Knowledge graph optimization

TL;DR

Knowledge graph optimization turns a brand or a person from a string of characters into an entity that Google, Claude, and Perplexity all recognize. The three moves that matter: earn a Wikidata QID with clean citations, ship sameAs coverage across the homepage schema, and claim the Google Knowledge Panel when it appears. Without a QID a brand is guessed. With one it is known.

The surface, in one paragraph

Knowledge graphs are structured databases of entities and their relationships. Google's Knowledge Graph powers Knowledge Panels, answer boxes, and a growing share of AI answer citations. Wikidata is the open-source graph that seeds it, and Wikipedia is the human-readable article surface on top. A brand or person exists inside these graphs when their entity has a unique identifier (a QID), when reliable sources describe them, and when downstream systems can trace machine-readable links from the entity back to authoritative pages. Once the entity exists, search engines stop guessing which "Apple" the user meant, AI answer engines stop confusing two people with the same name, and the brand shows up in structured surfaces (Knowledge Panels, Bing entity cards, ChatGPT citations, Perplexity source blocks) that a brand without a QID cannot reach.

Where this fits in Search Everywhere Optimization

KGO underpins several other surfaces.

SEO gets Knowledge Panels. Branded queries with a Knowledge Panel take up half the screen. That is real estate no ad can buy.

AEO cites entities, not strings. Claude and ChatGPT prefer to cite named entities that resolve to Wikidata or Wikipedia. Brands with clean KGO become preferred citations.

GEO builds on the same graph. Generative engines use knowledge graphs to disambiguate people, places, and organizations. A QID prevents mixups.

VSO and VxSO need entity clarity. Assistants and Lens both call the Knowledge Graph before answering. If the brand is not there, the answer defaults to a competitor or Wikipedia.

E-E-A-T signals compound. Author bios with sameAs to Wikidata, LinkedIn, and Google Scholar (where relevant) tell every engine that the author is a real, verifiable person.

The five levers

1. Wikidata QID with clean citations

The QID is the anchor. I create the Wikidata item with clear notability signals: press coverage, book publications, board seats, patents, funding announcements from reliable outlets. Every claim on the item cites an external source. I avoid promotional language. I add labels in the main languages the brand operates in, aliases where the entity has multiple names, and structured claims for country, industry, founding date, founders, and headquarters. If the brand does not meet Wikidata's notability bar, I do not fabricate one. I wait until the coverage exists.

2. sameAs coverage across the homepage schema

Organization schema on the homepage. Person schema on the About page. Each with sameAs links to: Wikidata QID URL, Wikipedia URL if it exists, LinkedIn company/personal page, Crunchbase, official YouTube channel, official social profiles, GitHub for technical entities, Google Books or Amazon Author Central for authors. The sameAs cluster tells Google which profiles all belong to the same entity, which speeds Knowledge Panel resolution.

3. Wikipedia article where notability supports it

Wikipedia has a stricter notability bar than Wikidata. If the entity qualifies (two or more independent reliable-source articles that discuss the entity substantively, not press releases), I support a community editor's article creation. I never edit the article about the client's own entity: it violates policy and reads as a conflict of interest. I supply source lists to editors. I let the community decide.

4. Knowledge Panel claim and management

Once the panel appears, I claim it. Google's "Claim this knowledge panel" flow requires an official social or YouTube account tied to the entity. Once claimed, the entity owner can suggest edits, upload a preferred logo image, and get correction requests reviewed faster. Panels without a claimed owner get stale and inaccurate over time.

5. Persistent identifier discipline

Beyond Wikidata, I collect and cross-link every persistent identifier the entity qualifies for: ORCID for researchers, ISNI for authors, VIAF for library-catalogued entities, LEI for financial-market entities, and DUNS for suppliers. Each ID goes into Wikidata as a claim. Each ID appears in sameAs on the homepage. Persistent identifiers are the mortar between the entity's identity and every database that references it.

First 30 / 60 / 90 days

Days 1 to 30: audit and claim inventory

Entity audit. Does a Wikidata item exist. Does Wikipedia have an article. Does a Google Knowledge Panel appear on brand-name queries. Is it claimed. What does it say. Is the info correct.

Notability audit. Which independent reliable-source citations exist. Press coverage from named outlets, industry awards, patents, published works, funding announcements. This determines what Wikidata claims I can create with real citations.

sameAs inventory. All official profiles across LinkedIn, YouTube, Crunchbase, GitHub, Instagram, X, TikTok, industry directories. Any dormant or brand-inconsistent profile gets flagged.

Persistent identifier check. ORCID, ISNI, LEI, DUNS, VIAF. Which the entity qualifies for. Which are missing.

Deliverable at day 30: an entity brief, a notability citation library, a sameAs inventory with cleanup list, a Wikidata plan (create, edit, or wait), and a Knowledge Panel status report.

Days 31 to 60: create and connect

Wikidata item created or improved. Labels, descriptions, aliases, and structured claims added with citations. If the entity already exists but has stale or incorrect claims, they get corrected with source citations.

sameAs schema shipped. Organization schema on the homepage. Person schema on the About page. Each with the full sameAs cluster.

