being mentioned, even if those mentions are unlinked.

Unlinked Mentions Are Valuable

In 2026, unlinked mentions on high-authority sites (like a mention in a Guardian article without a link) are as valuable as backlinks for entity building. Google and LLMs read the text, identifies the entity, and this increases the confidence score of the entity.

Sprouting Behaviour

The “sprouting” behaviour you might see (where a knowledge panel expands then shrinks on almost a daily basis) is the hallmark of a confidence score threshold issue. Google’s reconciliation engine is finding enough data to trigger the panel, but its trust engine isn’t finding enough corroborating, independent, high-velocity signals to keep the rich attributes (the “massive” panel) active.

Creating Your Own Entity Home

Your entity home needs to be bulletproof. Here’s what that means:

Structured Data Requirements

• Full bespoke Organization/Person schema with kgmid references where applicable
• Proper sameAs linking to all official profiles
• All social profiles must link back to your entity home
• At least five high-authority external mentions
• @ids linking all of the relevant entities on the page together into a Knowledge Graph, e.g. Product, Brand, Organization, Offer

Content Requirements

• Clear, consistent information about founding dates, credentials, achievements, qualifications, certifications, awards etc.
• Matching information to what’s on Wikipedia/WikiData/LinkedIn/Crunchbase
• Regular updates (freshness signals matter)

The Circular Reference Pattern

Ensure your social media, WikiData, and website all point to each other in a circle. Google’s entity reconciliation uses this to confirm that “Author X” is definitely “Person X” is definitely “Company Founder X.”

What About AI and LLMs?

The relationship between Knowledge Graph entities and LLM mentions is still evolving. From my testing:

• LLMs appear to read structured data when it’s in the HTML (not JavaScript-injected)
• Co-occurrence patterns influence both Knowledge Graph confidence and LLM citation likelihood
• Having a strong Knowledge Graph entry doesn’t guarantee LLM mentions, but it helps
• Unlinked mentions in high-authority sources feed both systems

My Recommendations

After losing my panel and rebuilding it, here’s my top 8 things that I’d focus on:

1. Get your Wikipedia/WikiData sorted properly – Use references, use qualifiers, cite high-authority sources
2. Publish something in Google Books or Scholar – Even a small technical guide creates a credibility boost
3. Get listed in alternative entity databases – Crunchbase, Grok, Golden.com, Wikitia with consistent information
4. Focus on co-occurrence – Get mentioned alongside your company/book/achievements in third-party articles
5. Use kgmids in your structured data – Create those hard-wired entity connections
6. Build the circular reference pattern – Website ↔ WikiData ↔ Social Profiles ↔ LinkedIn all pointing to each other
7. Pursue unlinked mentions – High-authority coverage matters even without backlinks
8. Press releases for major updates – Use high-tier wires to trigger Knowledge Graph re-verification

The Bottom Line

Building a Knowledge Panel isn’t about gaming the system. It’s about creating genuine, verifiable, consistent entity signals across multiple trusted sources.

Google needs to be confident about who you are, what you do, and how you relate to other entities

The “secrets” aren’t really secrets – they’re just understanding how Google’s entity reconciliation actually works and giving it what it needs to be confident about your entity.

And yes, you might lose your panel during transitions when Google isn’t quite sure of the facts. That’s normal. The key is understanding why it happens and how to rebuild stronger than before.

Want help building your brand entity? I offer brand entity audits and ongoing optimisation.