The Knowledge Panel Course: How to Change Information in a Knowledge Panel
Script from the lesson The Knowledge Panel Course
How to Change Information in a Knowledge Panel
Jason Barnard speaking: Hi and welcome. Huge question: how do you change information in a Knowledge Panel? There are multiple ways to do this, but my number one rule here is to remember that the algorithms ultimately decide, as you’ll see.
Jason Barnard speaking: So, long term, you need to work steadily and solidly towards being in a position to heavily influence the algorithms. I’d love to say control. But as with educating a child, all you can really do is help it, explain, re-explain, have things explained by relevant authoritative third parties, and gradually build its trust in what you are communicating. And by that long process, you’ll have a very, very strong influence.
Jason Barnard speaking: Once again, this ultimately means you first need to make sure that Google understands which page is the Entity Home and then learns to accept that that Entity Home is authoritative for the entity.
Jason Barnard speaking: That said, you can ask humans at Google to change things in a Knowledge Panel. Knowledge Panels have a feedback button. Anyone can click on that and suggest changes. So even if you haven’t claimed the Knowledge Panel, you can suggest changes. If you have claimed the Knowledge Panel, then your suggestions are prioritised and have a higher likelihood of being accepted.
Jason Barnard speaking: Either way, when doing this, you need to click on the pencil icon of the element you want to change or delete, for that matter, and give a short explanation of the change you want, and provide links to at least two web pages that corroborate what you are
saying. One of those pages can be a page you own, but the other needs to be at least partially independent. What you say on your own website is generally not enough. You can also give general feedback, for example, you want to add something.
Jason Barnard speaking: With the explanation you provide, remember that the person at Google looking at your request has never seen this Knowledge Panel before and has no previous understanding of the entity. So, a little context always helps, but make sure your request is brief, clear, and to the point. The Google employees are under pressure to work fast, so they won’t spend a lot of time examining your request. Make it easy for them.
Jason Barnard speaking: The time frame here varies. Anything from a day or two to a month. Once the person at Google has made a decision, you’ll receive an email telling you whether they have applied your suggestion or not. I haven’t seen a case where the request is not dealt with, so don’t keep submitting the same request over and over. Doing that will probably keep pushing you to the back of the queue, rather than bumping you up to the front.
Jason Barnard speaking: Bear in mind that even though the option to make a suggestion is available in the Knowledge Panel, there are some restrictions. For example, Google employees cannot edit the subtitle. They can remove the subtitle, and a new one will be chosen by the algorithm. The title can only be changed if there is a major issue with you, and you have a significant supporting evidence. In 10 years, I have never seen this happen.
Jason Barnard speaking: Humans at Google do not choose the description. That is algorithmic. The best way to get the description you want is, once again, the Kalicube 3 step process and using the same descriptions everywhere on the web. Eventually, you will probably get the description coming from your own website, like we have at Kalicube.
Jason Barnard speaking: Social profiles can be added or removed but not reordered. Some items simply cannot be changed by Google employees. Attributes often cannot be changed, and any information that contravenes Google’s policies will not be changed. You can make suggestions for People Also Search For and images, however Google employees can only remove them and will only do so when the information contravenes their policies. And also note that even here, the algorithm can revert the change.
Jason Barnard speaking: Look in the additional materials, and you’ll find a link to Google’s official help document about manual changes to a Knowledge Panel. A really good use of your time is to read the help documents at support.google.com. Google is surprisingly open about what the processes are and what rules you need to follow. I do encourage you to use the Suggest an Edit option on the Knowledge Panel and ask for additions and changes where you feel the information is important. However, ultimately the algorithm will decide, as we will see.
Jason Barnard speaking: Here’s a story to illustrate. A client came to me to help her correct the Knowledge Panel. She is an author, and the wrong book was appearing in her Knowledge Panel. The book was written by her namesake, another author with exactly the same name. Google’s algorithms had got it wrong and were showing the other author’s book in my client’s Knowledge Panel.
Jason Barnard speaking: She had claimed her Knowledge Panel and suggested the change using the Suggest an Edit option. The Google employee implemented the change as she requested and removed the book. So, why did she come to me? Because the algorithms switched the book back within a week. That book written by her namesake was back in her Knowledge Panel. The algorithms ultimately control. If they think the human got it wrong, they will revert to their truth.
Jason Barnard speaking: So, what is the solution? You need to investigate, play Sherlock Holmes, and figure out why the machine is misunderstanding, and then remove or correct the information that is causing confusion. In this case, we found that both authors were published by the same company, and the publisher website had an author profile page that had the books mixed up. When the publisher corrected their mistake, the Knowledge Panel corrected itself within a month.
Jason Barnard speaking: Now, the reason that the algorithms have ultimate control is because Google is using machine learning to grow the Knowledge Graph and to populate Knowledge Panels. It cannot build an understanding of the entire world using only human curated sources.
