Hustle #006 / AI Site Check

AI Site Check: eight steps, one build.

A planned plain-English website audit for small businesses. The site and sample report are public. Checkout and live scanning are not available.

What we did

01 / Choose one customer

2 October

Small businesses need fixes, not another score.

  • The founder chose a website-check service. The offer is for small businesses that need to know what to fix and how.
  • The founder pushed for plain English. The sample should explain the real changes, not leave the customer with an unexplained score.
02 / Check the idea and the costs

2-3 October

Check what the audit can honestly promise.

  • The agent researched primary search guidance. It checked Google Search Central and OpenAI crawler documentation before describing SEO and AI-search access.
  • The founder challenged the planned £99 report. She said the sample needed to explain the real things that drive GEO.
  • The agent kept price and scope separate. £99 is a planned one-off price; the page allowance and delivery time are not final.
03 / Give it a name and a home

2-3 October

Name it and make the public home.

  • The founder chose AI Site Check. The agent built the site on Cloudflare Pages and connected aisitecheck.co.uk.
  • The agent set up the Google Search Console domain property. It added verification and submitted the XML sitemap.
  • The agent checked Google URL inspection. All five canonical pages were reported indexed on 3 October; the sitemap was successfully fetched with five discovered pages.
04 / Make one thing people can buy

3 October

Show the report before selling it.

  • The agent built a nine-area sample report. Each finding shows illustrative evidence, priority, exact fixes and a verification step.
  • The agent labelled the sample as fictional. Example Bicycle Repairs is not a scanned customer site.
  • The agent left checkout and scanning off. The secure scanner and payment route are not ready, so the audit stays planned.
05 / Put it in front of real people

3 October

Publish useful explainers alongside the sample.

  • The agent published two blog explainers. One separates ChatGPT search from training; the other explains the website basics behind Google AI search.
  • The agent added crawlable discovery files. The site has robots.txt, a five-page sitemap, canonical URLs and an informational llms.txt.
  • The agent linked the public build. The site is available at aisitecheck.co.uk without a payment or scanning claim.
06 / Learn from the first response

3 October

Make the sample useful after real feedback.

  • The founder rejected the first sample. She wanted more GEO depth and practical fixes for the planned price.
  • The agent expanded the findings. The revised sample covers access, entity clarity, answer-first content, JSON-LD, structure, evidence, FAQ limits, llms.txt limits and retesting.
  • The founder approved the deeper sample. Her response was: "Sample report much better".
07 / Hand over one repeatable job

3 October

Keep a repeatable watch, not a launch claim.

  • The founder requested a weekly search blog. It should explain what is changing and how LLMs search, using original sources.
  • The agent set an hourly sitemap-status watch. Google fetched the sitemap successfully on 3 October. The agent checked all five pages and retired the completed watch.
  • The agent separated access from ranking. An indexed homepage is evidence of indexing, not a promise of rankings or AI citations.
08 / Improve it, change it or stop

3 October

Simplify the route before adding features.

  • The founder cut the navigation to three links. Home, How it works and Blog. The agent removed the complicated menu.
  • The agent checked desktop and mobile. The three-link navigation is readable at both sizes; no looping carousel or marquee is present.
  • The agent kept the launch boundary clear. Payment, secure scans and a final delivery scope are the next checks before taking orders.
  • The agent fixed timeout and cleanup handling in local tests. A service error had been labelled as website ineligibility. Local code now separates the two and refuses payment before a completed report; no complete hosted report has passed yet.
  • The agent checked the free browser service before another run. The service showed no active browser sessions, but the safe allowance and stopping behaviour were not sufficiently established for the planned test. No complete report was produced.
  • The founder challenged repeated test-preparation loops. She asked the agent to finish the result rather than keep reporting blockers. The next useful proof is a complete report and download, not another isolated check.
6 November 2026

Today's update

AI Site Check is not taking customer payments. Timeout and cleanup fixes have passed local tests, including keeping a service problem separate from a verdict on the customer's site. The free browser service still needs a safe, tested execution route. A complete report and download have not passed end to end.

Repeated setup checks and requests for approval did not deliver the product. The founder called that out. The work now has to earn its next update with a tested report, not another list of technical obstacles.

The build lesson

Biggest learning

A local test can prove what code does under a simulated failure. It cannot prove that the live browser service will run and stop safely, or that a buyer will receive a useful report. Technical trouble belongs with the builder, not as a daily homework list for the founder.

The next test

Next step

Complete the report and download in a private test with a safe browser lifecycle, then test payment and delivery together. No customer orders or live payments until that works.