Back

Knowledge Panel Optimization: Making Your Brand Say the Right Things in an Entity Card

Tengence Team · ·
GEO/SEO Entity & EEAT Schema Markup
Knowledge Panel Optimization: Making Your Brand Say the Right Things in an Entity Card

Summary: A knowledge panel is generated automatically from Google's knowledge graph. You cannot decide whether it appears; you can influence what it says through three movable positions: structured markup, facts that live at a citable URL, and independent third-party corroboration. This article gives the order, the pass/fail test for each step, and three failure cases — plus the expectation to set with your boss first: median time from complete facts to a stable panel is about two months, not two weeks.

Key Takeaways

1. What a panel is and is not

Strip the misconceptions first. It is not a ranking result, not a rich snippet, not an account entitlement you can switch on. It is an automatically generated entity card that answers "what is this thing" for organisations, people, places and things.

Its fields fall into three rough layers: hard structured facts (official site, founding date, headquarters, founders), a description paragraph aggregated from public content, and an image area. The image area is the most misread — some images come from public image sources once the entity is claimed, others are previews of the entity from Google Images. Changing your hero image does not guarantee the panel shows only that.

A reality check: panels disappear. Google's documentation has a section titled "Knowledge Panel No Longer Displays", stating that when and on which queries a panel appears is determined by automated systems, and that this is neither something the company influences nor tries to. There is no public timetable and no appeal channel. What you can do is keep a record: for the same entity and the same fixed set of queries, log the display scope, which sources are cited, whether the description changed — and compare month over month. A single screenshot proves nothing because the panel is not permanent. That paragraph exists precisely for this situation: it is not a failure, it is re-evaluation.

Worth noting too: the description paragraph is not an "official introduction". It is a summary of public content. You cannot edit that sentence directly; you can only make more credible sources describe you the same way. Trying to insert marketing copy into the description does not work. Getting five independent sources to state clearly what you do and whom you serve does.

So the first task is not optimization but expectation calibration: a panel is a probabilistic asset, not a switch.

Want to know how much of the ChatGPT and Perplexity answers your brand owns?

Explore TENGENCE GEO

2. Three input types: your leverage sits at two ends

Google divides graph facts into three sources, and that division determines which actions have any effect.

Input type Provided by Can you influence it Action
Public facts on the open web Any crawlable page, organisation pages, bylined reporting Indirectly, by building facts Structured markup, facts at stable URLs, independent sources
Licensed partner data Data vendors (film, music, sports scores and similar verticals) Essentially no None available
Corrections after claiming The entity itself Yes, factual fields only Claim the panel, submit field-level corrections

The middle column is a dead end for nearly every B2B company: you have no licensable vertical dataset and no reason to negotiate one. Effort therefore goes into the first and third blocks, and the third depends on the first. What you submit after claiming is essentially "I confirm this fact is correct" — if the fact does not exist on the open web, there is nothing to confirm.

The most frequently skipped action in the first block is entity connection. Google's documentation repeatedly cites Organization markup, one purpose of which is telling the search engine which logo belongs to the entity. sameAs is more fundamental: it is a list of other identities belonging to the same entity. Without it, machines cannot merge your presence across platforms, and the graph holds only isolated pages.

We put this action first not because it is advanced but because it is cheap. Ship the markup once and every subsequent external fact stacks onto the same entity. Do it the other way — pay for coverage now, add markup six months later — and those mentions wait for the next recrawl to merge, wasting the interval.

3. The four-step sequence

The order cannot be reversed, because each step's test depends on the previous step's output.

  1. Add Organization markup and connect sameAs. Cover at minimum the official site, the company LinkedIn page, GitHub or Crunchbase, and your primary industry directory listing. Test: the rich results test runs without errors, and every URL in sameAs returns 200.
  2. Lift facts out of prose into addressable URLs. Legal entity name, registration jurisdiction, founding year, service scope and public price bands each need a position that can be cited directly — not a homepage carousel, not the fifth screen of an About page. Internally we call this "URL-ising the facts"; it is the precondition for anyone external to cite you.
  3. Build independent corroboration. Priority order: industry association annual reports, bylined trade press, public corporate and licence registries, and encyclopedia entries last. One test: can a third party verify this mention, or did we simply publish a sponsored post?
  4. After launch, monitor fields, not existence. Once a panel appears, watch three things: whether the site link points at the right language version, whether social profiles from a same-named company have bled in, and whether founder or address data is stale. This step is the most skipped and the most damaging.

