MLforSEO Academy · The AI-powered Entity Audit
Knowledge Graph Reconciler
Match your extracted entities against Google's Knowledge Graph, and flag the ones where the returned type contradicts the type your content implies.
- What you need
- The same Google key, with the Knowledge Graph Search API also enabled
- Cost
- Free — 100,000 lookups a day
- Feed it
- entity_mentions.csv from the Extractor, or a pasted list
- You get back
- kg_reconciliation.csv — matched, ambiguous, unmatched, with type conflicts flagged
How to use it
- Paste your key and upload the entity export.
- Leave the match-score floor at 20 unless you have a reason to change it.
- Read the three verdicts before you read a single row.
- Filter to type conflicts — those are matches that are confidently wrong.
- Export, and keep the file for the Scorecard.
Worth knowing. A returned match is not a correct match. Check the type and the description before you trust the score.
MLforSEO · The AI-powered Entity Audit
Knowledge Graph Reconciler
Match your extracted entities to Google’s Knowledge Graph, and find the three ways it goes wrong: no match at all, an ambiguous match, or a confident match to the wrong thing.
1 · Your API key
Held in this tab only. 100,000 lookups a day are free.
The Knowledge Graph Search API is a legacy service. Google directs new
projects to Cloud Enterprise Knowledge Graph and warns it is “not suitable for use as a
production-critical service”. Fine for auditing; do not build a live product on it.
2 · Your entities
Paste one entity per line, or upload the
entity_mentions.csv from the Entity Extractor — that adds NLP types and MIDs, which unlocks type-conflict and over-linking detection.
Or paste below.
Add a key and some entities to begin.
Reading resultScore correctly: it is a relevance score for this lookup —
how well the entity name matched a graph node. It is not authority, not popularity, and not a
ranking signal. A high score means “confident match”, nothing more.
Reconciliation
Suspected over-linking
Same MID, divergent wording
These entities were given the same Knowledge Graph MID by the NLP API despite
describing different things. An over-linked entity is worse than an unlinked one: you are
actively teaching machines that your thing is something else.
This tool comes from The AI-powered Entity Audit
It is taught in Module 3, Lesson 3.2, on a real site, with the dataset bundled so your numbers match the video.