The Knowledge Panel Course: How a Knowledge Panel Is Built
Script from the lesson The Knowledge Panel Course
How a Knowledge Panel Is Built
Jason Barnard speaking: Hi and welcome. One thing that is often overlooked is how Google actually builds a Knowledge Panel. As we have seen, a presence in the Knowledge Vault or one of the Knowledge Graph verticals is the starting point. Each of these verticals, including the Knowledge Vault, attributes a kgmid to the entity. That kgmid is unique for each entity.
Jason Barnard speaking: So, Jason Barnard, me, has the kgmid /g/11cm_q3wqr, Jason Barnard, who teaches at the Golden Gate University, has the kgmid /g/11l48kvhjg, and Jason Barnard, the actor in the film, First Man, has the kgmid /g/11fw71_nbj.
Jason Barnard speaking: Bill Slawski mentioned to me a couple of years ago that a Google patent indicates that there is a centralised lookup table with these IDs so that all the algorithms can access them. The lookup table contains the ID and the name. Each Knowledge Graph vertical adds new entities to that lookup table as it creates them. That isn’t much information.
Jason Barnard speaking: Each vertical Knowledge Graph contains much more information than that. And they use that additional information in their own algorithms in their vertical, be it Podcasts, Books, or Google Maps, but they also use that information in the Rich Elements they provide to Google in the main SERPs, for example, podcast boxes, image boxes, and map packs.
Jason Barnard speaking: Previously, I’ve talked about the fact that Knowledge Panels are triggered by vertical Knowledge Graphs. But when building the SERP, the whole page algorithm is using the lookup table too. So, it has the kgmid, the name, and knows which vertical provided the entity.
Jason Barnard speaking: As with all the other Rich Elements on a SERP, there is a dedicated algorithm for a Knowledge Panel, that builds the panel and then submits it to the whole page algorithm. If you haven’t already done so, please do read my article about Darwinism in Search and my article about the whole page algorithm, since those will help you to understand how Google fits the SERPs together. You will find links to those in the lesson material below.
Jason Barnard speaking: Now, how does this specifically build a Knowledge Panel? I’m going to use a chest of drawers analogy. The kgmid plus the name is a reference to a chest of drawers containing information about the named entity. My chest of drawers is called /g/11cm_q3wqr Jason Barnard.
Jason Barnard speaking: The chest of drawers has multiple small drawers, each of which contains a piece of information about the named entity. The Knowledge Graph vertical the entity comes from adds drawers with labels and fills the drawers with the information it has. For example, a person might have a set of drawers that contains a description, photos, date of birth, net worth, squad number, role in a film, or whatever. A company might have a founding date, founders, subsidiaries, revenues, logo, and so on.
Jason Barnard speaking: At this point, the Knowledge Panel algorithm could simply build a Knowledge Panel for the entity from the information in each drawer. You’d think the story stops there and that this lesson is over, absolutely not. There are multiple vital aspects to understand before I continue. Hold onto your hat. This gets complicated very quickly. But once you’ve got the idea, it is very simple and incredibly powerful when managing your Knowledge Panel. In fact, it is absolutely fundamental to managing, enhancing, and correcting Knowledge Panels, so stick with me here.
Jason Barnard speaking: Firstly, some drawers are locked and cannot be changed by human beings: the entity name, the ID, and the Entity Home, for example. The contents of most of the others can be changed, and more drawers can be added using data from the Web Index. However, as we will see, depending on the Knowledge Vertical, there can be additional drawers that are locked for the Web Index, but these can be changed by human beings at Google.
Jason Barnard speaking: Looking at my Knowledge Panel, the Entity Home, represented by the world icon, is locked. The description is locked, since it comes from Google’s own human curated books knowledge vertical. I am working on seeing how to replace that with my own
description, but this is another example, as with Wikipedia, of handing control to a third party that is the disadvantage of these easy wins for Knowledge Panels. We cannot know for sure for the rest, but it is all in the Knowledge Vault. And I would imagine my date of birth, my partner, my mother are all locked, and the albums and TV shows aren’t.
