How We’re Growing Our Blog with AI: The Results and the Workflow

Our blog earned 3,929 Google search impressions in July and 5,253 in August 2026, a 33.7% increase across two complete months. September 1–14 added 6,464 impressions. That partial month averaged 461.7 impressions per day, compared with 169.5 in August.

We use AI to help research, write, improve and check the work, supported by reusable instructions, focused project records and scheduled reviews. Below is the evidence, what we changed, and a practical way to build your own version.

Download the free blog growth replication kit (ZIP) — four portable skills, automation prompts, a project template, review rubric, change log and our dated search data. No email required.

Our month-by-month search results

Our own search data · July–September 2026

More visibility. The clicks still need work.

Same blog cohort, compared per day. September covers 14 days; July and August each cover 31.

Impressions / dayClicks / day

Average daily impressions · scale 0–500

July126.7

August169.5

September 1–14partial461.7

July → August: +33.7% impressions over two full months. September’s observed daily rate is 2.72× August’s, not a full-month result or forecast.

Underlying totals — Google Web search, main-domain blog pages

Period	Days	Impressions	Clicks	CTR
July	31	3,929	8	0.20%
August	31	5,253	12	0.23%
September 1–14	14	6,464	10	0.15%

Source: Search Console daily table, read September 16, 2026. Web search; pages containing aibusinessboomer.io/blog/. Dates use Pacific Time. Filtered data can be partial. These are search appearances and clicks, not ChatGPT referrals, leads or revenue.

Download the daily data (CSV)

An impression means our page appeared in search; a click means someone followed a result to the site. Our monthly blog clicks were 8 in July, 12 in August and 10 during September 1–14. The click counts are small. Increased visibility is a useful early signal, but it is not evidence of a comparable increase in customers or revenue.

For an equal-length comparison, July 21–August 17 versus August 18–September 14 produced 4,209 → 9,106 blog impressions, 14 → 15 clicks, and a reported average position of 34.6 → 18.2. Lower position numbers are better. Average position also changes with the mix of queries; this does not mean every keyword moved up by the same amount.

The chart uses the daily table from Google Search Console, read September 16, with Web search and a page filter containing https://aibusinessboomer.io/blog/. Daily sums supply the exact impression and click totals; the interface rounds its summary cards. September ends on the latest complete date available in our review, September 14. Filtered reports can be partial. This is Google visibility, including eligible Google AI search appearances, not a ChatGPT traffic report. Google’s reporting explanation.

Which pages are driving the increase?

In the same two 28-day periods, a small number of existing pages supplied much of the growth:

Existing article	Earlier impressions	Later impressions	Later clicks
Video-editing tools	293	2,514	1
AI blog-writing tools	67	1,005	0
AI automation partners	198	443	1
AI website builders	283	476	1

The first two pages added 3,159 impressions between them. Our working interpretation is that specific comparison and workflow questions are earning visibility. The evidence supports improving those useful pages; it does not establish that every new article or every AI-generated paragraph will perform.

The next challenge is clear: give the right reader a compelling reason to click and a useful next step once they arrive. Our guide to impressions without clicks explains how to diagnose that gap.

What we changed—and when

The dates matter. Some of our current process improvements happened after the rise began.

When	Change	What the evidence supports
Before September 8	Existing comparison guides were already earning more visibility.	The increase was underway before the refresh package below.
September 8	We refreshed three priority guides, corrected source-backed product details, added three worksheets and improved discovery links. A separate Search Console guide also went live.	These were published improvements to already-relevant content. They cannot explain the beginning of the rise.
September 8 onward	Daily draft production continued even when publication was held.	A blocked release no longer prevented a complete new draft. A saved draft still was not counted as a published page.
September 9 operating revision	One focused task became responsible for each outcome; concise project records and source links replaced mandatory coordinator relays for routine work.	This describes our working structure. We have not measured its independent effect on rankings or productivity.
September 15	The daily publisher received standing authority to publish one current-day article after all required checks pass.	This began after the chart’s September 14 cutoff, so it did not cause the growth shown here.
September 16	We expanded the video guide with our Codex/HyperFrames process and new editorial images.	The update is live. Its search effect has not yet been measured.

We cannot separate our work from changes in search demand, competitors, Google’s systems or query mix. This is an observed improvement alongside a documented process, not a controlled claim that AI caused a particular uplift.

The structure: one owner, reusable skills, visible evidence

Our system keeps four kinds of information separate:

Record	What belongs there	Why it helps
Project memory	Current facts, decisions and one next action	The assistant can resume without guessing what is current.
Work scope	The requested outcome, boundaries and completion check	The task stays focused on a deliverable.
Sources	Links to original research, analytics and release evidence	Claims can be checked instead of repeated from memory.
Tools	Available capabilities and recent access checks	A configured connection is not assumed to work.

One focused task owns the outcome. It loads the relevant context, selects the procedure it needs, completes the work and records evidence. Additional agents are optional when an independent subtask warrants them; a large team is not a prerequisite.

