RefreshAgent Real-world case study
Observed Google Search Console data · October 2026
I use a recurring AI-agent workflow to find SEO opportunities, change the code and measure what happens next. Here's what it looks like on SnorkelForecast—including the parts that are still experiments.
September 10–October 9 vs August 11–September 9, 2026. This measures site growth, not traffic caused by an individual agent.
Two equal, consecutive 30-day Search Console windows. The bars represent totals, not a made-up daily trend.
1,297 more clicks across the site
32,983 more appearances in search results
We verified the numbers using the connected Search Console property for snorkelforecast.com, with comparison periods ending 9 October 2026. Google's search data can lag. Seasonality, indexing, new pages and earlier development work can also affect growth. No controlled experiment isolates an agent's contribution.
These are Google Search Console clicks, not GA4 sessions or booked tours. The site's consent setup and attribution changes also make historical GA4 comparisons less suitable for this claim.
Use the AI you already have. Give it reliable data, a bounded task and a way to verify changes.
RefreshAgent connects to Google Search Console and GA4. The agent sees the same page, query and period evidence we use to make human decisions—not guessed search volumes or generic AI SEO advice.
Compare opportunity size, intent and the site's ability to add original information. Favor computed snorkeling facts and better answers over producing more near-duplicate travel articles.
A scheduled Hermes SEO workflow and coding agents inspect the Django repository, make scoped changes, run checks, and record the work in Git. Normal deployment follows the site's existing delivery pipeline; code changes aren't proof of production impact.
Experiments include a baseline, guardrails and 14-/28-day reviews. Search Console needs time after publication; an opportunity detected today shouldn't be marketed as an uplift tomorrow.
Concrete, page-level work. Each shows an observed opportunity and the change being tested—without inventing a result.
Query: “maui snorkel report”
RefreshAgent flagged a high-impression search term with room to improve its click-through rate. Work on the Maui forecast report focuses on clearer search snippets and an unmistakable current-conditions verdict using stored marine data.
Outcome: latest changes under review; no uplift claimed yet.
Inspect the live Maui report ↗Query: “best places to snorkel”
The best places to snorkel page was visible in results but not attracting this query's clicks. Instead of making another listicle, the work focuses on useful destination decisions, backed by real conditions and derived data where coverage allows.
Outcome: search-intent and consolidation changes are recent; measure after indexing.
Inspect the destination guide ↗Example query/page observations: Search Console 30-day data ending 9 October 2026. Metrics identify opportunities; they do not establish that the experiments succeeded.
SnorkelForecast is a data product. It combines marine forecasts, water temperatures and historical conditions to answer questions like where or when to snorkel. AI can help decide which answers need improving, but it doesn't get to invent ocean conditions or safety guarantees.
Produce comparisons and derived facts from defensible sources, not paraphrased destination blurbs.
A Search Console CTR gap is observed. An explanation for the gap is a hypothesis until tested.
Bounded commits, correctness tests and logged experiments matter more than an agent publishing endlessly.
Start with a real decision, not a vague request to “do SEO.”
Analyze my site's Google Search Console performance for the last 30 days versus the previous 30. Find three opportunities supported by exact page and query evidence. Prioritize changes that create genuinely useful information. Recommend one bounded improvement, the metric to watch and a 28-day review plan. Never invent data or claim a result before measuring it.
What RefreshAgent does: securely supplies your Search Console/GA4 evidence and deterministic SEO analysis. Your existing ChatGPT, Claude Code, Codex or other supported MCP client does the reasoning with its own model access. The hosted RefreshAgent agent is optional.
What it doesn't do: guarantee rankings, automate every site change by itself or establish causality from observational traffic charts.
MAKE YOUR AI MORE USEFUL
Give your AI real search performance, discover one useful improvement, then measure the result. That's the part worth automating.