This is a submission for the AI Agents Challenge powered by n8n and Bright Data
What I Built
A tool to build Qualified Leads funnels at zero cost. AI automation agencies get alerted when businesses get bad reviews, so they can be the first to call and offer a free damage control SaaS product (also built in n8n using the "Google Business Profile" verified node), which serves to establish rapport so they can offer bespoke services down the line.
The tool identifies prospects that are likely to be interested in automated damage control, by filtering for businesses with overall_ratings > 4.0 (likely pay attention to reputation management), reviews with ratings <= 3 (likely to impact customers visiting their page), owner_response = is_null (likely need to write a response, so you can offer to save them the work).
The tool uses a hosted n8n Chat interface so users can indicate preferences conversationally, generate sales scripts tailored for specific business and recent reviews, and delivers them to users via Email or Slack. The workflow uses n8n's AI Agent node with tool chaining, including Bright Data tool calls, to allow for flexible requests while still providing sufficient guardrails to optimize for good user experience.
Demo
Video Walkthrough
(Actual demo from 5:40-8:40)
Please excuse the poor audio!
Chat & Slack Delivery
Email Delivery
n8n Workflow (GitHub Gist)
https://gist.github.com/myopicOracle/163fed9602906eb99e50da02211481d3
Technical Implementation
Node-by-Node Breakdown
1. Chat Trigger
Starts the workflow in response to a user message from the chat UI. Passes the message and session context downstream.
2. Chat Trigger Agent B1
Acts as the central AI orchestrator—interprets user input, manages workflow steps, and decides when to call each tool node. Ensures a smooth, structured conversation.
3. Create Snapshot Tool
Requests business review data for the user’s chosen country. Output is a unique snapshot_id for later retrieval.
4. Loading Display Tool
Keeps users informed during data processing, displaying a loading message and GIF while waiting for results.
5. Retrieve Snapshot Tool
Checks snapshot status and retrieves review data once available, enabling the workflow to proceed at the right time.
6. Extract Top Reviews Tool
Identifies and summarizes the top 5 negatively reviewed businesses from the collected data, preparing options for user selection.
7. Generate Sales Script Tool
Creates a customized phone outreach script and a suggested public review response for a selected business. (Currently disabled.)
8. B1 LLM
Handles language understanding and generation, supporting the agent’s conversational intelligence.
9. B1 Chat Memory
Maintains conversation history and workflow context, allowing for continuity and effective state management throughout the session.
10. Send Email Tool
Sends generated scripts to the user’s email, if requested.
11. Slack Message Tool
Delivers scripts via Slack, when selected by the user.
How Nodes Are Orchestrated
- Agent B1 is the command center: It interprets user input, maintains workflow state, determines which tool node to call and when, and coordinates the entire workflow with strict sequence control.
- Data flows from the chat entry, through the agent, then to data collection, extraction, script generation, and finally to the chosen delivery node.
- Memory and LLM nodes guarantee context awareness and conversational quality across multiple user interactions.
- Delivery nodes are only triggered upon explicit user request, ensuring privacy and control over script sharing.
Bright Data Verified Node
I mainly leveraged the "Filter dataset" and "Retrieve snapshot contents" modes of the Bright Data Verified Node as tool calls for my AI Agent, but they're also available as standalone nodes, and you can do stuff like: store snapshot, filter dataset, get dataset metadata, get snapshot data, get snapshot metadata, split snapshot, list datasets, list snapshot ids, download snapshot, get results, check batch status, extract from url, start batch extraction, unlock and extract url.
The Bright Data Tools sub-nodes for AI Agents were the key pieces to my workflow’s ability to provide timely, high-quality sales prospects. It enabled the automated retrieval of up-to-date Google Maps business reviews tailored to the user's selected country. When the workflow begins, once the user specifies the desired country, the agent validates and formats the country code, then passed it into the "Group Filter" parameter of the Bright Data node, which will generate a snapshot based on any number of filter parameters you set (although you can also apply single filters). Then the Bright Data node returns a snapshot_id that you can use to retrieve that specific snapshot downstream or save into memory.
By leveraging the Bright Data Verified Node, this workflow can programmatically gather, monitor, and process business review datasets without manual intervention. This integration is what empowers the automated review extraction, selection of target businesses, and ultimately, the hyper-personalized script generation for sales outreach.
In summary, the Bright Data Verified Node transforms raw online review data into an actionable sales asset, tightly integrated in the workflow for maximum automation, reliability, and value for business consultants and agencies.
Journey
n8n
I made the mistake of jumping right in without doing a deep dive on docs for each of the nodes I was using, or looking up related nodes to analyze trade-offs. Very quickly, I was putting out fires left and right without a clear path in mind. Looking back, that's also where a lot of the learning happened, but I wish I knew these things before I started:
- n8n has a "Chat with Docs" modal in n8n's docs you can use to get far more accurate answers on syntax and trade-offs, because a lot of the answers you'll get from generic LLM's reference older variants of the same node, so it might give you advice that is no longer optimal.
- Some of n8n's AI Agent and LLM/Tool calls use Langchain under the hood (n8n-nodes-langchain.llmChatOpenAI; @n8n/n8n-nodes-langchain.toolCode; @n8n/n8n-nodes-langchain.chatTrigger) so you can transfer over any knowledge you have over to get a better understanding of how to structure calls and prompt engineer your way to victory.
- The browser AI Assistant can directly access your workflow's JSON (I think) as well as the n8n docs, so it's very helpful for debugging issues that could be attributed to several points of failure in your workflow. However, the n8n docs remain the ultimate source of truth when setting up your nodes, especially with optional parameters or new nodes.
Bright Data
Firstly, shoutout to @depapp for letting me know that we can just pull snapshots instead of buying the whole marketplace dataset. Before I found this out, I was trying to create CommunityPulse's functionality with entirely manual HTTP requests to several Google Places API endpoints. Massive pain the derriere. Why? Because the Places endpoint only returns a list based on a lat/long, so you first need to call the Geocode endpoint to transform user requests for city/country.
But the list that the Places endpoint returns doesn't contain any reviews! So you need to call the Reviews endpoint. Guess what? The Reviews endpoint only allows calls to be made with individual place_id URL params, so you need to loop over your array of extracted place id's and call the Reviews endpoint individually for each business- quick way to get blocked!
Bright Data for save: In addition to custom web scrapers you can set up quite quickly by using the Bright Data Agent to generate a starter script, there's also a massive collection of unlocked data in Bright Data's Dataset Marketplace. And it's pretty wild the range that it spans. Off the top of my head, I can think of several commercial B2B products (Bloomberg, MergerMarket, S&P CapitalIQ, MorningStar, etc.) that charge boatloads of money for people to access similar data that's available at a fraction of the price. 👀
Parting Thoughts
I went from putting out fires to drinking from the firehose, but after that initial hump (which many of you probably don't relate to), it gets really fun with the speed that you can prototype ideas and discard them without looking back. I've only barely scratched the surface on the number of integrations that n8n has, or all of the data that's at your fingertips with Bright Data, without exorbitant costs or advanced scripting skills. If you're still all-in on BeautifulSoup, would highly recommend checking out Bright Data's custom Web Scrapers and IDE that you can use to build and test queries.
Top comments (1)
Super cool use of n8n agents + Bright Data—this feels like an AI SDR for reputation-led lead gen. It aligns with the broader shift to agentic GTM: real-time intent signals + hyper-personal outreach at near-zero cost. Curious—will you expand to multi-channel reviews (Yelp, FB) and auto-route responses for a full “reputation ops” loop?