Teardown No. 31 · SEO

How Maddy Osman built a Zapier agent that drafts her journalist query responses

Maddy Osman runs the content agency The Blogsmith. Journalist query services send her daily media opportunities, and many went unanswered because she lacked time to evaluate them and draft strong responses. She built an agent in Zapier Agents to handle the work. A Gmail label triggers the agent, and the agent parses each query and scores it against her topic guideline. It then sends the query to a small Replit app that searches her published content in Pinecone. The agent logs the result to a Google Sheet and posts a ready-to-edit draft to Slack. She edits and sends every response herself. She reports a domain rating of 72 on Ahrefs and 3,700+ backlinks without paying for link building. She says she went from one or two responses per week to several per day.

Maddy Osman · @maddyosmanVerified Oct 17, 2026Markdown copy

The problem

Journalist query services like Qwoted, Source of Sources and Featured “send out daily media opportunities.” For a while, many of those opportunities went unanswered. She did not have the time to evaluate which ones were worth pursuing and to draft responses strong enough to send. She uses these opportunities to build thought leadership. She also uses them “as a proof of concept for using content to achieve SEO successes for clients.”

Before the agent, a query would “sit unread for a day or two while client work took priority.” It then got a rushed response or no response at all. A longtime team member knew her content library well, but finding the strongest source material still took time and mental effort.

The blueprint (as she describes it)

  1. Pre-filter with a Gmail label. The workflow starts with a Gmail trigger. It fires whenever a new email arrives with the “PR Opportunities” label. She applies this label to any email from the journalist query services she subscribes to.

  2. Parse the email. The agent extracts “the query itself, relevant category, any requirements or qualifications they specified, contact information, and the due date.” The rest of the workflow runs on this structured data.

  3. Score against a topic guideline. She wrote a scoring guideline that assigns score ranges to subject areas:

    • AI, content marketing, SEO and entrepreneurship score between 80 and 100.
    • Tangentially related topics score between 20 and 40.
    • Unrelated topics score between 0 and 10.

    The agent uses keyword matching and contextual matching to reach a score. It filters out low-scoring queries before going further.

  4. Retrieve from her own content (RAG). For each query that scores high enough, the agent sends the query text as a webhook POST request to a Replit project she built. The Replit project connects to her Pinecone vector database. It searches Pinecone for the most relevant chunks from her content library and returns those passages with source citations. She uses Replit because it gives the app a public URL. That URL is the bridge between Zapier and Pinecone, so she needs no separate backend. Her aim is to ground each draft in content she has already published “rather than generating a generic response from scratch.” The pipeline that uploads her content and keeps the knowledge base current is covered in a separate RAG guide. That guide is not included in this source.

  5. Log to a Google Sheet. Each row holds the query details, contact information, source, due date, relevance score and its reasoning, and the RAG response. She uses the sheet to track what was sent, follow up on placements and review how the scoring performs over time.

  6. Deliver the draft in Slack. The agent posts a summary to her #blogsmith-expert-roundups Slack channel. The message includes the original query details, the due date, the source and the draft response.

  7. Human sign-off. The agent stops at the Slack message. “Nothing goes to the journalist automatically,” and she says that is deliberate. She reads the draft, adjusts the tone and checks that it answers the question. Some drafts need a full rewrite of the angle. She then copies the draft into Gmail and sends it using the contact details from Slack. She notes the workflow could instead create the draft directly in Gmail. In her words: “An automated response that misses the point of the query is worse than no response at all.”

She chose Zapier Agents “for its ease of use, flexibility” and for its integrations. In Zapier Agents she writes instructions in plain language and connects them to existing integrations without code.

Stated results

  • A domain rating of 72 on Ahrefs and 3,700+ backlinks, “all without paying for link building.”
  • The agent handles 95% of the query response process.
  • Before the agent: one or two journalist queries per week. Now: “several strong opportunities per day.”
  • Queries are parsed, scored and drafted “within minutes of arriving.”
  • She says it cut what she was spending on manual research and drafting for each response. No figure is stated.
  • Number of responses sent, placements won or backlinks earned through the agent specifically: None stated.

Caveats

  • The source is on ActiveCampaign’s “autonomous” stories site (autonomous.activecampaign.com), published 2026-06-18. That site is vendor-run. ActiveCampaign is not part of her stated stack.
  • The DR 72 and 3,700+ backlink figures are framed as the result of answering these opportunities. The headline says “landing 3.7k backlinks.” The source does not separate what the agent produced from earlier manual work, and it gives no timeframe.
  • The source does not name the AI model the agent uses.
  • She also wrote about the same agent on the Zapier blog, which is the platform it runs on. The source states no commercial relationship with Zapier, Pinecone or Replit.
  • She says the drafts usually need minimal editing, but some need a full rewrite of the angle.
  • Handles: X @maddyosman (verified 2026-10-17: linked from theblogsmith.com/team). LinkedIn linkedin.com/in/madelineosman (verified 2026-10-17: same page).

Sources

Want agents like these working your distribution? See the Agent Stack Teardown. Get new teardowns weekly: subscribe (double opt-in).

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.