Someone just posted “is [your product] a scam?” in a subreddit with 400,000 members. A competitor’s founder tagged you in a thread. A customer raved about you on a niche forum you’ve never heard of. By the time you stumble across any of this — if you ever do — the conversation is days cold.
A brand mention agent fixes that. It watches the places people talk about you, reads each new mention, decides whether it actually matters, and pings you only when it does. No dashboard you have to remember to check, no scrolling. And you can build one without writing a line of code. Here’s exactly how we do it.
What a “brand mention agent” actually is
Strip away the hype and it’s four steps wired together:
- Listen — pull in new posts, comments, articles, or reviews that contain your name.
- Judge — an AI model reads each one and classifies it: is this positive, negative, or neutral? Is it a sales lead, a support fire, or just noise?
- Route — send the ones that matter somewhere you’ll see them (Slack, email, Telegram), with a one-line summary.
- Repeat — on a schedule, forever, without you touching it.
The “AI agent” part is step 2. A plain alert (like Google Alerts) just forwards everything and buries you. The agent layer is what reads in context and filters, so you get five alerts a day that deserve attention instead of fifty that don’t. That difference is the entire point.
Decide first: do you even need to build anything?
Honest answer — sometimes you don’t. Before you open an automation tool, check whether a free off-the-shelf alert covers you:
- F5Bot emails you within minutes when your keyword appears on Reddit, Hacker News, or Lobsters. Free, up to 200 keywords, two-minute setup. If Reddit is the only place that matters to you, stop reading and just use this.
- Talkwalker Alerts is the closest free replacement for Google Alerts across news, blogs, and the broader web, delivered by email.
Build your own agent when these aren’t enough — and they stop being enough fast. You’ll want to build when you need sentiment filtering (only alert me on negative mentions), multi-source aggregation in one feed, alerts in Slack or Telegram instead of email, a logged history in a spreadsheet, or coverage of sources the free tools ignore (X/Twitter, review sites, niche forums). That’s where a no-code workflow earns its keep.
The tools you’ll wire together
You need three things: a place to get mentions, a brain to judge them, and a platform to glue it together and run on a schedule.
The glue (pick one):
| Platform | Best for | Starting cost | Watch out for |
|---|---|---|---|
| Make.com | True beginners; visual drag-and-drop; huge connector library | ~$9/mo (Core); free tier exists | Bills per operation — every module step burns credits, so busy workflows add up |
| n8n | People comfortable with a bit more tinkering; AI-agent workflows; cost control at scale | $20/mo cloud, or free self-hosted on a ~$6/mo VPS | Bills per execution (one full run = one execution), cheaper at volume, but the editor is less hand-holdy |
If you’ve never built an automation, start with Make for the gentler interface. If you expect heavy volume or want to self-host and stop worrying about per-step billing, n8n wins on cost. Both have ready-made brand-monitoring templates you can clone instead of starting from a blank canvas — search their template galleries for “Reddit monitoring” or “sentiment analysis” and you’ll find working starting points.
The brain: any LLM connector — OpenAI, Anthropic’s Claude, or Google’s Gemini. For sentiment classification you do not need a flagship model. A small, cheap one (GPT-class “mini” or Claude Haiku tier) reads a short post and returns a label perfectly well, for a fraction of a cent per mention. Don’t overpay here.
The sources: this is the part beginners underestimate, so be realistic:
- Reddit — easy. Reddit’s own search has a JSON endpoint; many templates hit it with no auth needed for basic keyword search.
- News & web — RSS feeds and free news APIs (like GNews or NewsData) plug straight in.
- X/Twitter — the honest part: X’s official API is expensive and restrictive. Most no-code builders route around it with a scraping service such as an Apify actor. It works, it costs a little, and it can break when the site changes. Budget for friction here.
- Review sites / niche forums — usually a scraper (Apify, Firecrawl) on a schedule.
Building it, step by step
Here’s the recipe we actually use, described in Make/n8n terms. The logic is identical on either platform.
Step 1 — Trigger on a schedule
Add a Schedule trigger. Every 15–30 minutes is plenty for most brands; hourly is fine if you’re not in crisis-prone territory. Don’t run it every minute — you’ll waste operations and hit rate limits for no real benefit.