Jason Barnard speaking: What Google engineers are doing is to teach the machine to learn how to learn. The machine is being taught to get better and better at figuring out what is fact, what is not fact, what it can show, what it can’t show, which named entity is which, what their attributes are, what the relationships are. It’s absolutely massive. Google engineers simply feed corrective and reinforcement data into the algorithm and tweak the algorithm from time to time. The machine effectively runs itself and makes its own decisions.
Jason Barnard speaking: What do I mean by feeding corrective and reinforcement data? Corrective data is information that the machine got wrong. Providing examples of where it made a mistake helps the machine evolve, helps it learn to learn. The feedback button is one example. Every time a human being at Google changes something in the Knowledge Panel through the feedback, they are giving the machine one example where it made a mistake, but this is one tiny signal to the algorithms, as we have seen. It can choose to ignore that specific correction.
Jason Barnard speaking: Reinforcement data consists of examples of information that the algorithms got right. As with a child, in order to learn, it needs to know both when it got things right as well as when it got things wrong. The feedback button is a miniscule proportion of the corrective and reinforcement data Google is feeding into the machine. We don’t know how much, but you can be sure that the machine regularly receives millions of pieces of corrective and reinforcement information, as well as vast amounts of additional human curated information.
Jason Barnard speaking: So, back to the practicalities for you and the Knowledge Panel you want to change. Suggesting an edit is great, and it works very well as long as you provide solid supporting evidence. Ultimately though, your capacity to change information in your Knowledge Panel is driven by your capacity to educate Google about the entity, the attributes of that entity, and the relationships between that entity and other entities.
Jason Barnard speaking: As always, closer, stronger, and longer relationships are the ones that you can leverage most powerfully. And that means having a strong, authoritative Entity Home, solid corroborative sources, and a solid set of signposts in place is vital.
Jason Barnard speaking: Another example of changing information in a Knowledge Panel I’d like to share with you is the case of a company for whom the Knowledge Panel showed an inaccurate figure for yearly revenues. The client had not communicated that on their site, so Google took the figure from an article about them written by a third party who had gotten the information wrong.
Jason Barnard speaking: We had the Entity Home under control, so we knew we had a good chance of correcting this algorithmically. We wrote a short article about the business figures of the company, including the revenues and some other financial details in a HTML table, added some internal links with an anchor text, financials 2021, from the Entity Home and two other relevant pages. We submitted all of those pages through Search Console. And in two weeks, the information in the Knowledge Panel was corrected.
Jason Barnard speaking: To get a description if there isn’t one or to get the description you want if Google is using one you don’t like, the solution is to take the description you wrote in step 2 of the Kalicube process and duplicate that everywhere around the web.
Jason Barnard speaking: A quick note on exceptions. As of late 2022, it is not possible to replace the information and description when the Knowledge Panel is using Wikipedia, Google Books, Google Scholar, or Google Podcasts as the source. That will certainly change in the future as Google gives the machine more freedom and moves away from that initial core of human curated data.
Jason Barnard speaking: If you want to add attributes, the trick is to provide that information in an HTML table or using a simple semantic triple near the start of your entity description, and then duplicating that around the web. You can also add the information to structured information on trusted sources such as Crunchbase, Wikidata, Wikipedia, IMDb, MusicBrainz, LinkedIn, et cetera, et cetera. At Kalicube, we maintain a list and share some of that data publicly on kalicube.pro.
Jason Barnard speaking: The best way to find which attributes you are likely to add to a Knowledge Panel is to look at what appears for Entity Equivalents. That is what appears for Entity Equivalents of the same type, the same geo region, and the same industry. If an attribute appears for them, you can expect to be able to get it for your entity too.
Jason Barnard speaking: Always remember that Google wants to show useful information about the entity to its users. So, a person’s shoe size is an attribute that only makes sense if they have particularly small or big feet. I know someone who managed to get his shoe size in
the Knowledge Panel. His shoe size is 9. The algorithms removed that attribute pretty quickly after it appeared.
Jason Barnard speaking: Description sources, attributes, People Also Search For, social channels, in short anything that appears in a Knowledge Panel is what we track at scale on the Kalicube Pro SaaS platform so that we know where to place information, so it is picked up by Google and also what information Google tends to show for specific entity-geo-industry combinations. Entity Equivalents is a hugely powerful concept. Spying on the competition always makes sense.
Jason Barnard speaking: To finish this lesson, I want to reiterate one last time: the algorithms ultimately decide what is displayed in a Knowledge Panel. So if you want any level of control long term, then you need to focus on an Entity Home, a clear presentation of the information that matters to you, and authoritative corroboration.
Jason Barnard speaking: Thank you so much for watching. I’ll see you soon.
