1k+ участников

Вступить в Telegram

n8n

•

27.08.2025

Automating deep research with AI Research System

•

8 articles

goal

Create a fast and reliable tool for automatically conducting complex studies that saves hours of manual work and provides a ready-made structured report in 5-10 minutes.

Scope of application

Business intelligence — analysis of markets, competitors and trends
Scientific research — collection and systematization of data by topic
Education — preparation of materials for term papers, diplomas, articles
Content marketing — creating expert articles and reviews
Startups — niche research and preparation of presentation materials

Process description

This automation is a comprehensive system for conducting in-depth research on any given topic. The system automatically breaks down the topic into 5 subtopics, researches each of them using 25 sources, creates a structured HTML report and sends the finished PDF document to the user by email.

‍Required API keys and settings

Main services:

OpenAI API key - for 8 different AI operations
Tavily API key - for 5 in-depth web searches https://app.tavily.com/home
APItemplate.io API key - to generate PDF from HTML https://app.apitemplate.io/manage-api/
Google Sheets OAuth2 - to save intermediate data
Gmail OAuth2 - to send a finished report

‍Google Sheets spreadsheet structure:

Required columns:

The topic of analysis is the main topic (key field)
Title - report title
Introduction - introduction
Chapter 1, Chapter 2, Chapter 3, Chapter 4, Chapter 5 - Chapter Titles
1 Topic: Sources, 1 Topic: Sections, 1 Topic: Content - Topic Data 1
2 Topic: Sources, 2 Topic: Sections, 2 Topic: Content - Topic Data 2
3 Topic: Sources, 3 Topic: Sections, 3 Topic: Content - Topic Data 3
4 Topic: Sources, 4 Topic: Sections, 4 Topic: Content - Topic Data 4
5 Topic: Sources, 5 Topic: Sections, 5 Topic: Content - Topic Data 5
Sources - a general list of all sources
ToC - table of contents

‍Detailed architecture

SECTION 1: PREPARATION (Red Block)

‍

1.1 Launch form - On form submission

Purpose: User entry point Settings:

{

“FormTitle”: “In-depth analysis”,

“FormDescription”: “Enter a topic for analysis”,

“FormFields”: [

{

“FieldLabel”: “Search Topic”,

“placeholder”: “theme”,

“RequiredField”: true

{

“FieldLabel”: “Email”,

“placeholder”: "name@example.com “,

“RequiredField”: true

}

]

}

‍1.2 AI Planning - Plan Topics

Purpose: Breaks the main topic into 5 sub-topics for research

Connected components:

OpenAI Chat Model (gpt-4o-mini)
5 Topics (Structured Output Parser)

Plan Topics System Prompt:

# Review

You're an AI research assistant built to create five highly relevant and in-depth search topics based on a given search topic. Your goal is to break down your search topic into five key sub-topics that will allow for in-depth research and comprehensive understanding.

## Instructions:

1) Break down the topic: Identify five separate but interrelated subtopics that, when researched, will provide a comprehensive understanding of the main search topic.

2) Ensure depth and relevance: Each search topic should be specific enough to provide in-depth research but broad enough to provide meaningful insights.

3) Avoid redundancy: Five search topics should be unique and cover different aspects of the main topic.

## Output format (JSON):

{

<Первая тема углубленного поиска>“ topic_1": “”,

<Вторая тема углубленного поиска>“ topic_2": “”,

<Третья тема углубленного поиска>“ topic_3": “”,

<Четвертая тема углубленного поиска>“ topic_4": “”,

<Пятая тема углубленного поиска>“ topic_5": “”

}

## Example:

Input: “AI-based business automation”

Conclusion:

{

“topic_1": “The role of AI in automating business processes”,

“topic_2": “AI automation tools and technologies”,

“topic_3": “Problems and ethical aspects of AI automation”,

“topic_4": “Successful cases of implementing AI in business”,

“topic_5": “The Future of AI Automation: Trends and Innovations”

}

‍Structured Output Parser Schema “5 Topics”:

{

“topic_1": “Understanding AI Agents”,

“topic_2": “Building AI Agents with n8n”,

“topic_3": “Optimizing Prompt Engineering”,

“topic_4": “Storing and Retrieving AI Knowledge”,

“topic_5": “Scaling AI Automation for Businesses”

}

‍WHAT IS Structured Output Parser Schema: This is a tool that makes AI return data EXACTLY in a given JSON format. Without it, AI can return text in any way it wants - in paragraphs, lists, in any form. The scheme tells AI: “Give me back exactly 5 topics in topic_1, topic_2, etc.”

