Teardown No. 136 · Affiliates

4,797 AI Citations, 11 Pageviews: How Paul Watson's Codex agent rebuilds his old affiliate site and keeps the scorecard honest

Paul Watson is using Codex to bring his neglected affiliate site, Make That Sea Change, back to life, and he publishes a weekly report on what it did and what it earned. He connected Codex to WordPress, Google Search Console, Bing, GA4 and GitHub, then turned his review-refresh process into a skill that ran for two days reviewing, editing and publishing old posts. In week 9, Bing's AI Performance report recorded 4,797 citations of the site, but three of the leading reviews (Pampered Chef, Lurn and Norwex) had 11 pageviews in the 28-day window he checked, and there were no new referrals that week. His goal is 100 Starter-to-Premium upgrades on Wealthy Affiliate by December 31. By week 17 he had one new free Starter and no verified new paid upgrade.

Try this step

Connect the agent to every data source first. In week 1 he used Codex to connect his WordPress account, Google Search Console, Bing and GA, and set up GitHub repositories to save files. His first lesson: "AI can teach you AI". He asks it to walk him through the setup and to document it, so he learns what happened.

The problem

Paul built Make That Sea Change years ago to earn Wealthy Affiliate (WA) commissions, then left it. After a Google update he had set many of his old MLM company reviews to noindex, yet they still attracted the most traffic. In July 2026 he started rebuilding the site with AI. He is clear that it is not a full-time project: he and his wife run fitness businesses in the US after moving from Australia. He is on the $20 plan and promised himself he would not upgrade until the site earned the money to pay for it.

The blueprint (as he describes it)

  1. Connect the agent to every data source first. In week 1 he used Codex to connect his WordPress account, Google Search Console, Bing and GA, and set up GitHub repositories to save files. His first lesson: “AI can teach you AI”. He asks it to walk him through the setup and to document it, so he learns what happened.
  2. Audit before writing. The first job was a full SEO check. Then they updated the money pages and any calls to action that were stale or pointed to the wrong place.
  3. Start with the pages that still get traffic. The old MLM reviews came first. For each one the agent checks whether the company is still active, whether the opportunity still exists, whether the review is now misleading, and whether there are broken links, old videos, stale screenshots, outdated compensation plans or old schema.
  4. Run it by hand, then turn it into a skill. They set up the task, ran it a couple of times, and he corrected layout and text issues. Then they made a skill to run it automatically. In his words: “For the next two days non stop, Chat codex clicked away reviewing, editing, updating, reworking and publishing updated posts to my site.” Then he ran out of credits.
  5. Use the credit gaps for other work. When credits run out he works on a sellable workbook. When he has credits, Codex reviews and updates the site and builds the weekly report.
  6. Build audit tools from the process. By week 9 he had five beta tools live on the site: Affiliate Site Reset, Internal Linking Audit, Affiliate Link Checker, Content Refresh Finder and Blog Post Brief Builder. They sit behind an email opt-in and are meant to go behind a paywall later.
  7. Publish an AI-written scorecard every week. The first section of each post is his own writing. The rest is written by AI from GA4, Search Console and WA data. Each report separates new referrals from renewals and says which reporting window each figure covers.
  8. Check every traffic source before retiring a page. In week 17 he found old YouTube videos that still had viewers while the articles they pointed to had been retired, redirected or left out of date. By November 14, eleven guides and their YouTube links had been refreshed.

Stated results

  • AI citations: Bing’s AI Performance report recorded 4,797 citations across its June to September reporting period. Across three of the leading reviews (Pampered Chef, Lurn and Norwex), “GA4 recorded only 11 pageviews in the 28-day baseline we checked.” The report calls this historical visibility, not new mentions from the rebuild.
  • Week 9 scorecard (12 to 18 September): 0 new referrals, 0 new Premium upgrades, and a $46.50 commission from an existing subscriber renewal.
  • Speed: one older article went from a mobile PageSpeed score of 44 to 78 after the YouTube player was set to load only on play.
  • Week 16: the SEO audit was finished. He wrote that he was “a little deflated” by the results so far.
  • Week 17: one new free Starter joined through his review link, and WA recorded a $40 commission from an existing referral, which he keeps separate from new business. Sessions rose by 222 while engaged sessions fell by 21.
  • Goal: 100 Starter-to-Premium upgrades by December 31. No verified new paid upgrade as of week 17.

Caveats

  • The report sections are written by AI from his data. He says so in each post, and in week 2 he copied Codex’s list of work “word for word”.
  • The 4,797 Bing citations cover a period that mostly predates the rebuild, and Bing’s report does not cover other AI platforms.
  • He sells AI audit tools on the site (free beta behind an email opt-in, paywall planned) and earns WA referral commissions. The weekly posts are on WA’s own blog platform.
  • He uses Claude Code for his studio business and Codex for this site, and Lovable for a companion app. This is not a Claude-centric setup.
  • He notes that AI is literal: his first SEO audit skill did not check meta schemas or spelling until he added them.
  • Handles: Make That Sea Change on Facebook and the YouTube channel @PaulWatsonMakeThatSeaChange, both linked from his site. No X, LinkedIn, Bluesky, Threads or Instagram profile was verified.

FAQ

What does the Codex agent do on the site?

It audits pages, refreshes old reviews and guides, fixes calls to action and tracking, and publishes the updates. It also writes the reporting section of his weekly post from GA4, Search Console and WA data.

Did 4,797 AI citations bring visitors?

Not many that he could see. Three of the leading reviews had 11 pageviews in the 28-day window checked, and there were no new referrals that week.

Has the rebuild made money?

One new free Starter joined in week 17. The commissions in the reports ($46.50 and $40) came from existing referrals, and no new paid upgrade had been verified.

What would he do differently?

His week 17 lesson is to check every existing traffic source before retiring a page, because an old article may still be the next step for a YouTube viewer.

Sources


Want agents like these working your distribution? Get It Built. Get new teardowns weekly: subscribe (double opt-in).

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.