A Weekly Listening Digest With an AI Agent, Brand24 and n8n

The short answer: if someone on your team reads Brand24 every Monday to tell leadership what was said about the brand and two rivals, let an n8n agent write that digest and keep a person between the draft and the inbox. The listening is the expensive part, not the agent: Brand24 Pro, the first plan with MCP access, costs $499 a month on monthly billing and n8n Starter $24, so $499 + $24 = $523 in all, of which the agent's share is the $24 (Brand24 pricing, n8n pricing, checked on 26 October 2026). For example, a digest that takes 90 minutes to write and 20 to check saves 70 minutes a week, 70 x 52 = 3,640 minutes a year.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

A listening feed is a river; a digest is the three things leadership should know by Monday lunch. What changed in how people talk about the brand, which mentions need a reply, and what the rivals were up to. The agent reads the week and writes those three. A person opens every mention the draft leans on before anyone else reads it. The gate is a Gmail draft: n8n creates it in the reviewer's mailbox, and it goes nowhere until someone presses Send.

What to settle first:

1. The plan, worked out. For example, a brand with 1,200 mentions a month and rivals with 2,000 and 1,800 need 1,200 + 2,000 + 1,800 = 5,000 a month. Team's 10,000 holds that, but Team has no MCP access, so letting an agent read the data costs $499 minus $349 = $150 a month more for Pro, per the pricing page.
2. The projects you already have. One project for the brand and one per rival, on the same language setting. Brand24 recommends separate projects when you want to compare brands (first project article); keywords and exclusions are covered in how to use Brand24.
3. The accounts. An n8n instance, on Cloud or self-hosted; a Brand24 login that can approve the MCP sign-in; and the reviewer's Gmail account.
4. Two digests by hand. Before building anything, write the digest yourself on two Mondays. Note the minutes and the items you chose. That is the bar the agent has to clear.

The workflow, from Monday trigger to sent digest

1. Trigger. Add a Schedule Trigger on the Weeks interval, Monday, 7am. Set the workflow's timezone in its settings; without one, a self-hosted instance runs on America/New York and n8n Cloud on the zone it detected at sign-up, or GMT, per the Schedule Trigger docs. Each firing is one execution (executions), so a year of digests is 52.
2. The agent and its brief. Add an AI Agent node with a chat model. In the prompt, ask for the digest of the seven days ending today, with the date mapped in. The System Message holds the format, five short sections: the brand's week, mentions that need a reply, each rival's week, the sources driving the talk, and volume against last week. It also holds the rules: every item links the mention it rests on, counts come only from tool results, sentiment only as Brand24 labels it, and "nothing notable" beats a padded section. Turn on Return Intermediate Steps, so the output keeps a record of what the agent asked for (Tools Agent).
3. Brand24 as the only tool. An AI Agent needs at least one tool (AI Agent node). Attach an MCP Client Tool pointed at https://mcp.brand24.com/v1/mcp, with OAuth2 authentication (MCP Client Tool). n8n's MCP credential registers itself with the server through Dynamic Client Registration, on by default (MCP credential), and Brand24 documents OAuth for any MCP client, custom agents included (Brand24 MCP). Under Tools to Include choose Selected and tick only the tools that read data. Test the credential before building further; if the sign-in will not complete in n8n, the same brief runs by hand in Claude, the client Brand24 documents step by step.
4. The draft lands in Gmail. Add a Gmail node with Resource Draft and Operation Create: the subject names the week, Email Type is HTML, the message is the agent's output, and To Email is the leadership list (draft operations). A draft sends nothing.
5. A person checks, then sends. The reviewer opens each linked mention and confirms it says what the digest claims. Counts get checked against Brand24's Comparison tab, which puts total, positive and negative mentions, reach and share of voice side by side for any range of days or weeks (comparison article). A sentence no mention supports comes out, however good it sounds. Then the reviewer presses Send.
6. The record. The executions list keeps each run with the agent's tool calls, and the Sent folder keeps every digest, so next quarter's question about when a complaint started has an answer.