WHY IS IT NEEDED HERE:

Ensures that you will receive exactly 5 topics (not 3, not 7)
Allows you to access each topic via $json.topic_1, $json.topic_2
The Switch node can determine exactly which topic to send where
Eliminates data parsing errors

HOW IT WORKS:

AI receives the prompt+ scheme as a strict instruction
AI must return JSON with the topic_1, topic_2, topic_3, topic_4, topic_5 fields
n8n automatically parses JSON and makes $json.topic_1 available for the following nodes

1.3 AI Structure Generation - Intro

Purpose: Generates a title, introduction, and chapter titles for the report

Connected components:

OpenAI Chat Model1 (gpt-4o-mini)
Title, Intro, Chapters (Structured Output Parser)

Intro system prompt:

# Review

You are an AI generator of report titles and introductions. Your task is to create a compelling title, a short introduction, and five clear chapter titles for the research report based on a given main search topic and five subtopics.

## Instructions:

1) Title: Create a clear and engaging title that reflects the overall theme of the report.

2) Introduction: Write a short introduction (2 paragraphs) that provides context, explains the significance of the topic, and outlines what the report will cover. This section should start with a title called “Introduction”.

<h2>3) Contents: At the end of the introduction, list five sections of the report, each of which is presented in the form of a clearly formatted chapter title in the headings.

4) Chapter titles: Make sure each chapter title is formatted as a title <h2>to make it visually distinct.

## Design:

- Name: an element <h1>in the center, decorated in a professional dark blue color.

- Introduction: enclosed <div>with a light grey background (#f4f4f4) for contrast.

- Paragraphs: readable font size (16 pixels), appropriate line height (1.6), and neutral text color (#333)

- The horizontal line (<hr>) should be used to separate the two introductory paragraphs.

- Chapter titles: clearly formatted as elements <h2>for easy reading.

## Output format (JSON):

{

“title”: “<Сгенерированное название отчета>”,

<Сгенерированное введение>“ introduction”: “”,

“chapter_1":" <h2>Chapter 1: [</h2>Title] “,

“chapter_2":" <h2>Chapter 2: [</h2>Title] “,

“chapter_3":" <h2>Chapter 3: [Title]</h2> “,

“chapter_4":" <h2>Chapter 4: [</h2>Title] “,

“chapter_5":" <h2>Chapter 5: [Title]</h2>”

}

## Example:

{

“title”: “<h1 style='text-align: center; color: #00366D;'>Artificial intelligence in modern business: a comprehensive analysis of automation</h1>”,

“introduction”: "<div style='background-color: #f4f4f4; padding: 20px; font-size: 16px; line-height: 1.6; color: #333;'><h2>Introduction</h2> <p>AI-based automation has become a key factor in the transformation</p> of modern business... </div><hr><p>This report explores five critical aspects of AI adoption...</p> “,

“chapter_1": “<h2 style='text-align: center; color: #00366D;'>Chapter 1: The role</h2> of AI in automation”

}

‍Structured Output Parser Schema “Title, Intro, Chapters”:

{

“title”: "<Generated Report Title>“,

<Generated Introduction>“introduction”: "“,

“chapter_1":" <h2>Chapter 1</h2>: [Title] “,

“chapter_2":" <h2>Chapter 2</h2>: [Title] “,

“chapter_3":" <h2>Chapter 3: [Title]</h2> “,

“chapter_4":" <h2>Chapter 4</h2>: [Title] “,

“chapter_5":" <h2>Chapter 5: [Title]</h2>”

}

‍WHAT IS Structured Output Parser Schema: Here, the scheme provides all components of the report structure in a strictly defined format. Without it, AI could return everything in one text or in random order.

