Google Knowledge Panel

Somebody hears your name in a meeting. They type it into Google that evening. What comes back in the next ninety seconds is, for most business relationships, the most consequential piece of content you will ever be involved with, and it is the one piece you almost certainly have never deliberately designed.

The Knowledge Panel is the most visible part of that answer. It is the box on the right of the search results, with a photo, a one-line description, and a set of facts Google is asserting about a person. When it appears, Google has stopped listing links about you and started stating who you are.

This page is the decision version. What the thing actually is, what it is honestly worth, what it costs, and the cases where the right answer is to keep your money. If you want the build mechanics instead, the companion manual on the execution site walks through the doing.

01What a Knowledge Panel actually is

Google maintains a Knowledge Graph: a structured map of entities, which are people, places, organisations, works and concepts, and the relationships between them. A Knowledge Panel is a rendering of one entity in that graph. When Google decides a query is about an entity it holds with enough confidence, it draws the panel.

Three consequences follow from that definition, and each one contradicts something people commonly believe.

1. There is no application. No form, no fee, no queue, no relationship at Google to buy. Anyone selling you expedited access is describing something that does not exist.
2. It is a confidence output, not a reward. Google is not deciding whether you deserve a panel. It is deciding whether showing one is safer than not showing one. Those are different questions with different answers.
3. You do not write it. Even after you claim a panel, you suggest edits and Google reviews them. The description line is assembled from what the web says, not from what you would like it to say.

02The three things people confuse it with

A large share of the conversations we have start with somebody describing something that is not a Knowledge Panel. Worth clearing up before anything else, because the fix for each one is different.

03The two stages nobody explains

This is the single most useful thing on this page, and it is missing from almost every article on the subject. A panel does not switch on. It arrives in two stages, and the gap between them is where most people conclude the work failed.

• Stage oneThe entity record goes live

  Google has resolved your scattered records into one entity and given it an identifier. Search that identifier and a card comes back. Search your bare name and you may still get nothing. This typically lands somewhere between two and five months into a real build.

• The gapNothing visibly changes

  The record exists. The assistants can already read it. Your name search looks identical to how it looked in month one. This is where people stop, and stopping here is how the work gets wasted.

• Stage twoThe panel fires on the bare name

  Google is now confident enough to show the panel when somebody types just your name. Typically six to twelve months from a standing start, longer on a contested name, and no date can be promised.

We publish four of our own records that are sitting at stage one right now, alongside four at stage two. Showing only the finished ones would make a better advert and a worse argument, and anybody in this field showing you nothing but finished panels is either very selective or very new.

Source: Our own eight client records, read live from Google on 31 August 2026

04What it is actually worth, without the marketing

The honest business case is not the box. The box is a visible symptom. The thing worth paying for is the underlying record, and it is worth paying for because of something that changed in the last two years.

Until recently, being findable meant ranking. Somebody searched, got ten blue links, and clicked. Now two things happen instead. Google increasingly answers the question directly, drawing on entities it holds rather than on pages it ranks. And a growing share of the same questions never reach Google at all, because they are asked of an assistant.

Both of those read from the same kind of source: crawlable, attributable, third-party text about you. The Knowledge Panel and the assistant answer are two outputs of one input. That is the real argument, and it is a better argument than the box.

Source: Ahrefs, 863,000 keyword SERPs and 4 million AI Overview URLs

You are not buying a box on a search page. You are buying the thing that answers when you are not in the room.

There is a second, less obvious value, and founders tend to feel it before they can name it. A panel is a disambiguation device. If you share a name with a cricketer, an academic and a man who was arrested in 2019, the panel is the mechanism by which Google tells the world which one you are. For anyone with a common name, that is worth more than the prestige.

05What it does not do, stated plainly

Four things it will not do. We would rather you read these now than discover them in month seven.

On schema, since it gets sold as an AI product. Ship structured data, because it earns rich results and makes your records machine-readable. Do not buy it as an AI visibility deliverable. A controlled study of 1,885 pages that added JSON-LD against 4,000 control pages found AI Overview citations went down 4.6 percent. The reason people believe otherwise is that 53 percent of AI-cited pages carry schema, which is correlation: competent sites do many things right at once.

Source: Ahrefs, 1,885 pages that added JSON-LD against 4,000 controls, four methods

06Who actually gets one

Google does not publish an eligibility list, which means everything written on this subject is inference from observation. Here is ours, from the builds we have run and the ones we have declined.

