obah sylva

How I Built AI Agents That Replaced The Trumpet’s Technical Team — and Quadrupled Our Google Traffic

Share this article

In June 2026, thetrumpet.ng was getting about 73 clicks a day from Google. In the last week of October, it averaged more than 350. I didn’t hire anyone to get there. I built a small team of AI agents that now handle the technical and repetitive work of running a digital newspaper.

In about five months, a set of Claude agents, Blotato and Grok took over jobs that used to need a web developer, an SEO analyst, a social media manager and a research desk. I didn’t buy a growth hack either. The agents handle the technical work so the people at The Trumpet can spend their time on journalism.

This is the full story: the tools, the agents, the SEO decisions that actually moved the needle, what broke along the way, and the parts AI still can’t do.

The numbers first

All figures come from Google Search Console, web search only, for thetrumpet.ng.

PeriodAvg. daily Google clicksvs. June
June 2026731.0x
July 2026821.1x
August 2026901.2x
September 20261782.4x
October 20262433.3x
Last 7 days (28 Oct – 3 Nov)3534.8x

Other signals moved in the same direction:

  • Impressions went from 54,088 in June to 155,985 in October, almost three times as many.
  • Average position went from about 19 (the bottom of page two) in June to about 10–11 in October (the edge of page one).
  • Peak days got bigger. On 11 September we had 894 clicks in one day. On 3 November we had 719.

These numbers cover Google search clicks only. They leave out the readers who now come in from Facebook, X, Threads, LinkedIn, Instagram and Pinterest through our automated distribution, so total traffic growth is higher than the table shows.

The problem every small Nigerian newsroom knows

Running a news website in Nigeria means doing two jobs. One is journalism. The other is a steady stream of technical work that never ends:

  • Formatting and publishing stories, fixing bylines, adding internal links and keeping WordPress healthy.
  • Posting every story to six or more social networks, each with its own caption style.
  • Watching beats like education, recruitment portals and state politics for anything new.
  • Tracking keyword rankings, reading Search Console and finding pages that are losing clicks.
  • Writing weekly reports for the editor-in-chief.

A proper technical team for all of that would include a developer, an SEO specialist, a social media manager and one or two researchers. For an independent publisher, that payroll is hard to justify, and the people you train often leave for bigger outlets.

I’m a software engineer, so I asked a different question: what if every one of those jobs were a written procedure an AI agent could run every day without being reminded?

The stack

Claude (Anthropic): the brain and the hands. Claude does the reasoning and the actual work. Three features made it workable:

  • Connectors (MCP). The Model Context Protocol lets Claude reach real systems instead of just talking about them. Through connectors, Claude can work directly in our WordPress site, read Search Console and keyword data through OpenSEO, send reports from Gmail and post through social tools.
  • Skills. A skill is a written playbook: a step-by-step procedure Claude follows the same way every time. Each agent below is a skill.
  • Scheduled tasks. Skills run on a timetable, so the work happens whether or not I’m at my desk.

Blotato: the distribution layer. Blotato is a social media publishing platform built for AI agents, with an API and an MCP server. Instead of a person logging into six dashboards, our agent sends each story to Facebook, X, Threads, LinkedIn, Instagram and Pinterest through Blotato.

Grok (xAI): the real-time X pulse. Grok’s biggest strength is live access to posts on X. In Nigerian news, X is often where a story breaks first, so Grok is our early-warning system for what people are saying and sharing right now.

Supporting tools: Google Search Console, Google News Publisher Center (where we’re listed as The Trumpet Newspaper), Metricool as a backup scheduler, and OpenSEO for rank tracking.

The agents, one by one

Each agent took over a specific job that a person used to do, or that simply wasn’t getting done.

AgentJob it replacedWhat it does
Publishing agentWeb editor / WordPress developerPublishes articles on the site with the correct byline, formatting and internal links, and refreshes our live trackers from newly published stories.
Social distribution agentSocial media managerFinds newly published articles that haven’t been shared yet, writes a caption for each network, posts through Blotato, and records what was shared so nothing goes out twice.
Delta social monitorSocial listening deskScans public conversation on X and Facebook among people in Delta State across politics, governance, religion, entertainment, technology and gossip, then emails a story-leads report to the editor-in-chief.
Education reporterBeat researcherWatches the Ministry of Education and its agencies, universities, polytechnics, colleges of education and the unions, and emails a daily leads report.
Opportunities monitorService-journalism deskTracks N-Power payment updates, NELFUND student loans, INEC ad hoc staff recruitment, CDCFIB, Police and FRSC shortlists, and empowerment programmes. These are among our highest-demand topics.
Delta rank trackerSEO analystChecks our positions every week for eight priority Delta State keywords, shows the cost before running, and reports the week-on-week change.
Weekly SEO reportAnalytics leadEmails the editor-in-chief our Search Console numbers, tracker updates and the pages that are getting impressions but too few clicks.