Step 4 in practice means maintaining a field ledger: every displayed field, its source URL, and the date you confirmed it, reviewed quarterly. The payoff is that when a field suddenly changes you can tell immediately whether your own data changed or an external source did.

4. Three failure cases and our measurement position

Failure one: betting everything on an encyclopedia entry. An encyclopedia entry is human-edited; a panel is an algorithmically aggregated card. The first can be created; the second can only be influenced. Bet wrong and the symptom is "the entry was approved, the panel still does not exist".

Failure two: chasing mention count. The criterion is not how many mentions but whether each carrier can be verified: fixed editorial process, named author, public corrections policy, traceable archive. Thirty scraped mentions and four from trade press plus an association report are not the same order of magnitude in the system's eyes — the former lacks "who is accountable", the latter has it. Volume barely registers in this mechanism; credibility does.

Failure three: watching one engine only. Bing, Baidu and generative engines each build their own entity understanding. The correct order is to clean your own structured markup first, then build external facts, then adapt per platform — because the first two steps are portable across all of them.

You can verify without tooling: fix four engines, prepare six to eight questions real users would ask, run one pass in a single time window, and record per question whether you were cited, who was cited, and the exact wording. Re-run the same set two weeks later and compare. The value is not in absolute numbers but in direction — holding steady, or drifting toward a competitor.

5. Boundaries: what cannot be done, and how long it takes

Three things must be stated up front.

You cannot decide whether the panel appears. Google's wording is that automated systems judge it. Any promise of "guaranteed appearance" is a promise from someone who has not read the documentation.

You cannot rewrite the description paragraph. It aggregates public content. You can get more credible sources to describe you the way you want; you cannot edit the sentence.

Timelines are counted in months. Across our projects the median is two to three months, with individual cases past four. A proposal promising a panel in one month can be rejected on sight.

One more empirical note: if nothing appears within three months, do not add budget for external coverage yet. Go back and check whether the markup was actually indexed — confirm parsing with the rich results test, confirm crawling with a site query, then check whether any sameAs URL 404s. On two of our projects this step is how we found "markup shipped equals markup absent": one was blocked by noindex, the other had the markup injected into the wrong template route.

FAQ

Q: Can we have a knowledge panel without a Wikipedia entry?
Yes, and many B2B companies have no entry. Graph facts come from multiple public sources; an entry is one type of source, not a prerequisite. What you need is a sufficient set of independently verifiable public facts.

Q: What does claiming a panel require?
You must be the entity or its official representative and pass verification, after which you can suggest corrections. Ordinary users submit suggestions through the feedback link at the bottom of the panel; the two channels differ.

Q: The site link in the panel points at the wrong language version. What now?
First confirm that url in your Organization markup and the sameAs entries point at the intended language version, then wait for a recrawl. If it is still wrong after a week, submit a correction through the panel feedback channel. Do not game it by changing site redirects.

Q: How should multi-brand and subsidiary structures be handled?
One set of Organization markup and one sameAs list per legal entity, with parentOrganization and subOrganization declaring the relationships explicitly. Mixing them into one markup makes the graph treat them as a single entity.

Q: Once a panel exists, how often should fields be reviewed?
Quarterly to start; immediately in any month with a rebrand, domain change or entity change. A panel with wrong fields is worse than no panel, because it shows the wrong information on your brand query.

Related Reading

Data Sources

  1. About knowledge panels — Google Knowledge Panel Help — official definition and what owners can and cannot influence.
  2. How Google's Knowledge Graph works — Google — how facts are collected and merged into entities.
  3. Establish your business details with Google — Search Central — the markup route for organisation facts.
  4. Organization — schema.org type definition — the sameAs and parentOrganization properties in detail.
← Previous: Enterprise Search Deployment Cost: A 3-Year TCO Breakdown Next: GEO for Education and Training Providers: Getting Recommended During Admissions Season →
Consult
QR code

WeCom