• You can name three things you have done that a stranger could verify without your help
• Somebody not on your payroll has written about or interviewed you in the last two years
• There is a domain with your name on it, or you are willing to buy one
• You will sit for content: podcasts, talks, written pieces, five hours a month
• You have one Google account you will still control in five years

If four or five of those are true, a build is a reasonable purchase. If two or fewer are true, the honest answer is that you are not buying a Knowledge Panel, you are buying the eighteen months of public work that has to happen first, and you should decide whether you want that on its own terms.

07What it costs

Three honest routes, and the cheapest one is genuinely viable for a lot of people.

Notice that the hours column barely moves between hybrid and full. That is not a pricing trick, it is the shape of the work. About thirty-five hours are structurally yours: deciding your identity, sitting for the content, holding the Google account, and submitting the claim with your own documents. Nobody can take those, and anybody offering to take the last one is asking for papers nobody should hand over.

On the paid build specifically: half at intake, half the day there is an entity identifier or a live card to show you. No monthly retainer and no upsell ladder. Four builds run at a time, because the part that takes the months is third-party corroboration and that cannot be parallelised by hiring more people.

08The build order, and why the order matters

Twelve steps. The sequence is not decorative: each step assumes the previous one is true, and skipping is the most common reason a build stalls at month five with nothing to show. The full mechanics live on the companion manual; this is the shape.

Source: Our own eight client records, read live from Google on 31 August 2026

1. Decide the identity. Three sentences: who you are, what you stand for, what result you drive and for whom. Plain language. If these cannot be answered, nothing downstream has anything to broadcast.
2. Build the brand brain. One private file holding every fact anyone will state about you: bios at four lengths, dates, registrations, the master headshot, transcripts, case studies with names and results.
3. Stand up the entity home. One site on your own name, separate from the company site and linked both ways, with the identity line as the first paragraph in plain text and the headshot above the fold.
4. Ship the schema. Person on the personal site, Organization on the company site, wired to each other. This is the machine-readable layer you fully control.
5. Align every profile. Same name string, same headshot file, same first eight words of headline, same city, same first bio line, one link format. Two unglamorous hours with the highest return in the project.
6. Instrument it. Search Console verified on an account you will still control in five years, because the same account is needed for the claim and every future edit.
7. Manufacture real proof. Two cornerstone pieces a month, one case study, weekly clips. Published to your own site first, then mirrored outward.
8. Densify the cross. Independent witnesses at the intersection of your trade and your city. Three to five of each. This is the part that takes months and the part that cannot be faked.
9. Add the accelerators. Authorship with an ISBN first, then Wikidata only if the notability bar is honestly cleared, then peer proximity through podcasts.
10. Claim it. When the panel appears, you claim it, from your own account, with your own documents.
11. Lock the entity. Panel appears on brand-only queries, verified edits land in days, off-brand images are gone, you consistently beat namesakes.
12. Run the cadence forever. Weekly review, monthly assistant baseline, quarterly audit. Drift is silent and it is the enemy.

09The geo-vertical cross, which is where the months go

If you take one structural idea from this page, take this one. Independent corroboration has two axes, and the value is at the intersection rather than on either axis alone.

The vertical axis is your trade: peers, suppliers, associations, conferences, industry podcasts. The geographic axis is your place: local press, chambers, allied trades, charities, civic bodies. A person who is named by five people in their trade looks like somebody with a good publicist. A person named by five in their trade and five in their city looks like somebody who actually exists and actually does the thing.

That second pattern is the one spammers do not bother to fake, because faking it is more expensive than earning it. Which is precisely why it works.

One India-specific caveat, since most of our clients are here. The standard playbook for this step is written for American local-service businesses, where chambers of commerce, city journals and local press all publish crawlable pages. In India that geographic layer is much thinner online. The substitution set is different: industry associations, trade bodies, regional business press, university and conference pages, and Indian-language publications that Google crawls but most agencies ignore.

10The claim, and the field that gets people rejected

When a panel appears, claiming it takes about ten minutes and one field decides the outcome. The field asks why you are claiming the panel. Almost everybody answers the wrong question.

Also watchHow I got my verified Google Knowledge Panel, step by step (part 1 of 3)4:04

One hard line, and it is not negotiable. The claim requires identity verification, and that verification is yours. We prepare the packet and rehearse the field. You submit it. We do not take identity documents, we do not store them, and we do not log into your account. Anybody willing to do those things for you is a risk you are hiring.

11Namesakes, and the case where we say no

Namesake collisions are the most under-discussed failure mode in this field, and they are common in India where a shared surname plus a common first name can put several hundred people in the same string.

There is a soft version and a hard version. The soft version is that other people with your name rank above you, which more record fixes. The hard version is a live collision: Google serves a different person's panel for the same query on different days, because it holds two entities it cannot cleanly separate.