Together they cover publishing, distribution, newsgathering leads, SEO monitoring and reporting, which is everything a technical team used to do for us.

The SEO decisions that actually moved traffic

Automation by itself doesn’t quadruple traffic. Doing the right things consistently does, and the agents made consistency cheap. These are the changes that mattered most.

1. One intent, one URL

We had a cannibalisation problem: several stories competing for the same search, so Google trusted none of them. In November we built a keyword map where each search intent is owned by one URL. The clearest example: we stopped publishing a new dated “Dollar to Naira Today” post every day and now update a single /dollar-to-naira-today/ tracker. Signals that used to be split across dozens of pages now build up on one.

2. Live trackers instead of one-off stories

For recruitment portals and government schemes, readers want the latest status, not yesterday’s news. Evergreen tracker pages that the agents refresh whenever a new story is published stay current and keep earning rankings.

3. Location hubs

We built a Delta State news hub, plus hubs for Ughelli, Sapele and Isoko. Hubs give Google a clear picture of what we cover and give readers a reason to stay on the site. Local keywords like Asaba, Warri and Ughelli are where a regional paper can beat national brands.

4. Clean author identity

We merged duplicate author profiles into one consistent byline. A small fix, but it matters for how search engines judge expertise and trust.

5. Competitor-led targeting

We named who we’re trying to beat: Ripples Nigeria and The Whistler on politics keywords, and CareerPointer and RecruitmentCrib on recruitment-portal keywords. The agents measure us against them every week.

To be clear, the agents didn’t come up with all of this strategy. What they did was make it possible to carry it out every day, which no small team I could afford had managed before.

Working with Claude: what the collaboration looked like

Claude has been part of this from the start, less as a tool I query and more as a colleague I train. Nearly every agent was built the same way:

  1. I describe the job in plain language. For example: “Share every new article to our socials, but never post the same story twice.”
  2. Claude does it once while I watch. We find the problems together: a wrong handle, a caption that’s too long for X, a story that was already shared.
  3. Each correction becomes a rule. Our X handle is @TrumpetNews24. Our writer’s byline is “Omotayo Edubi”, in that order. The agent learns these once and doesn’t forget them.
  4. The procedure is saved as a skill. It now runs the same way every time.
  5. The skill goes on a schedule. At that point a person no longer has to do the job.

The biggest change for me was how I think about the work. I stopped asking “who should do this?” and started asking “can I write this down clearly enough for an agent to do it?” If the answer is yes, it gets automated.

What went wrong

The process wasn’t smooth, and anyone copying it should expect to hit these:

  • Duplicate posts. Early social automation risks posting the same story more than once. The fix is a shared record of what has already been published, which the agent checks every time.
  • Small identity errors spread fast. A wrong social handle or a reversed byline gets repeated hundreds of times by automation. Lock these facts down early.
  • Our own content competed with itself. Daily dated posts felt productive but split our rankings. More content is not always better.
  • Access is the hidden bottleneck. Our Search Console lives on a different Google account from our main workspace, and every connector needs proper permissions. Expect to spend real time on setup.
  • Real-time social data needs checking. X moves fast and misinformation moves with it. Anything from Grok or the social monitor is treated as a lead, never as a fact, until a journalist verifies it.

What AI did not replace

The headline says the agents replaced our technical team, and that’s accurate. They did not replace our journalists.

The agents find leads, but people decide what is news. The agents publish and distribute, but people report, verify, call sources, weigh legal risk and take responsibility for every word under our name. If anything, automation made that human work more important, because it’s now what sets The Trumpet apart.

How to copy this in your newsroom

  1. Write down every recurring technical task for two weeks: what it is, how often it happens and how long it takes.
  2. Connect your CMS and Search Console to Claude through connectors.
  3. Do each task once with Claude interactively and correct it as you go.
  4. Save each working procedure as a skill, including every correction.
  5. Schedule the skills so they run without you.
  6. Add a distribution layer such as Blotato so every story reaches every network.
  7. Use Grok for real-time X monitoring, and verify everything it surfaces.
  8. Measure every week in Search Console, and fix one structural SEO problem at a time: cannibalisation, hubs, trackers, author identity.

The bottom line

Five months ago, The Trumpet’s growth depended on how many hours a few people could put in. Today it runs on a system that publishes, distributes, monitors, measures and reports every day, and our Google traffic is roughly four times what it was in June.

The lesson isn’t that AI replaces people. A small newsroom that writes its work down clearly can now run like a much bigger one.

If you run a newsroom or publication and want help with the human side of this, from editing to content structure, see my news editing services and content and AEO strategy services. For more on how The Trumpet approached search, read Is Hiring an SEO Company Worth the Money? and Preparing Your Website for AI Platforms That Now Drive Traffic Like Search Engines.

1 thought on “How I Built AI Agents That Replaced The Trumpet’s Technical Team — and Quadrupled Our Google Traffic”

  1. Pingback: How to Make Your Site 3x Faster: The Trumpet's WordPress Story

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top