Project Overview
Forekast is an 8-person content marketing agency serving B2B SaaS companies. Their writers were spending 4 hours per brief on research — opening 18 tabs across Reddit, Hacker News, Perplexity, and competitor blogs to figure out the angle before writing a single sentence. Their Copy.ai Pro subscription helped with drafts but did nothing for the research phase, which was the actual bottleneck.
NodeSparks built a content research agent that takes a topic + target audience as inputs and returns an outline with source links, a recommended angle, and the three strongest hooks pulled from real discussions. Research time dropped from 4 hours to 1 hour per brief. Forekast ships 3x more briefs per week with the same headcount.
The Problem
The agency had tried both ends of the spectrum. At one end: human researchers spending half a day per topic to land a good angle. Quality was high, throughput was low, margins were tight. At the other end: AI drafting tools like Copy.ai and Jasper that wrote competent paragraphs but had no idea what was actually being discussed in their client's niche this week. The middle — research-as-a-step that an AI could do well — was unowned.
Cost of Copy.ai Pro at 5 seats: $594/yr. Research time cost (at fully-loaded writer hourly): $43k/yr across the team. The AI subscription was a rounding error; the research time was the real expense.
The Build
The agent takes a topic (e.g. "B2B SaaS pricing page best practices") and target audience descriptor (e.g. "SaaS founders, 20–100 employees") and runs four parallel searches: Reddit (relevant subs), Hacker News, Perplexity (general web), and a configurable list of competitor blogs. It deduplicates, ranks by signal (recency + engagement), and returns: (1) a top-line summary of what is currently being discussed, (2) three angle suggestions with the source threads, (3) a draft outline with H2s and the source link per section.
Total build time: 11 working days. Cost: $3,400. Forekast killed their Copy.ai subscription on day 12 — the agent’s outline output was already cleaner than what Copy.ai was producing.
Stack
- Runtime: Claude Code with web fetch and parallel sub-tasks
- Sources: Reddit API, HN Algolia search, Perplexity API, direct fetch for competitor blogs
- Ranking: composite score from recency + thread upvotes + comment count
- Output: Markdown brief delivered via Slack to the writer's DM
- Hosting: client's GitHub repo + Vercel cron
Outcome
- Research time per brief: 4 hrs → 1 hr (75% reduction)
- Brief throughput: 8/wk → 24/wk with same team
- Killed Copy.ai Pro: $594/yr saved (small but real)
- Recovered writer time worth ~$32k/yr at fully-loaded cost
- Margin per client engagement improved by ~18% (more billable output per head)
“We were paying $49 a month for Copy.ai Pro and still spending four hours per brief on research. Now we spend an hour. The agent scrapes Reddit, HN, and Perplexity for whatever angle we feed it, drafts an outline with source links, and flags the strongest hook. Our writers do what they should be doing — writing — and not opening 18 tabs to figure out what the angle is. We ship three times more briefs a week with the same headcount.”
We replace the SaaS stack and the manual ops work eating your team's time. One custom system, owned by you.