Seph
Hawkins

I inherited a static, blog-first site with real technical debt and editorial drift. Pricing and company details didn't match from one page to the next. That cost us a weaker Google rank, fewer and fuzzier mentions in AI Overviews, and a softer conversion rate.

I fixed it on two tracks. First came quarterly blog health checks in ClickUp, focused on the highest-value pages. I picked these by reading Google Search Console data against each page's SEO history. Second, I rewrote moveBuddha's internal style guide and turned its rules into four Claude Cowork plug-ins the whole team could run. Each one handled a single check: voice consistency, subject-matter expert (SME) input, conversion best practices (CRO), and data accuracy. Prices and company facts finally lined up across the site.

The brand reads as one clear authority on moving again. It now shows up more often, and more clearly, in AI search for both LLM chatbots and Google's AI Overviews, roughly from 0–1 average AIO appearances to 7+. Priority pages climbed in Google rank, increasing approximately 5 positions on average within a week. Consistent facts and one voice are what readers and AI engines trust, and trust is what lifts rank and conversions.

I built residential solar from a minor topic into a flagship vertical. Over my time here I wrote hundreds of solar pieces across five-plus media partner websites, including guides on installation, costs, federal and state tax credits, and SRECs, and grew the category into one of the top three we supplied to those partners.

Getting there meant turning dense, scattered information into something a homeowner could act on. I synthesized solar industry and incentive data that was technically public but hard to parse, ran surveys of 3,000+ residential solar customers, and interviewed 10+ subject-matter experts to ground every recommendation. I also helped develop programmatic local campaigns that adapted the category to specific markets and search demand.

The vertical drove enough qualified leads to our solar partners to stand on its own as a flagship category. Clear, trustworthy explainers are what turn a curious homeowner into a lead, and that's what moved solar from a side topic to a pillar of what we delivered.

Here are a few representative pieces:

This is where I learned to produce at volume. I wrote articles, reviews, and round-ups across consumer verticals. Each one was accurate, search-optimized, on brand, on time, tied to a real question a reader was already asking, and built to perform once it shipped. To hit that bar piece after piece, I built a tight, repeatable production process and held myself to it.

I took constructive feedback to learn each brand's standard, then revised until the work met it. I brainstormed copy and visual ideas with the editorial team and our digital marketing partners, and made the case for which new ideas were worth building. From there I sourced the supporting images, charts, graphs, and infographics, loaded everything onto the sites, and QA'd every asset before it went live. I also kept older pieces current as the facts moved, such as pricing, promotions, and retail partnerships, and helped with outreach to media, brands, and influencers.

The home-services verticals grew faster than anything else we ran. They grew enough that Home Solutions was spun out of Three Ships as its own company, built to manage the revenue and performance this content created. A disciplined line: write, revise, illustrate, publish, QA, and update. This is what turned a content program into a business.

Here's a representative piece:

Some editorial work is judgment. A lot of it is assembly. It's the same shape of article, built from the same kinds of inputs, over and over. I built agents to take the assembly off people's plates so the judgment got more time and effort.

Two systems came out of this at moveBuddha. The first sits behind the Editorial Consistency System in Content Producer above: Claude agents standing in for fixed points in our review process, which is what let us push roughly twice the volume of revised pages per week during the site recovery. The second is the one drawn here. It's an n8n pipeline for our Better Moves project. Readers submit their move details through a survey that lands in a Google Sheet. The automation pulls each submission, runs consistency checks across the fields, and hands the data to a set of Claude agents that draft against a fixed article outline and our style guide in markdown. A few minutes later there's a rough draft. A writer takes it from there, with a human read, QA, and the visual elements.

The point was never to publish on autopilot. It was to skip the blank page. With the structure, the survey data, and the checks doing the first pass, we turned out Better Moves articles at two to three a week, up from one every couple of weeks. The writer still owns the final piece. They just start from a real draft instead of a blank document.

Here's a representative piece:

Good home-services content lives or dies on numbers most writers can't get: what a solar install actually runs, what people really pay movers, what a window replacement costs once the quote is in. We worked to get those numbers. They sat in a database of more than 5,000 consumer survey responses across solar, moving, and replacement windows. The problem was using them fast enough to matter.

So I built custom GPT and Claude models that read straight from that survey data. A writer could ask for the average cost in a category, run the formulas behind it, and pull real, defensible pricing in the time it used to take to track down a single figure. That last part is the hard part in home services, where honest pricing is tough to source at all. The models did the synthesis, such as calculating averages, breakdowns, and scaled cost data, and the prompt workflows behind them turned a slow manual research slog into something close to instant. They replaced hours of digging through spreadsheets for one stat.

Two things came out of it. The data-backed pages were the kind that Google's E-E-A-T guidelines reward. We could rapidly share first-hand experience and numbers a reader can trust, and we saw it in the rankings and in higher conversion, as readers believed what they read. The second was that a new writer could open a category and get an immediate, grounded read on where each customer cohort actually sits — their needs, their budgets, and their pain points. They could get straight to work rather than guessing their way in. Here's a published piece built on that synthesized survey data:

Performance data for a vertical never lives in one place. Search numbers sit in Google Search Console, conversion numbers sit in a stack of CRO tables, and the vertical-specific teams who need to act on them don't have time to stitch it all together every week. By the time a trend was obvious enough to notice on its own, the best moment to react to it had often already passed.

