AI Reputation Monitor — Know the Moment Anyone Says Something Important About Your Brand Online
Built by PulsarCore ·
About this build
A Claude and Perplexity powered workflow that monitors your brand name, founder name, product and key competitors across Reddit, Twitter/X, Google News, Hacker News and niche forums in real time — then classifies every mention by sentiment and urgency, filters out the noise and delivers only the mentions that actually
What was built
Built this after missing a Reddit thread where 40 people were discussing my product and most of the conversation was based on outdated information from 6 months earlier. By the time someone forwarded it to me the thread had run its course, the community had formed their opinion, and the moment to correct the record had passed completely. The workflow runs every 90 minutes around the clock. Here is exactly what it does: Step 1 — Continuous search sweep via Perplexity Every 90 minutes n8n triggers a series of Perplexity searches across five source categories. Brand mentions — the product name, company name and founder name searched across Reddit, Twitter/X, Hacker News, LinkedIn and niche forums relevant to the industry. Competitor mentions — the same pattern for up to 3 competitors to catch conversations where someone is directly comparing or switching. Industry keyword mentions — 5 to 8 keywords tied to the product category to catch broader conversations where the brand is not mentioned but the opportunity to add value exists. Crisis signals — specific searches for the brand name combined with words like scam, broken, cancelled, avoid, refund, lawsuit, security, data and similar terms that would indicate an emerging reputation problem. Positive signals — the brand name combined with words like recommend, love, switched to, replaced, best, favourite to catch organic advocacy that could be amplified. Step 2 — Mention classification via Claude Every raw Perplexity result gets passed to Claude with a structured prompt that classifies it across four dimensions simultaneously. Sentiment — positive, negative, neutral or mixed, with a confidence score. Urgency — rated 1 to 5 where 5 means requires a response today, 3 means worth monitoring, 1 means informational only. Action type — respond directly, amplify organically, flag for legal or PR review, no action needed, or save for product team. Authenticity — is this a genuine organic mention or does it look like a competitor post, spam, or an automated account. Step 3 — Intelligent filtering Only mentions that clear two thresholds get pushed to any notification. Sentiment must be clearly positive or clearly negative — neutral noise gets logged to Airtable but generates no alert. Urgency must be 3 or above. This filtering is the thing that makes the system actually usable. Without it you get notified about everything and start ignoring the notifications within a week. Step 4 — Tiered delivery Urgency 5 mentions — anything that looks like a crisis signal, a viral negative thread, or a direct attack — trigger an immediate Slack message to the founder with the full mention text, the classification and a suggested response written by Claude in the brand voice. Urgency 3 to 4 mentions get batched and delivered as a Slack message twice per day — 9am and 3pm — with a short summary of what was found and what type of action is suggested. Urgency 1 to 2 mentions get logged to Airtable only and included in a weekly digest email every Monday morning showing volume trends, sentiment shift over time and any emerging themes. Step 5 — Airtable history and trend tracking Every classified mention regardless of urgency gets written to Airtable with the source, the text, the sentiment score, the urgency rating and the date. After 4 weeks of data this becomes genuinely valuable. You can see whether sentiment is shifting week over week, whether a particular platform is generating disproportionate negative mentions, and whether competitor mentions are accelerating in a way that suggests a new campaign or product launch. What broke during development and how I fixed it: The crisis signal searches were generating too many false positives early on. Words like broken and cancelled are common in completely unrelated contexts. Fixed by requiring the crisis keyword to appear within 10 words of the brand name in the search query, and adding a secondary Claude filter that checks whether the context is genuinely about the brand before flagging as urgent. The competitor mention search was initially too broad and kept returning irrelevant results for competitors with generic names. Fixed by always including the competitor name alongside their primary product name and their most distinctive feature in the search query — three identifiers together are much more precise than one name alone. The Slack notification format went through 4 iterations. The first version was too long and people stopped reading past the headline. The final version is a 5 line card — mention source, one sentence summary, sentiment, urgency rating, suggested action. Everything else goes to Airtable. Short enough to read in 10 seconds, specific enough to know immediately whether to act. Results after 10 weeks: Caught an urgent thread on Reddit where a user was describing a bug as a scam and had already gathered 28 upvotes before the alert fired. Responded within 20 minutes with an explanation and a fix offer — the thread ended with the user updating their post to say the issue was resolved and the response was fast. Without this build we would have found that thread the following week after someone forwarded it. Caught a positive mention on Hacker News where someone listed the product in a comment that was getting significant engagement — was able to join the thread authentically and add additional context that generated 3 new signups directly. The weekly Airtable data showed that Reddit was generating 3 times more negative mentions than Twitter/X despite Twitter/X having higher total mention volume. That insight changed where we prioritised community engagement completely. Will share the full n8n workflow JSON and Claude classification prompt in comments for anyone who wants to deploy this.