WHY IS IT NEEDED HERE:

Divides the title, introduction, and chapters into separate fields
Allows you to access every element: $json.title, $json.introduction, $json.chapter_1
Google Sheets node can save each item to its own column
<h2>Ensures HTML formatting of chapters (required tags)

HOW IT WORKS:

AI gets a schema and is obliged to return JSON with 7 fields
Each field contains a ready-made HTML code
Send Intro can take $json.output.title and save it to the “Title” column

1.4 Maintaining the basic structure - Send Intro

Purpose: Preserves the report structure in Google Sheets

Google Sheets settings:

Operation: Append
Document ID: Your “Deep Analysis” table ID (or just choose from the list)
Sheet Name: Sheet1
Columns Mapping:
- Analysis topic: {{$ ('On form submission') .item.json ['Search Topic']}}
- Title: {{$json.output.title}}
- Introduction: {{$json.output.introduction}}
- Chapter 1: {{$json.output.chapter_1}}
- Chapter 2: {{$json.output.chapter_2}}
- Chapter 3: {{$json.output.chapter_3}}
- Chapter 4: {{$json.output.chapter_4}}
- Chapter 5: {{$json.output.chapter_5}}

1.5 Preparing data for separation

Sequence: Set Topics → Split Out → Split Out1 → Merge1 → Merge → Switch

Set Topics:

{

“assignments”: [

{

“name”: “topics”,

“value”: “['topic_1', 'topic_2', 'topic_3', 'topic_4', 'topic_5']”,

“type”: “array”

}

]

}

‍Purpose: Creates an array of topic titles for later separation.

Split Out:

Field to Split Out: output
Purpose: Splits the Plan Topics result into separate items.

Split Out1:

Field to Split Out: topics
Purpose: Splits the ['topic_1', 'topic_2', 'topic_3', 'topic_4', 'topic_5'] array into 5 distinct items.

Merge1:

Mode: Combine
Combine By: Combine by Position
Purpose: Combines Split Out and Split Out1 streams by position.

Merge:

Mode: Combine
Combine By: Combine All
Purpose: The final merger before Switch includes data from Send Intro.

Switch:

5 conditions for routing:
1. {{$json.topics}} equals topic_1 → Output 0
2. {{$json.topics}} equals topic_2 → Output 1
3. {{$json.topics}} equals topic_3 → Output 2
4. {{$json.topics}} equals topic_4 → Output 3
5. {{$json.topics}} equals topic_5 → Output 4

SECTION 2: RESEARCH FLOWS (Yellow Blocks)

TOPIC STREAM 1 (detailed analysis):

2.1 Web search - Tavily

Purpose: Searches information on the first topic through the Tavily API

HTTP Request settings:

Method: POST
URL: https://api.tavily.com/search
Authentication: Header Auth
press Create new credentials
Header Name: Authorization
Header Value: Bearer your_tavily_api_key

Get the API key from here

https://app.tavily.com/home

JSON Body:

{

“query”: “{{$json.output}}”,

“topic”: “general”,

“search_depth”: “advanced”,

“chunks_per_source”: 3,

“max_results”: 5

}

‍Parameter explanation:

search_depth: “advanced” - deep search with content analysis
chunks_per_source: 3 - 3 text snippets from each source
max_results: 5 - maximum 5 sources

2.2 Splitting results - Split Out2

Purpose: Splits Tavily results into separate sources

Settings:

Field to Split Out: results
The result: Each source becomes a separate item for processing

2.3 Source Numbering - Code

Purpose: Assigns numbers 1-5 to the sources of the first topic

JavaScript code:

//Create separate items for each URL with a number

return items.map ((item, index) => {

return {

json: {

number: index + 1,

url: item.json.url,

NumberedURL: `$ {index+ 1}. $ {item.json.url} `

}

};

});

‍

Explanation:

index + 1 creates numbers 1, 2, 3, 4, 5
NumberedURL creates strings like “1. https://example.com”
This is necessary for the correct links in the report

2.4 Content Generation - Writer

Purpose: AI creates an HTML report based on the information it finds

Connected components:

OpenAI Chat Model2 (gpt-4o-mini)

Writer system prompt:

# Review

You are an advanced AI research assistant specializing in writing professional HTML reports based on a provided title, research, source, and style guide. Your task is to create a fully formatted HTML report that is visually appealing and structured in accordance with professional writing standards.

## Reporting guide

1. Compliance with the style guide

- The report must strictly follow the HTML style guide provided, ensuring consistency in style, colors, fonts, padding, and layout.

2. Structure and formatting

- The report should be well structured and formatted in professional HTML.

- Title (<h1>) — provided as a “Title”

<h2><h3><p>- The main text is fully formatted research content with the correct separation (,,).

- Each section should be enclosed in a div styled according to the guidance provided to ensure consistency.