Jason Barnard speaking: Secondly, the chest of drawers are built on flexible templates of the drawers they should contain. For example, a film star would have a templated chest of drawers that includes film roles. A football star would have a template that includes squad number, salary, teams played for, and so on. And I say flexible, since a football star
can also be a film star. So, a specific chest of drawers could be a mixture of both. It’s a modular templated system.
Jason Barnard speaking: Thirdly, when building the Knowledge Panel, the algorithm will choose only the information that is judged to be actually useful to the searcher. As we have seen in previous lessons, when you go Knowledge Panel hopping, the Knowledge Panel building algorithm will add more information from additional drawers, if it has any.
Jason Barnard speaking: All of that immediately shows you why you shouldn’t focus too much on the Knowledge Graphs or the Knowledge Vault for what displays in the Knowledge Panel. As I have mentioned in previous lessons, some Knowledge Graph verticals have additional locked drawers and limited or no possibility of additional drawers.
Jason Barnard speaking: As of late 2022, podcasts are an example of this. Unless they are in the Main Knowledge Vault, the information shown in the Knowledge Panel is locked.
Jason Barnard speaking: So, now we have a chest of drawers, with multiple drawers containing a piece of information about the entity. At this point, an algorithm using the Web Index will go through each of the unlocked drawers and assess whether that information is, in fact, true today based on the information it has in the Web Index. If it determines that the web contains sufficient reliable evidence that the fact is wrong, the algorithm will replace the information in the drawer.
Jason Barnard speaking: An example of this is that in the Knowledge Vault, my subtitle is musician. But in the Knowledge Panel, the subtitle is author. When the Knowledge Graph updates, it will replace the information in the subtitle drawer with musician. That will appear in my Knowledge Panel for a few days, and the Web Index algorithm will then switch it back. Google’s Knowledge Graph API doesn’t give us much insights into the information the Knowledge Graph contains. So, it is unclear exactly which pieces of information in which drawers it controls. But after 7 years tracking, collecting data from Knowledge Panels and the Knowledge Graph API, and cross referencing the data, it is clear that some drawers are filled by the Knowledge Vault or other knowledge verticals and are locked.
Jason Barnard speaking: I can make very good educated guesses about which drawers are locked on a per template basis, but that is beyond the scope of this lesson. There are way too many different entity types, templates, and template combinations.
Jason Barnard speaking: But the huge trick to play here when managing a Knowledge Panel is to get into the Knowledge Vault and add information to it. Build its confidence in that information, to the point at which it is sufficiently confident to lock the drawer and prevent the Web Index algorithm overriding it. That takes months and possibly years to achieve.
Jason Barnard speaking: But this explains the volatility of the information in Knowledge Panels when the Knowledge Vault updates. If your information is switching back and forth, then the Knowledge Vault or other knowledge vertical hasn’t locked the drawer, and the Web Index is able to override.
Jason Barnard speaking: In my example, in the Knowledge Vault, I am seen as a musician because I have released 6 music albums and recorded over 100 songs, all of which are in the trunk of the Knowledge Vault. Because the information is in the trunk, it is very difficult to change. However, when I published my book, The Fundamentals of Brand SERPs for Business, and changed my primary semantic triple to “Jason Barnard is an author,” the Web Index algorithm disagrees and corrects the information in that drawer and puts it in the Knowledge Panel. Look in the additional materials, and you’ll find a link to the Kalicube case study on how I achieved the switch from musician to author in the Knowledge Panel.
Jason Barnard speaking: Note that the description and some other information the Knowledge Vault has taken from Wikipedia is locked. As of late 2022, if the description drawer contains a description from Wikipedia, then it is locked. I expect this to change in the future. The description drawer is also locked when it comes from Google Podcasts and Google Books. The drawers containing the books written by the author are not locked, however.