Persistent identifier applications. ORCID for researchers, ISNI for authors, LEI for financial entities. Each ID goes into Wikidata claims.

Wikipedia editor outreach if notability supports an article. Source list shared with volunteer editors. No direct editing.

Knowledge Panel claim attempt if the panel already exists. Verification through official YouTube or LinkedIn.

Deliverable at day 60: live Wikidata item with citations, shipped sameAs schema, applied-for or received persistent IDs, submitted Wikipedia editor request, filed Knowledge Panel claim.

Days 61 to 90: monitor and expand

Knowledge Panel appearance check. If the panel is now live, verify all facts are correct. Submit correction requests through the claim owner flow.

Wikidata item quality review. Add more claims as citable events occur (new funding, new hires, new products, new awards).

AI answer engine spot-check. Query Claude, ChatGPT, and Perplexity on the brand and its category. Does the brand appear. Is it cited correctly.

Content expansion. Long-form pages that reference the brand's key entities (people, products, events) with schema markup that ties back to the Wikidata QID.

Deliverable at day 90: a maintained Wikidata item, a claimed and accurate Knowledge Panel where earned, verified AI-engine visibility, and a quarterly maintenance cadence.

Tools I use

• Wikidata. Item creation and editing. QID assignment. Structured claims with references.
• Wikidata Query Service (SPARQL). Verify entity connections and relationships across the graph.
• Wikipedia. Notability check via WP:GNG and WP:NCORP guidelines. Editor community outreach.
• Google Knowledge Panel claim flow. Ownership verification and correction submission.
• Google Rich Results Test. sameAs and Organization schema validation.
• Schema.org validator. Deeper structured data checks.
• OpenRefine. Reconciling brand data against Wikidata.
• ORCID, ISNI, GLEIF, VIAF. Persistent identifier registration.
• Crunchbase and PitchBook. Source citations for company claims.
• Ahrefs and Perplexity. Reverse-check which entity Google associates with a brand-name query.

What kills the program

Editing the client's own Wikipedia article. Conflict of interest violation. Article gets tagged, edits reverted, brand loses editor goodwill.

Promotional Wikidata claims. "Leading provider of" is not a claim, it is marketing. Wikidata patrols delete promotional entries fast.

Manufactured notability. Press-release-only citations do not count. Two paid interviews on a podcast do not count. Real notability comes from independent, reliable, secondary coverage.

Ignoring the Knowledge Panel once claimed. Facts go stale. Product lines change. Founders leave. If nobody maintains the panel, the entity looks abandoned.

Inconsistent sameAs. LinkedIn URL points to the wrong slug. YouTube channel handle changed. Old Twitter handle still on the schema. The engines lose confidence.

Confusing similar entities. Two "Frederick Sona"s in the graph. Two "Acme"s. Without disambiguation claims (occupation, birth date, industry, location), the wrong entity gets referenced.

Chasing a panel before the entity qualifies. Some entities do not meet notability. Grinding away at panel claims for a nonexistent panel wastes time. Build coverage first.

KPIs that matter

• Knowledge Panel presence on brand queries. Binary. Does the panel appear.
• Knowledge Panel claim status. Claimed and verified, or unclaimed.
• Wikidata item completeness. Count of claims with references.
• sameAs cluster size. Number of profiles cross-linked in the entity graph.
• AI answer engine citation rate. Manual sampling: query the top ten branded and topical queries, count citations.
• Persistent identifier coverage. Which IDs the entity holds, cross-referenced across databases.
• Wikipedia article presence. Binary if applicable, plus edit-war health if the article exists.

FAQ

What is a QID?

A QID is the unique identifier a Wikidata item receives, like Q42. Google's Knowledge Graph, AI answer engines, and structured data validators all use QIDs to disambiguate entities. Earning one is the first step of KGO.

Can I create my own Wikidata entry?

Yes, but only if the entity meets Wikidata's notability policy. I create entries with clear source citations, avoid promotional language, and disclose conflicts of interest. Improper entries get deleted and hurt the brand.

How do I claim a Google Knowledge Panel?

Search for the entity in Google, click 'Claim this knowledge panel' at the bottom of the panel, and verify identity through an official account (LinkedIn, YouTube, verified social).

How long does a Knowledge Panel take to appear?

After Wikidata is clean and sameAs links propagate, panels usually appear in 4 to 12 weeks. For brands with existing Wikipedia articles the panel is often already there.

Does sameAs schema on my site help?

Yes. sameAs links inside Organization and Person schema on the homepage tell Google which social profiles, Wikidata QID, and Wikipedia URL belong to the same entity. It is the cheapest KGO move available.

What if the entity is not notable enough for Wikipedia?

Wikidata's notability bar is lower than Wikipedia's. Many brands qualify for Wikidata but not Wikipedia. Panels can appear from Wikidata alone if sameAs coverage and structured data are strong.

Related reading

• E-E-A-T playbook: entity trust as an author signal.
• Voice search optimization (VSO): assistants call the Knowledge Graph before answering.
• Visual search optimization (VxSO): Lens attribution flows through the entity graph.
• Feed and Discover optimization (FEO): publisher and author entities gate Discover.
• International optimization (GLOBO): multilingual labels feed the same graph.