A skill is a reusable set of instructions for doing a job. An automation runs a defined job on a schedule. Neither replaces the source data, grants publication permission or proves that a useful result was delivered.

The skills readers can reuse

Our public skills catalog includes AIBB’s writer, content optimizer, auditor and SEO quality-review procedures. Some catalog downloads retain older AIBB-specific roles and dependencies. Read and adapt them before use.

For this case study, we also made a portable public starter kit from the current working method. Its names and packaging are new; we are not claiming these exact files produced the historical results.

Skill in the starter kit	Job	Concrete output
blog-research-brief	Check reader intent, existing coverage and current primary sources	One evidence-backed brief with a refresh-or-new-page decision
blog-draft	Turn that brief into a complete useful article	Draft, metadata, internal links, sources and open checks
blog-refresh	Improve a page already earning relevant impressions	Revised page plus a dated before/after change log
blog-release-review	Review claims, duplication, build and rendered output	A verified release receipt or a specific hold

Download all four skills and supporting templates. In a Codex project, copy each skill folder into .agents/skills/, read its instructions, and replace the project template with your own domain, offer, audience and release rules. Other assistants can use the Markdown as procedural instructions. Analytics access, publishing tools and any paid services must be configured separately.

The automations we currently run

These are the saved schedules as of September 16, in America/New_York time. The replication kit contains editable prompts, not an installer that silently enables them.

Schedule	Job	Completion check
Daily, 9 AM	Research and produce one complete current-day article	Dated draft and index entry exist, with evidence and any unresolved checks
Daily, 2 PM	Improve that candidate and publish only if eligible	Editorial threshold and every required check pass; public content and canonical URL are verified
Monday, 10 AM	Read-only website and SEO health review	Tested routes, source-period metrics, defects and prioritized actions are saved
Friday, 1 PM	Weekly growth scorecard	Equal-period comparison, data limitations and one useful next action are delivered

Our publication threshold is 85/100 plus passing factual, duplicate, source, build, rendered-preview and destination checks. The score is an internal editorial checklist, not a Google ranking metric. A failed check blocks release regardless of score. The ZIP includes a clearly labeled starter rubric you can adapt.

The September 15 scheduled afternoon run produced a live article about testing AI receptionist hours, with a saved release receipt and public readback. That is evidence of an executed publishing cycle, not proof that every future scheduled run will succeed.

Daily publishing is our current operating choice, not a proven ranking requirement. If your business cannot produce a distinct, useful article every day, use a slower cadence. Never invent a topic merely to fill a slot or publish an old draft as today’s work.

We have also had analytics authentication failures. A successful schedule configuration does not prove the underlying data was fetched. When access fails, use an explicitly dated relevant export or mark the metric unavailable; do not turn missing values into zero. The growth chart here was verified directly in the signed-in Search Console report.

Replicate the process with one page first

1. Choose one audience and one offer. Write down the practical question a potential customer needs answered. Inventory existing pages so you do not create a near-duplicate.
2. Save a baseline. Export page and query data for a complete period. Record cohort, dates, impressions, clicks and relevant positions. Keep leads and revenue in separate verified fields.
3. Select a useful existing page. Look for relevant impressions and a clear gap: an incomplete answer, stale comparison, missing example or weak next step.
4. Make a material improvement. Add checked information and something the reader can use: a worked example, decision table, checklist or worksheet. Preserve a useful canonical URL.
5. Review the real page. Inspect desktop and mobile, links, images, metadata and sources. Publish within your authority, then read the public result back.
6. Record the change and measure again. Check delivery immediately; compare the same cohort after a complete post-change period. A ranking movement is an observation until you can justify a causal claim.
7. Schedule only after a successful manual cycle. Start with the draft and weekly review prompts. Grant a publisher narrowly defined authority only when your checks and recovery path are reliable.

Use the kit’s blank scorecard and change log for your own results. Our CSV is included so you can inspect this case study’s arithmetic, not copy our performance into your business’s report.

Google allows useful applications of generative AI, but warns against producing many pages without added value. Our aim is to make the answer, evidence and delivery process better—not to treat volume as the outcome. Google’s guidance on generative AI content.

What we are improving next

The growth in impressions gives us better information about which questions Google is surfacing our pages for. We will use it to improve query relevance, search presentation, original examples and the path from useful advice to a relevant service conversation. We will measure qualified inquiries separately before claiming business growth.

Get the replication kit and start with one page. If you want help adapting the process to your business, book a consultation and bring your site, one customer question and your current search data.

Run your own measurable experiment

Choose one page, save its current search data, and write down what you expect the change to improve. Record the release date before reviewing the results. Compare equal complete periods and show clicks alongside impressions and position.

Download the SEO experiment worksheet and six-step measurement guide. The worksheet is blank apart from a labeled example; it does not promise a result. Keep reader clicks, downloads, inquiries, and sales separate.