Jason Barnard speaking: Now, just like the Knowledge Vault, all the knowledge verticals will go through the chests of drawers they own from time to time and update the information. But as with the Knowledge Vault, the Web Index algorithm has the last word on all drawers that these Knowledge Graph verticals do not lock. So, looking at Google Books again, this example explains why sometimes books become unattached or lost from Knowledge Panels of the author.
Jason Barnard speaking: But this process of losing and gaining information in a Knowledge Panel is easy to understand once you have grasped this analogy of a chest of drawers, with locked and unlocked drawers filled by the Knowledge Verticals and optionally corrected by the Web Index.
Jason Barnard speaking: Two things strike me here. Number one is don’t panic when the information changes. It is either the knowledge vertical doing a spring clean or the Web Index driven Knowledge Algorithm correcting the information in a drawer, removing a drawer, or adding a new drawer.
Jason Barnard speaking: Generally speaking, with the exception of the locked drawers where the information is truly fixed, the rest is pretty flexible and changeable. The Web Index algorithm updates constantly. In Kalicube Pro database, we see that the contents of Knowledge Panels are very volatile.
Jason Barnard speaking: The Knowledge Vault also updates constantly, so information about an entity can change from one day to the next, even outside an explicit Knowledge Graph update. New entities are added on a day to day basis. And the Knowledge Graph updates are simply where we see the most action. That said, day to day, the Knowledge Vault will only change 1% of entities or less.
Jason Barnard speaking: Back to the Knowledge Panel. What Google is displaying in a Knowledge Panel is similar to changes in the left rail or in the blue link algorithm. Remember, the aim of the Knowledge Panel is to show the information Google judges to be the information that the searcher would want to see and is designed simply to save the searcher the effort of visiting multiple sites in the left rail results. And what Google perceives to be the information that is the most helpful will change over time as it understands more about the entity, but also the points of interest about that entity for its audience.
Jason Barnard speaking: Add to that that Google is constantly in a state of A/B testing, and things can get quite messy. In short, don’t freak out when something changes day to day, since most of the time, it isn’t a fundamental change in Google’s understanding. It’s the Web Index Knowledge Algorithm changing information.
Jason Barnard speaking: One last point is that Knowledge Panels are geo sensitive. Although some information is universal across the world, some information is localised, such as the customer service number and social media accounts for international corporations. The actual information in the Knowledge Panel algorithm chooses to show in different countries can also vary, since the audience in one country might have different cultural focuses. The audience in one country might be interested in the revenues of a company. Another might be more interested in the C-level employees. In these cases, Google adapts the Knowledge Panel country by country.
Jason Barnard speaking: So, Knowledge Panels are, to a great extent, fluid and change over time along with the Web Index. They are more or less real time summaries of the facts as understood by the dynamic Knowledge Algorithms that use the data that flows through the Web Index.
Jason Barnard speaking: Now, a Featured Snippet on Google is a best guess of the most accurate answer to a question. Many elements in a Knowledge Panel work in a similar manner. The algorithm for the Featured Snippet and the algorithms that fill the Knowledge Panels with information are close relatives. So, above and beyond the foundational fact that a Knowledge Panel triggers for an entity because it has been confidently understood by one of Google’s knowledge verticals, think of the information in a Knowledge Panel as being a combination of multiple tiny Featured Snippets.
Jason Barnard speaking: As you can see from that list, there are three types of questions you need to ask yourself: number one, what is Google’s foundational understanding, number two, which knowledge vertical am I dealing with, and number three, what is the dominant information about this in Google’s dynamic Web Index.
Jason Barnard speaking: And the last point is that the Entity Home has a special drawer that is locked. And from our experiments, as the Knowledge Algorithms see that the information on the Entity Home is correct, over time, they build trust for the information provided about the entity by that page and by extension, that website. This leads to a vastly increased capability to convince these algorithms to add what you want to those drawers and ideally, lock them.
Jason Barnard speaking: Thank you, and see you soon.