- Use <p>for paragraphs,<ul>/<ol>for lists, and <hr>to separate sections when needed.

- The headings “Review” or “Conclusion” are not needed, just output the full report.

- The horizontal line (<hr>) should be used to separate sections for better visual clarity.

3. Research-based content

- Synthesize information logically and ensure that the report is well researched, fact-based, and professionally written.

- Use informative writing. Please provide as much detail as possible.

- Maintain a logical sequence of ideas by effectively summarizing key points.

4. Source attribution and clickable links

- Use the number in front of the source as a clickable attribute in the report.

- Links to sources should be embedded in the text using: <a href="SOURCE_URL" target="_blank">[1]</a>

- Don't include the “Links” section — instead, all sources should be embedded in the content.

‍

## Example output structure:

‍

<html>

<head>

<title>Research Report: [Title]</title>

<style>

body {font-family: Arial, sans-serif; line-height: 1.6; margin: 20px;}

h1, h2, h3 {color: #333;}

.report-section {background color: #f4f4f4; padding: 20px; font-size: 16px; line-height: 1.6; color: #333; margin-bottom: 10px;}

</style>

</head>

<body>

<h1>Research Report: [Title]</h1>

</div>

<h2>Introduction</h2>

<p>Research content with embedded links <a href="https://example.com" target="_blank">[1]</a>.</p>

<hr>

<p>Additional analysis with reference to <a href="https://example2.com" target="_blank">[2]</a>.</p>

</div>

</body>

</html>

‍The input data for Writer are:

Title: {{$ ('Split Out2') .item.json.title}}

Research: {{$ ('Split Out2') .item.json.content}}

Source: {{$json.NumberedURL}}

Style Guide Example: {{$ ('Switch') .item.json ['Chapter 1']}}

‍2.5 Data Aggregation - Aggregate & Aggregate1

Aggregate (for content):

Field to Aggregate: output
Purpose: Combines all HTML reports from Writer into one array

Aggregate1 (for sources):

Field to Aggregate: NumberedURL
Purpose: Combines all numbered URLs into one array

Why do we need aggregation:

Writer processes each source separately (5 times)
Aggregate combines 5 separate HTML chunks into one array
This allows you to keep all the content of the topic as a whole

2.6 Combining streams - Merge3

Purpose: Synchronizes aggregated content and sources

Settings:

Mode: Combine
Combine By: Combine by Position
The result: One item with full content and all theme sources

2.7 Structure extraction - HTML

Purpose: Extracts section titles from generated HTML

Settings:

Operation: Extract HTML Content
Data Property Name: output
CSS Selector: title
Extraction Values:
- Key: section
- CSS Selector: title

Why do you need this: To automatically create a table of contents, you need to know what sections AI has created in the report.

2.8 Combine Sections - Combine

Purpose: Collects all section titles into one array

JavaScript code:

return [{

sections: $input.all () .map (item => item.json.section)

}];

‍2.9 Saving results - Google Sheets1

Purpose: Saves all data from the first topic to a table

Settings:

Operation: Update
Document ID: “Deep Analysis” table ID
Matching Column: Analysis topic
Columns Mapping:
- 1 Topic: Sources: {{$ ('Merge3') .first () .json.NumberedURL.Join (”\n“)}}
- 1 Topic: Sections: {{$json.sections.join (”\n“)}}
- 1 Topic: content: {{$ ('Merge3') .first () .json.output.join (”\n\n“)}}

TOPICS 2-5 FLOWS (short description):

The structure is identical to topic 1, the only differences are:

Topic 2: Tavily1 → Split Out3 → Code1 → Writer1 → Aggregate2 → Aggregate3 → Merge4 → HTML1 → Combine1 → Google Sheets2

Code1: sources 6-10 (index + 6)
Writer1: uses the style from {{$ ('Switch') .item.json ['Chapter 2']}}

Topic 3: Tavily2 → Split Out4 → Code2 → Writer2 → Aggregate4 → Aggregate5 → Merge5 → HTML2 → Combine2 → Google Sheets3

Code2: sources 11-15 (index + 11)

Topic 4: Tavily3 → Split Out5 → Code3 → Writer3 → Aggregate6 → Aggregate7 → Merge6 → HTML3 → Combine3 → Google Sheets4

Code3: sources 16-20 (index + 16)

Topic 5: Tavily4 → Split Out6 → Code4 → Writer4 → Aggregate8 → Aggregate9 → Merge7 → HTML4 → Combine4 → Google Sheets5

