We're fixing our own search visibility in the open. Every finding, every change, and every traffic update is documented here — so you can run the same process on your business site.
Live experiment — started 22 May 2026Before any SEO work, this is what the custom analytics showed. The numbers are small enough that the problem is obvious: the site isn't being found. Not a conversion problem. A visibility problem.
Baseline data below is from the pre-upgrade period (95 days to 22 May 2026, manual analytics).
Diagnosis: traffic is essentially zero from search. The site had been live for months with almost no organic visibility. Google was either not crawling the main service pages, or wasn't ranking them at all. We needed to find out which — and fix it.
We ran a full technical SEO audit across the four main pages: homepage, Agentic Engineering, Automation, and the Websites demo. Here's what came back.
| Finding | Page(s) | Severity | Status |
|---|---|---|---|
| 3 service pages missing from sitemap.xml — Google may never crawl them | AE, Automation, Websites | Critical | Fixed |
| No H1 on flagship Agentic Engineering page | agentic-engineering.html | Critical | Fixed |
| Homepage title 63 chars, truncated in Google SERPs (limit: 60) | index.html | High | Fixed |
| Meta descriptions over 155 char limit on 3 pages (168, 174, 183 chars) | Homepage, AE, Automation | High | Fixed |
| Automation H1 ("Small, sharp automations…") contains no searchable keyword | automation.html | High | Fixed |
| Muted text contrast fails WCAG AA — rgba(241,241,241,0.4) on #000a1a = 3.3:1 (need 4.5:1) | Homepage (12 instances) | High | Fixed |
| harbour-demo.html has no canonical tag and no schema | harbour-demo.html | Medium | Fixed |
| Brand name inconsistency: "SolVX" / "SolvX" in schema and meta tags | Automation, Homepage schema | Medium | Fixed |
| Sitemap lastmod dates all showing 2026-02-15 (stale) | sitemap.xml | Medium | Fixed |
| Google Search Console not yet set up — no impression or ranking data | Whole site | High | Fixed |
Nine issues fixed in a single session. These are table-stakes changes — Google won't rank pages it can't crawl, and it won't display truncated titles well. None of this creates traffic on its own, but without it, nothing else matters.
The sitemap only listed lab and utility pages. The three main revenue pages had never been submitted to Google. Fixed by adding all service pages with current lastmod dates.
The flagship page had no H1 at all — headings jumped straight to H2. Added a visually-hidden H1 so Google reads it without disrupting the visual design.
Low-opacity text (rgba(241,241,241,0.4)) on a near-black background blended to an effective grey that only achieved 3.3:1 contrast — below the 4.5:1 WCAG AA minimum. Raised opacity to 0.6–0.65 across 12 elements.
The technical fixes were table stakes. Since then the work moved from "make the site crawlable" to "make it findable and worth ranking" — keyword-targeted titles and descriptions, then a full pass for AI search (ChatGPT, Perplexity and Google's AI answers), plus real service pages for what we actually sell. Google Search Console is now connected, so impressions and positions are tracked properly.
Organic search is still small, but it has moved off the floor and is now a real, growing channel rather than a rounding error — which is exactly the shape you expect as a new site earns authority. The live counter at the top of this page updates on every visit. The full running log below tracks each change and why we made it.
This is the same checklist we ran on solvX.uk. If you have a business website, run through these in order. Each one either passes or fails — no expertise needed.
We audit small business websites and fix the issues we find — same process as above, done for you. Free 15-min call to start.
Book a free call Back to solvXEvery change made to the site — and the reasoning behind it — logged here in chronological order.
A week ago we logged the first-ever visitors arriving from ChatGPT — 3 of them, and the open question was whether it was a fluke. It wasn't. That number has tripled to 9, and most of the growth is recent: six of the nine arrived in the last seven days, including four in a single day on the 27th.
This is the pattern we hoped for. A brand-new site is effectively invisible to Google for months while it earns authority — but the AI answer engines read the structured, well-labelled content we built and started citing it in weeks. The takeaway is unchanged but now evidenced twice over — if something is time-sensitive, AI search is the faster route to being found.
Traditional organic search is still climbing quietly in the background too — 42 all-time referrals, up from 5 at baseline in May. Both channels are now real. Neither is large yet, but a year ago both were zero.
Three weeks after building the site for AI search, ChatGPT started referring real visitors — the first chatgpt.com traffic we've ever recorded. Small in absolute terms, but it's a channel that was completely shut before the AI-search work, and it opened on its own once the service pages and structured content were in place.
The striking part is the turnaround: AI answer engines picked the pages up in weeks, where traditional Google ranking is a months-long compounding game. Worth remembering for anything time-sensitive — a launch or a limited promotion has a real shot at showing up in AI answers far faster than in classic search. Organic search is still climbing steadily in the background (all-time search referrals up from 5 to 37 since May).