We have watched exactly that happen, twenty minutes apart, on somebody we know well. When that is the state, more content makes it worse rather than better, because you are adding weight to a query that is already ambiguous. The fix is a different and longer project, and we will say so before taking money rather than after.

12What we refuse to do

Worth stating, because most of these are on sale somewhere.

• We will not promise a panel, or a date. Google offers no guarantee, no application and no service level, and neither can anyone reselling it.
• We will not write or buy a Wikipedia article. A deleted page leaves a public record of the deletion and leaves you worse off than never trying.
• We will not handle your identity documents or log into your Google account.
• We will not quote the Knowledge Graph match score as a Google trust metric. It is a query-relative relevance number and treating it as a confidence rating is the most common piece of misinformation in this field.
• We will not take the engagement if the entity is not ready. Refusing a doomed build is cheaper for both of us than a refund in month eighteen.

13Our own record

Eight records, four at stage two and four at stage one, every one of them a real person you can look up right now. The stage-one four are published deliberately.

One of the eight is labelled worked on rather than built, because that person already had a partial record when we started and claiming otherwise would be a lie that is easy to check. One further client is deliberately excluded from this table: his entity is in a live namesake collision with an academic of the same name, and unstable proof is not proof.

14The nine diagnostic queries

Before you buy anything from anyone, run these. They take twenty minutes in a private window and they tell you which of three completely different problems you have. Most people who think they need a Knowledge Panel build actually have problem two, which is cheaper and faster to fix.

Score it honestly: how many of the nine return results that are about you, accurate, and that you would be happy for a client to read. Under four and the record genuinely needs building. Four to seven and you probably have a correction job rather than a construction job. Eight or nine and you should be claiming a panel rather than commissioning one.

15The assistants are now the same problem

Two years ago this page would have been about Google alone. It is not any more, and the reason is worth understanding because it changes what you are buying.

When somebody asks an assistant who the best person is for a given problem, the model is not checking your engagement rate or your follower count. It is drawing on text about you that exists in crawlable, quotable, attributable form. A LinkedIn post is a poor source for that. A published interview on a domain with a history is an excellent one.

Which means the Knowledge Panel work and the assistant-visibility work are not two products. They are one input with two outputs. Anybody selling them to you as separate line items is either confused or counting on you being.

Source: Ahrefs Brand Radar, 1.9 to 4 million queries per platform, read 25 Aug 2026

Two readings of that chart that matter for how you spend. First, social and user-generated surfaces dominate almost everywhere, so material that lives only on your own site earns very little citation share. Second, and this surprises people: the New York Times sits 35th on AI Overviews at 0.4 percent mention share. A tier-one press placement bought specifically for AI visibility is buying prestige rather than citations. It may still be worth buying. Just not for that reason.

Also watchHow LLMs recommend you and your brand: clarity and transparency on LLM SEO3:24

16How to compare us against anyone else

You should get other quotes. Here is what to ask them, and what the answers tell you. These are the questions we would want asked of us.

1. Ask them to guarantee a panel. If they will, walk. Google offers no application, no guarantee and no service level, so a guarantee of the outcome is either ignorance or a sales tactic. A guarantee of a refund is a different and legitimate thing.
2. Ask what the Confidence Score means. If they present it as a Google trust metric, they have read a blog post rather than the documentation. It is resultScore from the Knowledge Graph API and it is query-relative: it measures how well an entity matched the string you typed, not how much Google trusts the entity.
3. Ask whether they will write your Wikipedia article. If yes, walk. A deleted article leaves a public deletion record and leaves you materially worse off than never trying.
4. Ask whether they need your ID documents or your Google login. If yes, walk. The claim is the owner's own hands, always, and there is no version of this work that requires handing over identity papers.
5. Ask to see a client whose panel has not fired yet. If every case study is a finished panel, they are either very selective about what they publish or very new. Most of this work, most of the time, looks like a stage-one record.
6. Ask what happens in month six if it is not working. The right answer involves changing the corroboration strategy or stopping, not publishing more.

We publish four unfinished records for exactly this reason. It is a worse advert and a better argument, and if a competitor will not do the same, that tells you something.

17A real twelve months, month by month

What an engagement actually feels like from the inside, including the two months where it feels like nothing is happening, because those are the months people quit.

• Month 0The verdict

  We run the nine queries, the Knowledge Graph lookup, the namesake count and a three-assistant baseline before you pay past the deposit. If it comes back saying wait, we say wait and you keep the rest.

