How to Answer Questions from Your Docs with AI in Make.com

Build a lightweight retrieval flow that pulls relevant notes from a Data Store and answers questions with AI.

11 minAdvanced

Generic chatbots make things up. This scenario grounds answers in your own content by retrieving matching snippets from a Make Data Store first, then asking AI to answer using only that context. It is a simple retrieval augmented pattern with no external vector database.

A Make.com account
An OpenAI API key
A Make Data Store filled with your FAQ snippets
A trigger such as a webhook or form for incoming questions

Step 1: Build the knowledge Data Store

In Make go to Data stores and create one with fields for topic, keywords, and answer. Add a row for each FAQ entry. This is your searchable knowledge base, kept entirely inside Make.

Make.com - Data store

key | keywords | answer

1 | refund, money back | Refunds are issued within 14 days...

2 | password, reset | Reset your password from Settings...

3 | export, csv | Export data via the Reports tab...

Each row holds a snippet the model can be grounded on.

Step 2: Receive the question

Add a Webhooks Custom webhook trigger and run it once with a sample question so Make maps the question field.

Step 3: Retrieve matching snippets

Add a Data store Search Records module. Use a filter that matches the keywords field against words in the question, and return the top few rows. These become the grounding context for the model.

Keyword search versus embeddings

This uses simple keyword matching, which is fine for a small FAQ. For a large corpus, generate embeddings with the OpenAI embeddings endpoint and store the vectors, then rank by similarity instead.

Step 4: Answer with grounded context

Add OpenAI Create a Completion (Chat). Pass the retrieved answers as context and instruct the model to answer only from that text and to admit when it does not know.

OpenAI system message

Answer the user's question using ONLY the context below. If the answer is not in the context, reply: "I do not have that information." Do not invent details.

Context:
{{join(map(2.array; "answer"); "\n---\n")}}

Make.com - RAG flow

Webhook -> Data store Search -> OpenAI Chat -> Webhook Response

question retrieve snippets grounded answer reply JSON

Step 5: Return the answer

Add a Webhooks Webhook response module to send the AI answer back to the caller, or swap it for a Slack or Telegram module if you want the answer delivered to a channel.

Result: a question answering assistant that speaks only from your documented answers, so it stays accurate and refuses to guess on anything outside your knowledge base.