Blog vs no blog: the rules of my six-week experiment

How this all started. My friend and I tried to build a couple of products in the last year or so. We are very new to SEO and to marketing, and we tried different things. One of the things I tried was to write a blog.

I know it is a slow strategy, and it has changed a little with the arrival of AI. Now everyone tries to rank not in Google but in AI chats, which makes sense. However, it is not a real science, because you don’t have measurable results. Both of us are engineers, and we are used to a clear input and a clear result that can be tested. Writing a blog post and hoping it ranks is nothing like that. Our product does show up in ChatGPT and in Google AI results, so I think the blog works. But I can’t be sure, because a lot of other things happened at the same time. The blog is only one of them.

I also built a tool for myself to write blog posts faster, in my voice, promoting my product, with illustrations in my style. I think it writes better than most of the tools on the market, if not all of them. But better writing is not the same as results, and I can’t prove the results yet. That is why I’m running this challenge.

The setup

I can put the reason for this experiment in one sentence: the only way to say that my tool, and blogging in general, helps is to run the experiment. Two products, the same product with different names, and the only difference is the blog.

I built two products this month: Gripelens and Quibblemap. Underneath they are the same product: review mining for founders. You type a product or a category and get its competitors and their recurring complaints from public reviews, with a quote behind every claim.

The Gripelens and Quibblemap home pages side by side on launch day: the same product, the same five pages, two names and two looks.

I gave both sites the same price, the same five pages, the same waitlist popup and the same stack, Astro on Cloudflare Pages. I launched both on the same day, Monday 19 October 2026, with zero backlinks each. The sign-up button on each site opens a waitlist that gives access in waves, and it asks for an email only. I wrote the copy for each site separately from one brief, so the name, the domain and the look are different.

Which one got the blog was not my preference. On 7 October 2026 I ran Python’s secrets.choice over the two names, and Gripelens came out.

The one difference

I give Gripelens a blog with four posts a week. I research, write and publish them with my personal Claude Code plugin, the one I use for my own blogs and plan to release when it is ready. I approve each post before it goes out.

Quibblemap gets nothing. I froze its pages on launch day and they stay frozen.

The rules

I wrote the rules down before launch so nobody can say later that I rigged the test.

Everything is identical except the blog. The static pages on Gripelens stay frozen like the ones on Quibblemap; only blog posts get added.
I will not buy, ask for or build backlinks. No ads. No directory submissions unless both sites get the same ones on the same day.
Neither site links to the other, to this post, to the series page or to the plugin.
Every claim in a Gripelens post comes from a scrape I run for that post. No user counts, no customer claims.
I name both sites, or neither, in every post, tweet and video.
The test runs six weeks, from Monday 19 October to Sunday 29 November 2026. Then I decide whether to run it to 90 days.

What I measure

I measure both sites every Monday, the same day for both, starting with run 0 on launch morning. I will explain three terms before the table. Search Console is Google’s own report of how a site shows up in its results. AI crawlers are the bots that fetch pages for ChatGPT, Perplexity, Claude and Google’s AI answers, the ones Google calls AI Overviews and AI Mode. The prompt panel is my own set of buyer questions that I put to the AI assistants and score.

What	Where it comes from
Visits from AI crawlers (GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended)	Cloudflare AI Crawl Control
Pages indexed by Google	Search Console, plus a `site:` search
Impressions (how often a page appeared in Google results), clicks, queries and positions	Search Console
Impressions and clicks from AI Overviews and AI Mode	Search Console, the Generative AI report under Performance, exported by hand because it is not in the API
Visitors referred from chatgpt.com, perplexity.ai and gemini	Umami, the site analytics
Whether ChatGPT, Perplexity and Google AI Mode name or cite the site, for the questions I pick	My prompt panel: 11 questions a buyer would ask, three turns each, run twice a week through the OpenAI and Perplexity APIs and once a week by hand in the chat apps
Waitlist signups, and which page and link they came from	The sites’ own form, with a UTM tag on every link that records where the click came from

I picked Cloudflare because it shows every crawler, the AI ones included, it costs nothing, and both sites get the same view.

How the prompt panel works: a buyer's question, three engines, and a scorecard that starts at zero for both.

What I expect

I’m writing this down before the data so I can’t move the goalposts later.

In six weeks I expect both sites to be indexed. For Gripelens I expect impressions in Search Console, the first queries that are not its name, and maybe a ranking on page two. I expect AI assistants to describe Gripelens when asked about it by name and, with luck, to name it for a category question. For Quibblemap I expect to see it indexed, brand queries, and less from AI than Gripelens gets. I don’t expect a lot on one side and nothing on the other, just a difference. Stable rankings come later than six weeks, so I don’t expect those yet.

If the blog did nothing, the two Search Console charts and the two crawler counts will look the same after six weeks. And if I don’t see results, that is a result on its own.

Day 0

I recorded this table on launch morning, and it is the baseline. Both domains are brand new.

	Gripelens	Quibblemap
Pages indexed	0	0
Impressions	0	0
AI crawler visits (GPTBot, the week before launch)	1	1
Prompt panel mentions	0	0
Waitlist signups	0	0

The crawler row is not quite zero. In the week before launch the same bots knocked on both doors: Googlebot 14 times, Applebot 10 times, GPTBot once each, and Bingbot twice on Gripelens. Same bots, same counts, before either site had a single link pointing at it. Both sites sat behind a login page until launch morning, so every one of those visits saw the login page and nothing else.

Cloudflare AI Crawl Control for Gripelens and Quibblemap, the seven days before launch: 27 and 25 crawler requests, the same bots on both.

Follow along

I list every post and video on the series page as they come. I publish a post every Friday. I also post on X and YouTube, and the blog has an RSS feed.