Code4: sources 21-25 (index + 21)

SECTION 3: FINALIZATION (Green Block)

3.1 Combining all results - Merge2

Purpose: Collects results from all 5 topics in one place

Settings:

Mode: Combine
Number of Inputs: 5
Combine By: Combine All
Entrances: Google Sheets1, Google Sheets2, Google Sheets3, Google Sheets4, Google Sheets5

3.2 Data limit - Limit

Purpose: Optimizes data processing

Settings:

Max Items: 1
Why: Takes only the first element for further processing

3.3 Getting Sources - Get Sources

Purpose: Retrieves all sources from Google Sheets

Google Sheets settings:

Operation: Read
Document ID: “Deep Analysis” table ID
Filters: Equals analysis topic {{$ ('On form submission') .first () .json ['Search Topic']}}

3.4 Generating a list of sources - Sources

Purpose: AI creates an HTML list of all 25 sources

Connected components:

OpenAI Chat Model7 (gpt-4o-mini)

Sources system prompt:

# Review

You're an AI assistant specializing in creating well-formatted sections of HTML source code for research reports. You will be given a list of sources, each of which has been assigned a number. Your task is to create a “Sources” section in HTML format.

## Formatting rules:

- The section title should be <h2>Sources</h2>.

- Use an unordered list (<ul>) to display sources.

- Each source must be in the format:

<li><a href="SOURCE_URL" target="_blank">[NUMBER] Source name</a></li>

- It is necessary to maintain the original numbering of sources.

## Example output structure:

<h2>Sources</h2>

<ul>

<li><a href="https://example1.com" target="_blank">[1] Research on AI Development</a></li>

<li><a href="https://example2.com" target="_blank">[2] The Impact of Automation</a></li>

<li><a href="https://example3.com" target="_blank">[3] Machine Learning Trends</a></li>

</ul>

</div>

‍Input data:

Sources:

‍3.5 Saving Sources - Send Sources

Purpose: Saves the finished list of sources to a table

Google Sheets settings:

Operation: Update
Matching Column: Analysis topic
Column: Sources: {{$json.output}}

3.6 Get All Data - Get All Content

Purpose: Extracts the full report structure to create a table of contents

Settings: Identical to Get Sources, gets all table columns.

3.7 Creating a Table of Contents - Table of Contents

Purpose: AI generates a structured table of contents

Connected components:

OpenAI Chat Model8 (gpt-4o-mini)

Table of Contents system prompt:

# Review

You're an AI assistant tasked with creating a structured table of contents (ToC) in professional HTML formatting for a research report. Your output should be consistent with the style guide provided and be clean, readable, and well-structured.

## Guidelines

1) Format chapters as elements <h2>

<h2>- Each chapter should be a separate title.

<ol>2) Format sections as an ordered list ()

- Each section in the chapter should be a list item (<li>) inside<ol>.

- Maintain a clear hierarchy and indentation for easy reading.

3) Use this style guide

- Wrap the table of contents in <div>the same indent, font size, and color as the style guide.

- Add a horizontal line (<hr>) to split.

## Sample output:

<h2>Table of contents</h2>

<hr>

<ol>

<li>

<h2>Chapter 1: Introduction</h2>

<ol>

<li>Context and importance</li>

<li>Research Objectives</li>

</ol>

</li>

<li>

<h2>Chapter 2: Key Concepts</h2>

<ol>

<li>Theoretical bases</li>

<li>Practical applications</li>

</ol>

</li>

</ol>

</div>

‍

Table of Contents inputs:

Chapter 1: {{$ ('Plan Topics') .first () .json.output.topic_1}}

Sections: {{$json ['1 Topic: Sections']}}

‍

Chapter 2: {{$ ('Plan Topics') .first () .json.output.topic_2}}

Sections: {{$json ['2 Topic: Sections']}}

‍

Chapter 3: {{$ ('Plan Topics') .first () .json.output.topic_3}}

Sections: {{$json ['3 Topic: Sections']}}

‍

Chapter 4: {{$ ('Plan Topics') .first () .json.output.topic_4}}

Sections: {{$json ['4 Topic: Sections']}}

‍

Chapter 5: {{$ ('Plan Topics') .first () .json.output.topic_5}}

Sections: {{$json ['5 Topic: Sections']}}

‍

Style Guide: {{$json ['Introduction']}}

‍3.8 Save the table of contents - Send ToC

Purpose: Saves the finished table of contents to a table

Google Sheets settings:

Operation: Update
Document ID: “Deep Analysis” table ID
Matching Column: Analysis topic
Column: ToC: {{$json.output}}

3.9 Final Data Retrieval - Get All Content1

Purpose: Retrieves the full final version of all report data

Settings: Identical to previous Google Sheets read operations, retrieves all completed columns.

