RefreshAgent Real-world case study
Observed Google Search Console data · through 27 October 2026
A 90-day study of search visibility, page-level contributions and documented changes to SnorkelForecast. This is evidence of website growth—not proof that AI agents caused it.
30 July–27 October versus 1 May–29 July 2026. This measures total website search growth, not traffic caused by a particular agent or deployment.
Two equal, consecutive 90-day Search Console periods, measured for the whole domain. These bars show period totals, not a fabricated daily trend.
6,138 net additional clicks across the domain.
245,614 more search-result impressions.
A surprising result
Search clicks grew 6.14×, yet aggregate click-through rate fell slightly, from 2.62% to 2.52%. The overall click increase is associated with substantially more impressions, not a higher aggregate CTR. Changing query mix and positions also matter.
Google Search Console for sc-domain:snorkelforecast.com. Previous: 1 May–29 July 2026. Current: 30 July–27 October 2026. Retrieved on 30 October through the owner's authorised RefreshAgent connection. The results are a fixed snapshot; Google can revise historical records.
This is observational website-wide evidence. AI contribution, seasonality, indexing, new content and parallel changes have not been causally isolated. Google Search clicks are not GA4 sessions, bookings or revenue. A change in average position does not mean every query improved.
The Maui, Oahu, Kauai and Big Island pages all received Google clicks in the earlier comparison window. Their net gains total 3,004 clicks, equal to 48.9% of the site's 6,138 net additional clicks.
Reproducible calculation
(1,889 + 467 + 326 + 322) ÷ 6,138 = 48.94%. The denominator is the net domain-wide click change, not the sum of only winning pages. These four selected pages are a subset, not the complete decomposition of all existing versus newly published URLs.
Maui rose from 522 to 2,411 clicks; Oahu from 41 to 508; Kauai from 2 to 328; Big Island from 18 to 340. Earlier-period clicks establish that these pages existed and had search visibility before 30 July; they do not establish which individual changes caused the gains.
The dated internal growth log and experiment register document concrete changes. A planning entry, code implementation and verified production deployment are different levels of evidence; none alone establishes a traffic effect.
The growth log records a Maui-focused title and description, clearer coast comparisons, forecast-based best/weakest beach explanations and official observation links. The original goal was to improve report-query rankings toward position 5. The log does not independently prove the precise production deployment timestamp or its traffic effect.
Evidence: current Maui report · internal SEO growth log, 3 August.
The project log describes expanding the Maui report approach to Oahu, Kauai and Big Island, including coast-level forecasts and relevant internal links. The log calls for a post-deployment review; the exact live date and page-level traffic effect have not been verified for this study.
Evidence: four current island reports · internal SEO growth log, 12 September.
The experiment register records a Hawaii-specific change intended to avoid leading evening visitors with a nighttime 'Poor' verdict. A completed 14/28-day performance review and independent production verification are not established here. No uplift is attributed to this experiment.
Evidence: internal experiment EXP-2026-09-22-hawaii-daytime-lead · outcome pending.
At the August baseline, the recorded goal was better Maui report-query positioning and clearer information for visitors—not a prediction of sixfold sitewide growth. The query-specific position-5 objective has not been verified as met.
New pages, expanding query coverage, seasonality, parallel releases and changes in Google's indexing are plausible contributors. This dated timeline is an evidence map, not an experiment isolating the causal contribution of RefreshAgent or other tools.
The published evidence is a dated snapshot, not a silently changing dashboard. The page export covers the twelve largest positive gainers returned, while the domain totals cover all URLs.
Use Google Search Console for the same property and two non-overlapping 90-day windows. Normalise the canonical page URLs, subtract prior-period clicks from current-period clicks for each of the four Hawaii reports, add those differences and divide by the domain-wide net click increase. The top-gainers CSV alone does not reproduce the whole site's total.
Not computed: the complete share of growth from all existing versus all newly published pages. That requires the complete GSC URL inventory and validated publication/deployment history. A seasonally adjusted counterfactual, attribution to an individual AI agent, conversion uplift and revenue uplift have not been established.
Check the shorter window too
The latest 30-day comparison (28 September–27 October) recorded 3,197 clicks versus 2,799 in the previous 30 days (29 August–27 September): +14.2%. This provides a more recent, less dramatic measure alongside the longer-term 90-day result, not a directly comparable claim.
Data provenance: owner-authorised GSC property retrieved through RefreshAgent on 30 October 2026; data through 27 October. The page-level CSV is a selected set of positive gainers, not the full search-performance inventory. No customer account or private authentication data is included. Snapshot corrections should produce a new version rather than silently rewriting historical evidence.
Three concise, independently quotable observations with exact dates and limitations.
+514.1% across consecutive 90-day periods for the whole website.
Four pre-existing Hawaii report URLs accounted for 3,004 of the 6,138 net additional clicks.
No controlled study isolates the effects of RefreshAgent, coding agents, indexing or seasonality.
This first-party study covers a website operated by RefreshAgent's founder, Duncan Trevithick. It is not an independent customer case study. The site develops original calculated marine-condition information rather than publishing generic AI travel articles. Please cite the specific source snapshot and methodology; researchers seeking additional context can contact the author through his public profile.
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, propose bounded changes, run checks and record work in Git. RefreshAgent provides SEO evidence and analysis; the coding agents and owner implement and review changes. A Git commit is not proof of production deployment.
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.
These are dated opportunity snapshots from 9 October, not evidence that the follow-up experiments succeeded. Newer sitewide and page-level results appear above.
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.
