Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)

Claude and Codex Took a Blog's CTR from 0.4% to 0.9% in 3 Months and Delivered 1 Million Impressions

The Hypescribe team built an SEO pipeline on agents: Claude pulls queries from Search Console and writes articles aimed at positions 8–20, while Codex checks them for "AI slop." In three months the site received 1 million impressions in Google and 9,000 clicks, and CTR rose from 0.4% to 0.9%. For competitors, the agent describes only what it finds in quotes from their sites; otherwise it invents features that don't exist. The takeaway: automate the entire cycle, not just the writing.

AI-processed from TG: @tips_ai (Tips AI / IT & AI); edited by Hamidun News
Claude and Codex Took a Blog's CTR from 0.4% to 0.9% in 3 Months and Delivered 1 Million Impressions
Source: TG: @tips_ai (Tips AI / IT & AI). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @tips_ai (Tips AI / IT & AI) ↗
Editor-in-Chief
Key metric #1Target queries ranking in positions 8–20, the easiest ones to push into the top
Key metric #2If clicks drop 30%+ after measurement, the edit is rolled back
Key metric #3The agent compiled a list of 58 link directories and wrote submission requests for each

A familiar story: the site is in Google, but almost nobody clicks through. Back in the spring, Hypescribe got 3–4 visits for every 1,000 people who saw the site in search results. The pipeline's author, who describes himself as a SEO newcomer, put together a combination of Claude and Codex, and over the last three months the site received 1 million impressions and 9,000 clicks. CTR rose from 0.4% to 0.9%, and clicks were nearly double the previous three-month period. The numbers are self-reported and have not been independently verified, but the process itself is described in enough detail that it can be replicated.

How the agent chooses what to write about

First, Claude pulls queries for the past 1–3 months through the Google Search Console API and an MCP server, then splits them into two groups. The first covers topics where the product itself is the answer: these are what drive sign-ups. The second covers everything else: there are clicks, but almost no sign-ups. The second group is put on hold.

From the first group, the team takes articles stuck on the second page of results, in positions 8 through 20. Pushing them into the top is easier than starting from zero. The agent then checks who ranks above and how their texts are structured. A separate subagent collects facts about competitors from their official websites and saves quotes, links, and dates. This guards against a typical LLM problem: without sources, Claude is quite likely to attribute features to competitors that they do not have.

If the site is new and has no impressions, Search Console will show almost nothing. In that case, the author advises connecting SE Ranking or a similar service so the agent can see what people are actually searching for.

How the article is written and checked

The author gives a major role to the title and description, since those are what people see in search results. The title repeats the query, and the description opens immediately with a TLDR-style answer: who it suits, how much it costs, and what the catch is. The text openly acknowledges where a competitor is better. According to the author, such articles convert better, and AI search engines cite them more often. This is his observation; he laid out the supporting data in a separate post.

The text is broken up with images, tables, and lists so the reader isn't faced with a wall of text. The agent makes the images itself:

  • illustrations and assets are generated via image_gen in Codex CLI;
  • anything containing text or an interface is laid out in HTML, after which the agent takes a screenshot.

Quality control comes next. Codex rates the article on a 1-to-10 "AI-slop" scale and cycles edits until the score drops to 3 or lower. The list of banned patterns is taken from the humanizer skill: no "not just X, but Y," no em dashes, and similar tics. Then Codex plays an ordinary reader. It browses screenshots of the page on desktop and mobile and answers whether it found what it was looking for and whether it would try the product.

After a human approves the article, Claude publishes it via API, submits it to Bing, and requests Google indexing through browser-use. Measurements are taken one week and three weeks later. If clicks fall 30% or more and the position doesn't recover, the edit is rolled back. The agent has access to Search Console, the Bing Webmaster API, Playwright with Chromium, Codex CLI, SE Ranking, and PostHog.

What else is part of the SEO strategy

The blog is only one part of the system. The team builds standalone pages for specific tasks where a user can upload a file right away instead of reading an article. There are also small tools linked from the footer and from articles. And there is link building: arrangements with neighboring services to mention each other in roundups, and listings in directories such as There's An AI For That, which still bring in sign-ups. Claude also compiled a list of 58 sites and wrote a submission request for each.

What to take from this

The value of the case lies in the architecture, not in prompt magic. The agent works with real data (Search Console rather than guesses about demand), competitor facts are stored with their sources, the text passes through an independent reviewer, and every publication has a measurement point and a rollback condition. A human stays at a single checkpoint: approval.

The limitations are also clear. Growth from a low base of 0.4% CTR is less impressive than it sounds, and the author does not prove a causal link between specific steps and the results: seasonality and search algorithm updates could have played a role. Giving the agent access to all these resources requires careful permissions and logging. For a product manager or founder, a practical version looks like this: start with positions 8–20, make the agent save competitor quotes, set a rollback threshold, and compare results week by week.

Story

Key Questions Answered

How the agent chooses what to write about?

First, Claude pulls queries for the past 1–3 months through the Google Search Console API and an MCP server, then splits them into two groups. The first covers topics where the product itself is the answer: these are what drive sign-ups. The second covers everything else: there are clicks, but almost no sign-ups. The second group is put on hold. From the first group, the team takes articles stuck on the second page of results, in positions 8 through 20. Pushing them into the top is easier than starting from zero.

How the article is written and checked?

The author gives a major role to the title and description, since those are what people see in search results. The title repeats the query, and the description opens immediately with a TLDR-style answer: who it suits, how much it costs, and what the catch is. The text openly acknowledges where a competitor is better. According to the author, such articles convert better, and AI search engines cite them more often. This is his observation; he laid out the supporting data in a separate post.

What else is part of the SEO strategy?

The blog is only one part of the system. The team builds standalone pages for specific tasks where a user can upload a file right away instead of reading an article. There are also small tools linked from the footer and from articles. And there is link building: arrangements with neighboring services to mention each other in roundups, and listings in directories such as There's An AI For That, which still bring in sign-ups. Claude also compiled a list of 58 sites and wrote a submission request for each.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…