3.10 Combining into a single document - Combine Content

Purpose: JavaScript combines all parts of the report into a single HTML document

JavaScript code:

//N8N Code Node to combine multiple fields into a single field with new lines

‍

//Access the incoming data

const item = items [0];

‍

//Create a new field called “CombinedContent” that joins all fields with newlines

const CombinedContent = [

Item.json.Title ||”,

Item.json.intro ||”,

Item.json.toc ||”,

item.json ['Chapter 1'] || “,

item.json ['1 Topic: content'] || “,

item.json ['Chapter 2'] || “,

item.json ['2 Topic: content'] || “,

item.json ['Chapter 3'] || “,

item.json ['3 Topic: content'] || “,

item.json ['Chapter 4'] || “,

item.json ['4 Topic: content'] || “,

item.json ['Chapter 5'] || “,

item.json ['5 Topic: content'] || “,

Item.json.Sources ||”

] .join ('\n\n');

‍

//Add the combined content to the item

Item.json.combinedContent = CombinedContent;

‍

//Return the modified item

return [item];

‍

What's going on:

Takes all parts of the report from Google Sheets
Combines them in the right order: title → introduction → table of contents → chapter 1 → content 1 → chapter 2 → content 2 → etc → sources
Creates a CombinedContent field with a complete HTML document

3.11 Generate PDF - Generate PDF

Purpose: Converts HTML into a professional PDF document

HTTP Request settings:

Method: POST
URL: https://rest-us.apitemplate.io/v2/create-pdf-from-html
Authentication: Apitemplate.io account (pre-defined in n8n)

Get the API key here https://app.apitemplate.io/manage-api/

Query Parameters:

filename: {{$ ('On form submission') .first () .json ['Search Topic']}} .pdf

JSON Body:

{

“body”: “{{$json [" CombinedContent "] .replace (/” /g, '\\ "') .replace (/\n/g, '\\n')}}”,

“css”: "<style>.bg {background: red};</style> “,

“data”: {

“name”: “This is a title”

“settings”: {

“paper_size”: “A4",

“orientation”: “1",

“header_font_size”: “9px”,

“margin_top”: “40",

“margin_right”: “10",

“margin_bottom”: “40",

“margin_left”: “10",

“print_background”: “1",

“DisplayHeaderFooter”: true,

“custom_header”: "<style>#header, #footer {padding: 0! important;}</style>\n<table style=\ "width: 100%; padding: 0px 5px; margin: 0px! important; font-size: 8px\ ">\n<tr>\n<td style=\" text-align:left; width: 30%! important;\ "><span class=\" date\ "></span></td>\n<td style=\" text-align:center; width: 30%! important;\ "><span class=\" pageNumber\ "></span></td>\n<td style=\" text-align:right; width: 30%! important;\ "><span class=\" totalPages\ "></span></td>\n\n</tr></table>“,

“custom_footer”: "<style>#header, #footer {padding: 0! important;}</style>\n<table style=\ "width: 100%; padding: 0px 5px; margin: 0px! important; font-size: 8px\ ">\n<tr>\n<td style=\" text-align:left; width: 30%! important;\ "><span class=\" date\ "></span></td>\n<td style=\" text-align:center; width: 30%! important;\ "><span class=\" pageNumber\ "></span></td>\n<td style=\" text-align:right; width: 30%! important;\ "><span class=\" totalPages\ "></span></td>\n\n</tr></table>”

}

‍

Explanation of PDF settings:

paper_size: “A4" - A4 format
orientation: “1" - portrait orientation
margin_* - margins in millimeters
DisplayHeaderFooter: true - includes headers and footers
custom_header/footer - HTML code for footers with page numbers

The result: Apitemplate.io returns an object with download_url to download the finished PDF.

