NEO4J-GENAI study guide: topics, format and a 4-week plan

The Neo4j and Generative AI Certification (NEO4J-GENAI) validates that you can build retrieval-augmented generation applications that combine Neo4j knowledge graphs, vector search and large language models. It suits developers who already work with Neo4j or are building GraphRAG systems. The exam is free, taken online through GraphAcademy, open-book with Neo4j documentation, and has 50 questions in 60 minutes with an 80% pass mark.

NEO4J-GENAI at a glance

Details as of September 2026 — confirm on the official exam page before booking.

Results appear as soon as you submit, and a certificate is issued straight away. If you do not pass, you can retake after 24 hours; Neo4j states retakes are unlimited and always free.

Who this exam is for

The exam is a good fit for backend developers, ML engineers and data scientists who build RAG applications and want to show both core Neo4j skills and the GenAI-specific ones. The exam page lists no formal prerequisites. Neo4j says candidates who complete its recommended preparation path are best positioned, and GraphAcademy's free courses cover both halves of the exam: graph fundamentals and Cypher, and GenAI topics such as building knowledge graphs with neo4j-graphrag and working with Neo4j MCP tools.

Look at the topic split before assuming this is purely an AI exam. Almost half of the questions (23 of 50) are about Neo4j itself (fundamentals, modelling, Cypher, schema and drivers), and you will need to write Cypher, not only read it.

What the exam covers

Neo4j publishes question counts per topic rather than percentages. The percentages below are those counts out of 50.

GraphRAG & Vectors (15 questions, 30%)

The largest topic. It covers how graph structure improves retrieval: combining a vector similarity search with a traversal that pulls in related context, generating Cypher from natural language, and building a knowledge graph from unstructured documents with an LLM. Know the retriever patterns in the neo4j-graphrag Python package and when each fits a question the user might ask.

Data Modeling (10 questions, 20%)

Graph modelling decisions: what should be a node, a relationship or a property, how specific relationship types should be, and how a model follows the questions the application needs to answer. For GenAI work this includes how documents and chunks relate to the entities extracted from them.

Vectors (6 questions, 12%)

Embeddings and vector indexes on their own: what an embedding represents, why the index dimension must match the embedding model, how similarity functions differ, and how a vector query returns nodes with a score.

MCP (6 questions, 12%)

The Model Context Protocol, and how an agent or AI assistant connects to Neo4j through an MCP server to read the schema and run Cypher. Understand what tools such a server exposes and the risk of letting an agent run write queries.

Neo4j Fundamentals (5 questions, 10%)

The property graph model, labels, relationship direction, and the basics of running Neo4j, including AuraDB.

Cypher (5 questions, 10%)

Pattern matching, filtering, aggregation, and the difference between CREATE and MERGE. Few questions carry the Cypher label, but Cypher appears throughout the other topics and in the hands-on questions, so its real weight is larger than 10%.

Schema & Indexes (2 questions, 4%)

Constraints and the index types Neo4j offers, including full-text and vector indexes, and when a query can use them.

Drivers (1 question, 2%)

Connecting from application code: creating a driver once and reusing it, sessions and transactions, and passing parameters rather than building query strings.

A 4-week study plan

The plan groups topics that build on one another and gives the most time to the topics with the most questions, with Cypher practice running through every week because the hands-on questions depend on it. Set up a free AuraDB instance or a local database first so you can run every query you read.

Week 1: Fundamentals, Cypher and modelling (modelling is 10 of this week's 22 questions, so give it the largest share). Complete the GraphAcademy fundamentals and Cypher courses. Model a small domain of your own, load it, and write ten queries against it, including aggregations and optional matches. Add constraints and indexes, then confirm with EXPLAIN that your queries use them.

Week 2: Vectors. Add embeddings to your graph, create a vector index and run similarity queries. Change the similarity function or the embedding model and note what breaks. This week lays the ground for the heaviest topic.

Week 3: GraphRAG. Build a small GraphRAG pipeline with neo4j-graphrag: build a knowledge graph from a few documents, then compare plain vector retrieval, vector retrieval with a graph traversal, and Text2Cypher on the same questions. This topic alone is 15 questions, so spend the extra hours here.

Week 4: MCP, drivers and rehearsal. Connect an MCP-capable assistant to your database through the Neo4j MCP server and watch which queries it runs. Write a short script with the official driver. Then take full-length timed mocks and spend what remains on your weakest topic.

Where candidates lose marks

• Reading the counts as the whole story. Cypher has only five labelled questions, but GraphRAG, modelling and the hands-on items all assume you can write a correct query quickly.
• Relying on open book. Sixty minutes for 50 questions is just over a minute each. The docs help to confirm a function signature; they are too slow for learning a concept mid-exam.
• Mixing up retrieval patterns. A question about finding similar text and one about following relationships from the matched text need different retrievers. Read what the user in the scenario is actually asking for.
• MERGE semantics. MERGE matches or creates the whole pattern. Merging a long pattern when only one node already exists can create a duplicate of that node, which makes it an easy point to examine.
• Multiple-select questions. If a question says how many options to choose, select exactly that many, and judge each option on its own merits rather than looking for a pair that sounds good together.
• Vector index mismatches. An index whose dimension or similarity function does not match the embeddings produces errors or poor results. Know which setting belongs to the index and which to the model.

Practising for NEO4J-GENAI on CertifyAIx

This exam includes hands-on questions where you write Cypher against a real dataset, and it is open-book. A question bank cannot rehearse writing and running queries, so use our sets to firm up the concepts and use the GraphAcademy exercises and your own database for the query-writing. As a partial substitute, many of our questions give you a small, fully described graph and ask what a query returns or which query answers a question. What the [NEO4J-GENAI practice questions](/certifications/neo4j-generative-ai-certification) page offers:

• 516 original questions covering all eight topic areas, with the most on GraphRAG & Vectors
• Single-answer and multiple-select formats
• 3 full-length mock exams with 50 questions in 60 minutes, following the same per-topic counts as the real exam
• 24 mixed domain papers and 23 topic papers of up to 25 questions each, timed at 60 minutes per 25 questions
• An explanation for every question, covering why each option is right or wrong

See [how CertifyAIx practice questions are written](/blog/how-certifyaix-practice-questions-are-written) for the method, and [how to choose an AI certification](/blog/how-to-choose-an-ai-certification) if you are weighing this exam against others.

Frequently asked questions

How hard is the Neo4j and Generative AI Certification?

The 80% pass mark and one-minute pace make it tighter than the free, open-book format suggests. Developers who already write Cypher and have built a RAG application usually find the GenAI topics familiar and need to focus on speed and on the Neo4j-specific details.

How long should I prepare?

Two to six weeks is a sensible range. Someone who already uses Neo4j can concentrate on GraphRAG, vectors and MCP; someone new to graphs needs time for fundamentals and Cypher first, since they run through every topic.

Are CertifyAIx questions real exam questions?

No. Every CertifyAIx question is original, written against the topics Neo4j publishes for this exam. We never use questions from the live exam.

What if I fail?

You can retake after 24 hours. Neo4j states retakes are unlimited and free.

Is the exam open-book?

Yes. Neo4j allows its documentation during the exam. Check the official page for the current list of permitted resources before you start.

CertifyAIx is independent and not affiliated with Neo4j. Neo4j and Generative AI Certification is a trademark of its owner. All CertifyAIx practice questions are original.