Company research agent that fills a Google Sheet

Company research agent that fills a Google Sheet

An n8n workflow that researches companies for you. Put a company name or a domain in a Google Sheet. An OpenAI agent
searches Google and reads the company's website. It writes what it finds back into the same row. That covers the domain,
the LinkedIn address, the cheapest plan, a free trial, an enterprise plan, the market, a case study link and up to five
integrations.

What is inside

• workflow.json: the workflow as n8n exports it, 14 nodes and eight sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The agent's prompt asked for the market as "B2B" or "B2B", so B2C could never come out. It now says "B2B" or
  "B2C".
• Both Google Sheets nodes pointed at their author's own sheet. They now hold YOUR_SHEET_ID and the first tab, and the
  saved sheet name and link are gone.
• Two node names lost their spaced hyphen: SerpAPI - Search Google is now SerpAPI Search Google, and
  Google Sheets - Update Row with data is now Google Sheets Update Row with data.
• The two small workflows inside Get website content and Search Google with ScrapingBee were cleaned. Their
  instance ID, their credential reference and their sample data are gone, and the example address is example.com.
• The eight sticky notes are rewritten and hold no outside links.
• Every credential reference is removed, so you pick your own. The workflow is switched off until you publish it.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key that can use gpt-4o.
• A SerpApi account and its API key.
• A Google account, a Google Sheet built as described below and a Google Sheets OAuth2 credential.
• Optional: a ScrapingBee key, if you want to search with it instead of SerpApi.

The Google Sheet

Row 1 holds these headers, spelled exactly like this:

| Column | Filled by | What it holds |
|---|---|---|
| input | You | A company name or a domain |
| domain | The agent | The domain, in the form name.tld |
| linkedinUrl | The agent | The company's LinkedIn address |
| has_free_trial | The agent | True or false, and false when the agent finds no strong evidence |
| cheapest_plan | The agent | The lowest paid plan as a number, or empty when it is not sure |
| has_entreprise_plan | The agent | Whether there is an enterprise plan |
| last_case_study_link | The agent | The first case study link it finds |
| market | The agent | B2B or B2C |
| integrations | The agent | Up to five tools the company integrates with |
| enrichment_status | The workflow | Leave it empty. The workflow writes done |

You do not make a row_number column. n8n adds it when it reads the rows. Fill input for each company and leave
the other cells empty.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Google Sheets OAuth2 credential. Select it in Get rows to enrich and Google Sheets Update Row with data.
3. In both nodes replace YOUR_SHEET_ID with your sheet's ID, the long part of its web address between /d/ and
   /edit. Both read the first tab. To use another tab, change the Sheet field in both.
4. Add an OpenAI credential and select it in OpenAI Chat Model.
5. Add a SerpApi credential with your key and select it in SerpAPI Search Google.
6. Put two companies in the sheet, press Test workflow and watch the rows fill.
7. Save and press Publish to run the workflow every 2 hours.

How a run goes

1. When clicking "Test workflow" and Schedule Trigger both start the same line. The schedule fires every 2 hours.
2. Get rows to enrich reads every row whose enrichment_status cell is empty.
3. Input keeps two things from each row: company_input, taken from input, and row_number.
4. Loop Over Items hands over one row at a time.
5. AI company researcher gets the company and a list of what to find. It may take up to 10 steps. OpenAI Chat Model
   runs gpt-4o at a temperature of 0.3. The agent can call SerpAPI Search Google and Get website content, which
   fetches a page and returns its text without the page head. Structured Output Parser makes it answer in the fixed
   fields.
6. AI Researcher Output Data pulls those fields out of the answer.
7. Merge data joins them to the row's row_number by position.
8. Google Sheets Update Row with data finds the row by row_number, writes the columns and sets enrichment_status
   to done. The loop then takes the next row. When no row is left, the run ends.

Change the look

• What the agent looks for: the prompt in AI company researcher, then the fields in Structured Output Parser, the
  mapping in AI Researcher Output Data and the columns in Google Sheets Update Row with data. Change all of them
  together.
• Which rows run: the filter in Get rows to enrich. Clear a row's enrichment_status to research it again.
• How often it runs: the interval in Schedule Trigger.
• The model: gpt-4o in OpenAI Chat Model.
• How hard it tries: the maximum number of iterations in AI company researcher.
• The search: connect Search Google with ScrapingBee to the agent as a tool, add a Query Auth credential that holds
  your ScrapingBee key under the name api_key, and remove SerpAPI Search Google.

Good to know

• OpenAI's deprecations page, read on 12 November 2026, lists the dated gpt-4o versions for shutdown on 14 January
  2027, with gpt-6-luna as the replacement. It does not list the bare name gpt-4o, which this copy keeps.
• The agent also returns has_API, but Google Sheets Update Row with data does not write it. To keep it, add a
  has_API column and map the field in that node.
• The sheet column is spelled has_entreprise_plan, while the agent's field is has_enterprise_plan. Keep the
  column header as it is, or change it in both places.
• Get rows to enrich sets a lookup column, enrichment_status, and no lookup value. With no value, the Google Sheets
  node matches the rows where that cell is empty.
• A row the agent fails on stays empty, so the next run tries it again.
• A long run can still be going when the next one starts. Both then read the same empty rows and research them twice.
• Get website content makes a plain web request. A page that needs a browser to draw it, or that blocks bots, gives
  back little text.
• Search Google with ScrapingBee is not connected to anything and does nothing until you connect it.
• The answers come from a language model and a search, so check them before you use them in outreach.
• OpenAI bills a gpt-4o call for each step the agent takes, up to 10 a company, and SerpApi bills each search.
• Your data: the company names go to OpenAI and SerpApi, and the page text goes to OpenAI.