3.12 Downloading a PDF file - Download PDF

Purpose: Downloads the finished PDF file from the link received

HTTP Request settings:

Method: GET
URL: {{$json.download_url}}
Options: Download Binary Data included

What's going on:

Gets download_url from the Generate PDF response
Downloads a PDF file as binary data
Prepares a file to be attached to an email

3.13 Sending the finished report - Send Report

Purpose: Sends a PDF report to the user by email

Gmail settings:

Send To: {{$ ('On form submission') .first () .json.email}}
Subject: Deep Research Report: {{$ ('On form submission') .first () .json ['Search Topic']}}
Email Type: Text
Message:

Your Deep Research Report on {{$ ('On form submission') .first () .json ['Search Topic']}} is complete.

You can find it attached below.

‍Attachments:

Append Attribution: false
Binary Attachments: Include binary data from Download PDF

The result: The user receives an email with an attached 15-25 page PDF report with a full study on his topic.

Complete connection scheme for OpenAI models

8 OpenAI Chat Model nodes use the gpt-4o-mini model:

OpenAI Chat Model → Plan Topics (breakdown by topic)
OpenAI Chat Model1 → Intro (report structure)
OpenAI Chat Model2 → Writer (topic content 1)
OpenAI Chat Model3 → Writer1 (topic content 2)
OpenAI Chat Model4 → Writer2 (topic content 3)
OpenAI Chat Model5 → Writer3 (topic content 4)
OpenAI Chat Model6 → Writer4 (topic content 5)
OpenAI Chat Model7 → Sources (list of sources)
OpenAI Chat Model8 → Table of Contents (table of contents)

Google Sheets spreadsheet structure:

Required columns:

The topic of analysis is the main topic (key field)
Title - report title
Introduction - introduction
Chapter 1, Chapter 2, Chapter 3, Chapter 4, Chapter 5 - Chapter Titles
1 Topic: Sources, 1 Topic: Sections, 1 Topic: Content - Topic Data 1
2 Topic: Sources, 2 Topic: Sections, 2 Topic: Content - Topic Data 2
3 Topic: Sources, 3 Topic: Sections, 3 Topic: Content - Topic Data 3
4 Topic: Sources, 4 Topic: Sections, 4 Topic: Content - Topic Data 4
5 Topic: Sources, 5 Topic: Sections, 5 Topic: Content - Topic Data 5
Sources - a general list of all sources
ToC - table of contents

The result of automation

What the user gets:

15-25 page PDF report with professional design
Structured study on 5 aspects of the topic
25 numbered sources with active links
Table of contents with a hierarchical structure
HTML formatting with the same style
Delivery time: 5-10 minutes after submitting the form

Quality characteristics:

Study depth: 5 topics × 5 sources = 25 sources
Content volume: ~15,000-20,000 words
Professional design: uniform style, headers and footers, page numbering
Automated quoting: embedded links to sources
Structuring: a clear hierarchy of sections and subsections

It's done. This automation turns any topic into a professional research report fully automatically in 5-10 minutes.

You can always get a JSON file and instructions in video format by joining our unique Automation club.

write to us and we will show you the way to efficiency

Olga

@olga_startduck

Automating deep research with AI Research System

goal

Scope of application

Process description

Main services:

SECTION 1: PREPARATION (Red Block)

1.1 Launch form - On form submission

1.3 AI Structure Generation - Intro

1.4 Maintaining the basic structure - Send Intro

1.5 Preparing data for separation

SECTION 2: RESEARCH FLOWS (Yellow Blocks)

TOPIC STREAM 1 (detailed analysis):

2.1 Web search - Tavily

2.2 Splitting results - Split Out2

2.3 Source Numbering - Code

2.4 Content Generation - Writer

2.6 Combining streams - Merge3

2.7 Structure extraction - HTML

2.8 Combine Sections - Combine

TOPICS 2-5 FLOWS (short description):

SECTION 3: FINALIZATION (Green Block)

3.2 Data limit - Limit

3.3 Getting Sources - Get Sources

3.6 Get All Data - Get All Content

3.7 Creating a Table of Contents - Table of Contents

3.9 Final Data Retrieval - Get All Content1

3.10 Combining into a single document - Combine Content

3.11 Generate PDF - Generate PDF

3.12 Downloading a PDF file - Download PDF

3.13 Sending the finished report - Send Report

Complete connection scheme for OpenAI models

Google Sheets spreadsheet structure:

The result of automation

What the user gets:

Quality characteristics:

Читайте также

write to us and we will show you the way to efficiency