Toward the end of my time at Home Solutions I learned Zapier and built a set of vertical-specific reporting automations to close that gap. Each one pulled from a sheet linked to GSC data, plus the CRO tables, and pushed everything into a fresh working sheet. There a ChatGPT step synthesized the raw rows into a handful of clear datapoints, and a Claude model translated those datapoints one step further, into a precise, plain-language read on what had changed and what to do about it. The finished synthesis went out as a scheduled email to each vertical's team.

The point was speed of reaction. Instead of a dashboard nobody opened, each team got a short written brief that told them where performance was moving and handed them a concrete starting place, so they could devise plans and respond to trends in the same week the data surfaced them rather than a month later. It's the analytics-and-reporting layer of a marketing team (KPI tracking, week-over-week reads, and A/B-test calls), wired to run on automation instead of a manual weekly pull. Here's a simulated, anonymized example of what landed in a team's inbox:

Most editorial calendars are just a list of due dates. They tell you what publishes and when, not whether any of it moves the business. I plan the other way around. The calendar starts from the goals leadership actually measures, and every planned piece has to earn its slot.

At Home Solutions the operating rhythm was OKRs, the objectives and key results a team commits to for a quarter. I proposed a set each quarter, then built the editorial calendar to serve them. A pillar refresh mapped to a ranking goal. A conversion test mapped to a lead goal. Each week I tracked progress against those numbers instead of against publish dates, so a piece that wasn't pulling its weight got caught early. I carried the same method into moveBuddha, where the planning ran across thousands of programmatic pages rather than a handful of posts.

Run this way, the calendar becomes a decision tool instead of a wish list. Weekly tracking shows what's off target while there's still time to fix it. Quarterly review keeps the team pointed at what leadership cares about, and any work that doesn't ladder up to an objective gets cut or rethought. That's how content stops being busywork and starts being accountable.

A Special Attention Review (SAR) is the process I ran at Home Solutions when a page or vertical stopped performing. Something specific set it off: leads or conversions slipping, a KPI or OKR target getting missed, or content losing ground because it no longer matched search intent or showed enough first-hand experience and expertise (EEAT). Rather than patch the symptom, I opened a review.

I owned the review end to end. First I pulled the data: Search Console numbers, the conversion tables, and the page's own history. From there I traced the drop to a root cause instead of stopping at the surface number. Then I brought in the writers whose pages were in scope, because the people who made the content know things the analytics don't. The result was never a list of complaints. It was a clear read on what went wrong, why, and what to do first.

Leadership got something they could act on the same day: the cause, the fix, and the order to do it in. Because every SAR traced back to a real root cause, whether an intent mismatch, thin expertise, a conversion problem, or a gap in the process, the fixes addressed the actual issue and the same failure was less likely to return. The reviews did double duty, too. They taught the team how to spot the warning signs on their own.

Building a new vertical isn't a writing problem. It's a coordination problem. When Home Solutions made clean energy a priority, the work touched content, SEO, analytics, subject-matter experts, and more than five media partner sites. If those groups don't point the same direction, the launch ships as a pile of disconnected pages.

My job was to get everyone aligned before anything published. We settled the positioning and the conversion goals up front, so SEO, content, and the partner sites were all building toward the same targets. Then I gave every team one shared foundation of facts: dense public data on solar costs and incentives, more than 3,000 customer survey responses, and over ten interviews with installers and other experts. From there the rollout had a clear order. Lock the message and the KPIs, build the content and the programmatic local campaigns, ship across the partner sites, then measure against the lead and conversion goals.

Because the functions were aligned first, clean energy launched as one coherent push instead of scattered posts. It grew into one of Home Solutions' top three verticals across the partner network and brought in enough qualified leads to become a pillar of the business. The lesson stuck with me. The hard part of a launch is rarely the content. It's getting every team to agree on what "done" looks like before the work starts.

Allergy sufferers make the same calls every day: whether to go out, whether to medicate, whether to open the windows. Most of them do it on guesswork or a forecast that flattens a whole metro into one number. MyPollenPal, built with Shawn Hill, gives them a local read instead, today's tree, grass, and weed pollen for their exact ZIP, in plain language, with clear guidance on what to do about it.

Every reading routes to an action. The site turns raw counts into "stay in this morning," "take your medication before you head out," or "this afternoon is a safe window," and points people who need more help to a hand-picked directory of board-certified allergists. Seasonal guides, daily worst-city rankings, and per-city pages round it out for anyone managing a real allergy problem, not just checking a number.

The second half of the product is the alert. A free daily email fires only when it matters, warning subscribers before tree, grass, or weed pollen spikes in their area. More than 1,500 people are signed up for it.

The bet was simple: allergy sufferers deserve better than a forecast number buried three taps deep, and a two-person team could ship something more useful if the daily busywork ran itself.

So the site is largely run by a system of Claude agents wired to a GitHub repository. Each day the agents pull fresh data, pollen counts from Google's Pollen API plus weather and air-quality readings from WeatherAPI, NOAA, and the EPA, then update every city and state page, refresh the worst-pollen rankings, and send the alert emails, with no manual step in between. New surface area gets built the same way: I use Claude Code against the repo to stand up new sections, seasonal guides, the allergist directory, per-city pages, then hand them to the agents to keep current.

It is live and growing, with more than 1,500 alert subscribers and coverage across all 50 states and hundreds of cities. What is next is depth: more guides, a fuller allergist directory, and tighter, more personal alerts, most of which the agent system can roll out without rebuilding the site by hand.