The biggest change yet. We stopped optimising a site that only implied what we do and built real pages for what we actually sell: video production, game development and 3D animation, each answering — in plain machine-readable text — what it is, who it's for, how a project works, and the evidence behind it. The homepage was reframed from "practical AI for small business" to a creative-technology studio that leads with those three services. Three case studies and a buyer-intent guide were added as evidence, and robots.txt now explicitly welcomes the AI-answer crawlers (OpenAI's GPTBot & OAI-SearchBot, PerplexityBot, Google-Extended) so ChatGPT, Perplexity and Google's AI answers can read and cite us.
We also closed the conversion gap: a short enquiry form now sits on every service page, so an interested visitor can say what they want made without composing an email. Traffic to date: 438 visits / 178 unique since launch, with organic search up from 5 to 24 — small, but off the floor and climbing.
Ran real keyword research (free — web search, no paid tools) across the core services and rewrote every title and meta description to fit Google's SERP limits and target researched long-tail terms instead of generic brand copy. Several pages had titles over 60 characters and descriptions of 196–261 — all truncating in results — now trimmed to 54–58 and 143–158. As part of it, the incoming-invoice tool was repositioned around "accounts-payable automation", which is what it actually does, rather than competing with invoice-creation tools.
Started tracking the site properly and ran the first full technical audit. The core problem was visibility: three service pages had never been added to sitemap.xml, meaning Google may never have crawled them. Also found a missing H1 on the flagship Agentic Engineering page, over-limit title and meta descriptions on multiple pages, and 12 instances of text failing WCAG AA contrast.
All 9 fixable issues were resolved in the same session. Google Search Console reconnection is pending — it was previously verified but fell out of use.
Everything documented on this page is reproducible. The same checklist, the same contrast formula, the same schema templates — packaged as a free, open-source skill for Claude Code. You install it once and it runs on any website project you point it at.
Since the first release the skill has grown into two layers: the 10-step technical audit (crawlability, tags, schema, contrast, speed), and a zero-budget growth loop on top — keyword research for titles, reading your own first-party analytics, tracking the organic-vs-social traffic mix, and keeping a living progress log so every run builds on the last instead of starting over. The whole thing takes about 10 minutes to set up; after that, a full audit is one command.
Claude Code is a CLI that runs an AI agent alongside your code files. A subscription is required — a free tier exists but has limited usage. The Pro plan resets your usage every 5 hours.
The skill reads your actual files — not a URL, not a pasted snippet. If your site is in a folder on your computer or in a Git repo, you're ready. On managed hosting (Wix, Squarespace, etc.)? See the section below.
Install the CLI via npm (requires Node 18+):
Then start it and log in:
This opens a browser login flow. Sign in with your Anthropic account. Once authenticated, Claude Code is ready in any terminal session.
You'll also need the GitHub CLI — for installing the skill and (optionally) contributing improvements. Paste this into a Claude Code session to have it install it for your OS:
On Mac it uses Homebrew. On Windows, winget or the MSI installer. On Linux, apt or rpm. Claude Code detects your OS and handles it.
Skills live in a folder in your home directory and are available across every project. Open a Claude Code session and paste:
Or manually:
Open a Claude Code session in the folder containing your HTML files and type:
Or target specific pages:
Claude Code reads your files, runs all 10 checks, and outputs a findings report with severity levels and concrete fixes. To apply the fixes: "fix the critical and high issues" — it edits the files directly.
You don't have HTML files to audit — you have a locked platform. Claude Code can fix that. It mirrors your live site, rebuilds it as clean HTML/CSS, and creates a GitHub repository you own. Once that's done, you run the audit on the clean files and you're free to host anywhere.
Rebuilding a full site uses more of your 5-hour window than a normal audit — potentially half or more for a large site. It's a one-time cost. Once the repo exists, every future session costs a fraction of that. A site with 5 pages or fewer will use well under half a window.
Paste these prompts into a fresh Claude Code session, in order:
Benefits beyond the audit: you own the code, can host anywhere, can make any change directly, and have full version history in Git.
If the audit misses something on your site, finds a false positive, or you've improved a check — raise a pull request. The skill gets better with real-world use from real sites.
Every pull request must include before/after audit output from a real site showing the improvement. For contrast changes, include the computed ratios. Opinion-only PRs will be closed.
Ask Claude Code to handle the PR for you:
SKILL.md, CLAUDE.md contributor guide, README — all open source.
View on GitHubNo time or subscription? We can run the audit and apply the fixes — or rebuild your site from managed hosting into a proper repo you own.
Book a free 15-min call