• Months 1-2Identity and the home

  The three sentences, the brand brain, the entity home, the schema. This is the part that feels like progress, because things visibly appear.

• Months 2-4Alignment and instrumentation

  Every profile made to agree. Search Console verified. Unglamorous, fast, and the highest-return work in the whole engagement.

• Months 3-6The quiet part

  Witness work. Asking, waiting, following up, waiting. Nothing visible changes on your name search. This is where people conclude it failed, and it is where the outcome is actually being built.

• Months 4-8Stage one

  An entity identifier appears. Search it and a card comes back. Search your bare name and it may still be nothing. This is the milestone, and it does not look like one.

• Months 6-12Stage two, usually

  The panel starts firing on the bare name. Sometimes it flickers first, appearing on some searches and not others, which is normal and resolves as the record thickens.

• Month 12Claim and hand over

  You claim it from your own account. We hand over the brand brain, the cadence and the measurement set, and you can run it yourself from there.

The two months in the middle are the honest reason this is a twelve-month engagement rather than a ninety-day one. If somebody tells you a panel in ninety days, they are either describing the stage-one record and calling it a panel, or they are describing something that did not happen.

18What arrives in week zero, before you commit

The first thing we send is a verdict, not a proposal, and it arrives before you have paid anything past a deposit. It is deliberately the part of the engagement most likely to lose us the sale.

• The nine diagnostic queries, run signed out, with screenshots and a score out of nine
• The Knowledge Graph lookup: whether an entity already exists for your name, and its identifier if so
• The namesake count, and whether any of them is a live collision rather than an ordinary competitor
• A three-assistant baseline: nine prompts across ChatGPT, Claude and Perplexity, scored, dated, saved
• A route: full build, correction job, or wait, with the reasoning written down rather than asserted
• If the answer is wait, what specifically would have to change for it to become go

About a third of the time the honest route is not the full build. Sometimes an entity already exists and the job is correction, which is shorter and cheaper. Sometimes there is a live namesake collision and the correct advice is to solve that first. Occasionally the honest answer is that the underlying record is too thin and the money is better spent on the eighteen months of public work that has to happen anyway.

We would rather say that in week zero than in month seven. It is also the cheapest possible test of whether the person you are talking to is selling a method or selling hope, and you should apply it to everyone you speak to, including us.

19What happens if you do nothing

A fair question, and the honest answer is: for many people, not much, for a while. Then two things tend to happen at once.

The first is that somebody else with your name builds a record and takes the query. This is not hypothetical and it is not reversible on a short timescale. Entities consolidate, and the one with the deeper record wins the bare name.

The second is that assistants start answering the category questions in your field, and they answer from sources that do not include you. Being absent from those answers is not neutral, because the answer still gets given. It just names somebody else.

The question is not whether an answer about you exists. It does. The question is whether you had any part in building it.

20What a namesake collision actually costs

Worth its own section, because it is the one situation where the standard advice is actively wrong and where we most often tell people not to buy.

The ordinary version is that other people share your name and some of them rank above you. That is a volume problem and more record fixes it. It is annoying and it is not dangerous.

The dangerous version is a live entity collision. Google holds two entities it cannot cleanly separate and serves one or the other depending on the query, the day, and the data centre. We have watched Google serve two different people's panels for the same name twenty minutes apart. When that is the state, publishing more about yourself adds weight to an already ambiguous query and can make the blending worse rather than better.

The distinction matters commercially. The ordinary version is a normal build. The live collision version is longer, less predictable, and we will say so before taking money rather than discovering it together in month seven. One of our own clients is deliberately excluded from the published record for exactly this reason.

21How to start

Do not start by buying anything, including from us. Start with the ninety-second diagnosis, because it changes which of three routes you are on and one of the three is much cheaper than this page.

1. Search your full name in a private window. Screenshot page one. Note whether a panel appears and whether the results are about you.
2. Look your name up in Google's Knowledge Graph. If an identifier comes back, you already have an entity and your job is correction, not construction, which is a shorter and cheaper job.
3. Ask ChatGPT, Claude and Perplexity who you are and what you are known for. Save the answers with today's date.
4. Route yourself honestly. Clean panel showing means claim it this week and stop reading. Wobbly or wrong panel means fix the record. Nothing at all means a full build, and ninety days is the floor before anything is visible.
5. If you are doing it yourself, the day-by-day calendar and the copy-paste toolkit are both free and linked below. If you want it built, the intake is an email, not a form.

Either way, re-run step one in ninety days with the same wording. The delta is the only measurement in this whole field that means anything, and almost nobody has a baseline to compare against.