A common way to connect documents to the AI is to split them into chunks. When you ask a question, the tool finds the chunks with the most similar words and hands them to the AI. That works for simple questions. It goes wrong when an answer needs two facts joined correctly. It can mix up two people with the same name, or quote a rule that was replaced later. A written map of how things connect fixes this. It is called a knowledge graph.
The mistake most people make
They rely on word search alone for questions that need two or three facts joined in the right order.
“A logistics company asked its document search tool: 'Which supplier makes boxes for our Chennai export unit, and what credit do they give?'”
“Agarwal Packaging supplies boxes for Chennai on 15-day credit.”
How it works: the metro map
Think of the Delhi Metro. Each station is a thing: Rajiv Chowk, Kashmere Gate, Hauz Khas. Each line is a link. To get from Noida to Gurgaon you don't pick stations that sound alike. You follow the lines: Noida City Centre → Blue Line → Rajiv Chowk → Yellow Line → HUDA City Centre. A map of your business works the same way. The AI follows written links from a customer to a contract to a delivery term.
The parts of the map
1. Things
The main nouns in your business: customers, products, suppliers, contracts, warehouses and staff.
2. Three-part facts
Each fact is thing, link, thing. For example: 'Sharma Enterprises → distributes → organic jaggery' and 'organic jaggery → stored at → Bhiwandi hub'.
3. Following the links
With the links written down, the AI can answer 'If the Bhiwandi hub floods, which Sharma Enterprises orders are late?' It follows two links instead of guessing.
How to keep a simple map
Turn this document into a table of facts. Here is the document: """ [Paste your contract, process note or operating notes here] """ Give me a table with these columns: | Thing | Link | Thing | Conditions | Rules: 1. Use plain verbs for the link: supplies, approves, stores, invoices. 2. Put conditions such as "within 15 days" or "domestic orders only" under Conditions. 3. Use only what the document says. If a link is unclear, write [CHECK]. 4. After the table, list anything that isn't linked to an owner.
Paragraphs become a table of links you can check, save and paste into later requests.
Working File Templates
Copy or save these plain text files to use directly in your workspace.
A plain-text entity relationship table linking suppliers, inventory, logistics, and customer accounts.
# Business Entity Relationship Graph: Supply Chain & Distribution Version: 1.0 · Updated: 2026-09-15 ## 1. Relational Triples | Subject (Entity) | Predicate (Relation) | Object (Entity) | Conditions / Constraints | Source Reference | | :--- | :--- | :--- | :--- | :--- | | Agarwal Packaging | supplies | Corrugated Boxes Type-A | 60-day credit term | Contract AGR-2026 | | Agarwal Packaging | supplies | Heavy Duty Wooden Pallets | Advance payment | Contract AGR-2026 | | Bhiwandi Hub | stores | Corrugated Boxes Type-A | Minimum buffer 2,000 units | SOP-WH-04 | | Sri Balaji Logistics | transports_from | Bhiwandi Hub | Requires 24-hr advance notice | SLA-LOG-2026 | | Sri Balaji Logistics | delivers_to | Pune Automotive Plant | Dispatches twice weekly | SLA-LOG-2026 | | Pune Automotive Plant | receives_from | Sri Balaji Logistics | Strict delivery window: 8 AM - 12 PM | PO-AUTO-991 | | Sharma Enterprises | acts_as | Regional Distributor | Covers Maharashtra West | DIST-AGR-01 | | Sharma Enterprises | orders_through | Portal ID: SHARMA-WH | Max credit ceiling ₹5,00,000 | KYC-SHARMA | ## 2. Unlinked Entities Under Observation - Nagpur Transit Yard: Mentioned in route discussions but no active transporter SLA signed.
Additional Prompts for This Lesson
Tailored prompts for specific workflows and departments.
Extracts subject-predicate-object triples with clear constraints from multi-party contracts or notes.
You are a knowledge graph data assistant. Here is the operational document: """ [Paste your contract or operating procedure here] """ Your task: Extract all explicit relationships into a clean markdown table of relational triples. Table format: | Subject (Entity A) | Predicate (Relationship) | Object (Entity B) | Constraint / Condition | Source Line | Rules: 1. Use exact entity names as stated in the document. 2. Predicates must be lowercase active verbs: "supplies", "authorises", "stores_at", "invoices", "escalates_to". 3. Place time limits, rupee amounts, and caveats under "Constraint / Condition". 4. If a relationship is implied but not stated with certainty, omit it. 5. End with a list of "Isolated Entities" that appear in the text without clear connections.
Converts messy contractual clauses into discrete factual triples that an AI can trace without hallucinating connections.
What does a written map of links give you that word search alone doesn't?
Show the answer
Word search finds text that sounds similar, so it can join facts that don't belong together. A map states which things are linked and how, so the AI follows the real links.