For the n8n side in general, error workflows and publishing included, see how to use n8n.

Doing it by hand, for nothing

The free version is a Monday routine, and for a brand mentioned a few times a week it is the right one. Create a Google Alert for the brand, each product and each rival; the options cover how often, which types of sites, language, region and how many results, and matches arrive by email (Google Alerts help). On Monday, read the week's alerts, search the same names on the two networks your buyers use, and write the five sections above. Keep the alerts running after you buy anything; they are the cheapest second opinion you will have.

The agent can be tried free as well. n8n's Community Edition is self-hosted from GitHub at no charge (n8n pricing), and Brand24's trial runs for 14 days on Pro with 10 keywords and up to 30,000 mentions, refreshed every 24 hours (trial article). The trial article does not say whether MCP is included, so check that on day one. Two weeks is enough for two agent digests to compare with your hand-written ones.

What the paid tools add is reach and regularity. Google Alerts watches the web; Brand24 Pro's source list adds Reddit, LinkedIn, TikTok, YouTube, reviews and newsletters, with sentiment on each mention, on the pricing page. The agent adds a draft that is waiting at 7am whether or not anyone had time on Friday.

If your team already works in Agorapulse, its listening sits in an add-on called Advanced Listening, with competitive searches for up to 9 competitors (listening article). Its MCP connector reads Listening searches, filters them by source, date and sentiment, and defaults to the last seven days when no range is given, which matches a digest week (MCP article). The article documents Claude, ChatGPT and Copilot Studio, not n8n, so on that route a person runs the Monday prompt and edits the answer before forwarding it. The inbox side of the tool is in how to use Agorapulse.

What it costs, side by side

All prices were checked on 26 October 2026. At one execution a week, the digest uses a sliver of Starter's 2,500 a month (executions); the plan is sized by whatever else you run on n8n.

The three-project setup has a catch worth naming. Brand24's help centre recommends a separate project for each rival, and on Pro the AI Insights and AI Topics features cover up to 2 projects, while Business at $699 a month raises AI Insights to 5 and makes Topics unlimited (pricing page, first project article). So Brand24's own advice for a clean comparison puts a brand and two rivals one project past Pro's AI features, and the tier that lifts the limit bills $699 minus $499 = $200 a month more. Run the digest on all three projects during the trial, and judge the third project's section by what it says before treating that as a reason to upgrade.

How to know it worked

Three numbers, kept in a sheet beside the two hand-written baselines. First, the reviewer's minutes per digest. Second, the cut rate: the items the reviewer deletes as unsupported or beside the point. My rule of thumb: if more than 2 items in 10 come out two weeks running, rewrite the System Message before blaming the model. Third, the actions: replies sent, sales notes passed on, content ideas logged because of the digest. A digest nobody acts on is a newsletter with one reader.

Then a blind holdout, every fourth Monday. Before opening the draft, the reviewer lists the five things from that week's Brand24 feed they would have led with, then counts how many the agent missed. One miss is noise. The same kind of miss twice is a missing line in the brief, so add it.

If the hard part turns out to be the agent rather than the listening, building, documenting and handing over agents like this one is the agentic marketing work I sell.

FAQ

Can the digest go to Slack instead of email?

Yes. The Slack node's Send and Wait for Response operation, with Response Type set to Approval, holds the workflow until a listed approver clicks, provided the instance is reachable over public HTTPS and the credential carries the Slack app's signing secret (Slack approvals). Approve or Decline is the whole choice there, while a Gmail draft can be edited before it goes, which suits a digest.

How much history will the first digest have?

Not much. New Brand24 projects start with a sample of the last 30 days and collect forward from there (historical data article). Create the projects at least two weeks before the first digest, so the volume line has a previous week to compare with.

Does the agent need its own model API key?

Not on n8n Cloud's Starter plan, which lists AI models without API keys among its features on the pricing page. On the self-hosted Community Edition, you connect your own provider's key to the chat model and pay that provider for the tokens.