Step 2 — Fetch new mentions
Add an HTTP request (or a native Reddit/RSS module) that searches each source for your brand name. Use specific keywords: your exact brand, common misspellings, and your main product names. Avoid generic words — if your brand is “Apple,” you’ll drown. Tighten with phrase matching where the source allows it.
Step 3 — Filter out what you’ve already seen
This is the step people skip and then get spammed by their own agent. Store each mention’s unique ID (post URL or ID) in a Google Sheet or the platform’s data store, and at the start of each run, drop anything you’ve already processed. No dedup = the same Reddit thread alerting you twelve times a day.
Step 4 — Let the AI judge each mention
Pass each new mention to your LLM module with a tight prompt. Something like:
“You monitor mentions of [BRAND]. Read this post. Return JSON: sentiment (positive/negative/neutral), category (lead / complaint / question / praise / noise), urgency (low/medium/high), and a one-sentence summary. Text: {{the mention}}”
Asking for structured JSON is the trick that makes the rest of the workflow easy — you can branch on those fields directly. This is where the system stops being a dumb alert and starts being an agent: it’s deciding, not just forwarding.
Step 5 — Route by what the AI decided
Add a Router / Filter:
- Negative + high urgency → immediate Slack/Telegram ping to your support channel. This is the alert that justifies the whole build.
- Lead or question → a “warm mentions” channel or your CRM.
- Praise → a feel-good channel, or a list of potential testimonials.
- Noise → log it silently and move on. No ping.
Step 6 — Log everything
Append every processed mention to a Google Sheet: date, source, link, sentiment, summary. Costs you nothing and gives you a searchable history plus raw material for monthly “what are people saying about us” reviews.
Wire those six steps, run a test with your brand name, and tune the prompt until the labels match your judgment. That tuning loop — read the misclassifications, sharpen the prompt — is 80% of getting a genuinely useful agent versus a noisy one.
The honest limitations
So you build with eyes open:
- Coverage is never total. No setup catches private groups, most of TikTok, Instagram comments, or closed Discords. Anyone who promises “every mention everywhere” is selling.
- Scrapers break. The X/Twitter and review-site pieces are the fragile parts. Expect to fix them occasionally.
- The AI mislabels sometimes. Sarcasm fools it. Keep a human glance on the high-urgency channel; don’t fully automate replies to angry customers.
- Costs are small but real. A few dollars a month for the platform, a few cents per mention for the model, plus scraper credits if you go beyond Reddit and RSS. Cheaper than any paid monitoring suite, not literally free.
If your needs are big — multi-language, TV/radio/print, deep historical analytics — a paid platform like Brand24 or Awario (roughly $39–$79/mo) is genuinely the better buy. The DIY agent shines for small-to-mid brands who want control, custom routing, and low cost. Be honest with yourself about which you are.
FAQ
Do I need to know how to code to build this?
No. Make and n8n are visual — you connect boxes and fill in fields. The only “code-like” part is writing the AI prompt, which is plain English, and occasionally pasting a search URL. If you can set up a Zapier zap or build a spreadsheet formula, you can build this. Cloning a template instead of starting blank removes most of the remaining difficulty.
How much does it cost to run per month?
For a small brand monitoring Reddit and news: realistically $0–$15/month. n8n self-hosted on a cheap VPS plus a small AI model can come in under $10 all-in. The cost climbs when you add X/Twitter or review-site scraping (scraper credits) and with mention volume (more posts = more AI calls). Still well below paid social-listening tools, which start around $39–$79/month.
How is this different from just using Google Alerts or F5Bot?
Those are pure alerts — they forward every match and leave you to sort it. Your agent adds judgment: it reads each mention, filters out noise, separates an angry customer from idle chatter, and routes each type to the right place with a summary. If a simple email firehose is all you want, use the free tools. If you want to be pinged only when it matters, you want the agent.
Your next step
Don’t try to build the whole thing today. Do this in 20 minutes: sign up for F5Bot, add your brand name and top product names, and start getting Reddit alerts now — so you’re covered while you build. Then open Make or n8n, clone a “Reddit monitoring + sentiment” template, swap in your keywords and your Slack or email, and run it once by hand. Get that single source working end to end before you add a second. A small agent that reliably catches Reddit beats an ambitious one that’s half-built and silently broken.
Sources: F5Bot · Talkwalker Alerts · Make vs n8n comparison · n8n Reddit brand-monitoring template · Brand24: Google Alerts alternatives · Awario: Google Alerts alternatives