18 to 163 AI citations in two weeks. Here's exactly what I changed
The breakthrough I got, the one page that drove most of the increase, and the free playbook is now live if you want it
Cristian Lascu
June 04, 2026
Three weeks ago I wrote that AI-grounded clicks were converting roughly ten times better than traditional search clicks, and that earning AI citations was the highest-leverage move available right now. I believed it then based on a small signal: 18 citations on Microsoft Copilot, which felt promising but easy to dismiss as noise.
Then I pulled the dashboard again this week. The number was 163.
A nine-fold jump in about two weeks, with no ad spend, no backlink campaign, and no new traffic source. Just a handful of structural changes to how my pages are built. This issue is the honest breakdown of what happened, because the pattern is replicable and most of it is counterintuitive.
What the data actually showed
The first thing I did was check whether the jump was spread across the site or concentrated. It was concentrated, dramatically so.
One page drove most of it. A single page in the "best newsletters for [audience]" format pulled 91 of the 163 citations on its own. And underneath it, one grounding question did most of the work: people asking Copilot some version of "best AI newsletters for engineers in 2026," which the assistant answered by citing my page 43 times.
The rest of the citations spread across the same shape of page: best-of and side-project lists aimed at specific roles. Data engineers, staff engineers, DevOps, full-stack. Same format, different audience, each one getting grounded when someone asked a recommendation question that matched.
That concentration told me something important. This wasn't luck or a fluke of one viral page. It was a format working exactly as the mechanism predicts, which means it can be built on purpose.
Why this format wins
Here is the counterintuitive part. The pages getting cited are not my most sophisticated essays or my deepest technical pieces. They are structured lists.
The reason is mechanical. When you ask an AI assistant for a recommendation, it needs a source it can lift, attribute, and stand behind. A page that is already a curated, structured list answering exactly that question is the easiest thing in the world for the assistant to ground on. It does not have to interpret a flowing argument and extract a recommendation; the recommendation is right there, structured, with the reasoning attached.
So the shape of the question determines the shape of the page that wins. Recommendation questions ("best X for Y") get answered from list pages. Definitional questions ("what is X") get answered from pages that state the definition plainly in the first two sentences. The pages that get cited are the ones whose structure matches the question's structure.
Three things made the difference, and I want to be specific because the specificity is the point:
The pages state the answer immediately, in the first hundred words, before any preamble. Assistants lift the opening passage, so the opening passage has to contain the actual answer.
The pages are built as parseable structure. Clear headings that map to sub-questions, real lists where the content is a list, a comparison table where options are being weighed. Clean structure is what lets a machine extract a citable passage.
The site is actually reachable by AI crawlers too. This sounds obvious and it is the single most common hidden failure. Many sites block the exact crawlers they want to attract, often at the CDN edge where a default setting overrides whatever the robots file says. If your content is invisible to assistants despite being live, this is usually why.
The one caution I'd give anyone copying this
There is a tension worth naming, because ignoring it will burn you.
The same listicle format that wins AI citations is also the format that traditional search algorithms scrutinize most heavily. Thin, mass-produced "best of" pages, the same skeleton with the audience swapped, are exactly what recent Google updates (May 2026 Core Update) have pushed down hard. So you cannot just spin up forty identical lists and expect to win both surfaces. You will win citations briefly and then watch your search rankings fall.
The resolution is depth per page. Each list has to carry real assessment: why this entry made the list, who it is for, what its weakness is. A list of names with one-line descriptions is thin. A list with genuine evaluation and a point of view is a product. The format is not the problem; a shallow execution of the format is. Build each one as though it is the only page you are making, and it wins both the citation and the ranking.
What this means going forward
I'm reallocating real effort here. The math is hard to argue with: a citation from an AI answer reaches someone who is already most of the way to trusting the recommendation, because the assistant they trust just handed them my name. That converts better than almost anything coming off a traditional results page.
So the plan for the coming weeks is to take the format that's proven and extend it carefully to the audiences where I have something genuine to say, while keeping each page deep enough to survive on the search side too.
I'll report back on whether the 163 holds and climbs or settles.
I wrote the whole method down
Enough people asked how I did this that I turned it into a proper playbook, and it's live now. It covers the full process: finding the questions worth answering, building pages shaped like answers, leading with the listicle format the right way, making your site machine-readable including the CDN-edge trap, and measuring citations so you can see what's landing.
It's a free plain-markdown file, no signup beyond the newsletter you're already on. Same idea as the others: paste it into Claude, ChatGPT, or Gemini and have the assistant walk you through applying each phase to your own site.
If you run a content site, documentation, or even just a personal site where you want to be the source assistants cite, this is the one to grab.
The One Thing
The pages getting cited by AI are not the most sophisticated ones. They are the ones whose structure matches the structure of the question being asked. A recommendation question gets answered from a well-built list; a definitional question gets answered from a page that states the answer in its first two sentences.
Match the shape of your page to the shape of the question, give it genuine depth, and make sure the crawlers can reach it. That is most of the game.
What I'm Thinking About This Week
Whether the citation concentration is a starting point or a ceiling. One page driving 91 of 163 citations is a strong signal, but it raises the obvious question: is that page special, or is it just first? I'm going to find out by building two or three more in the same format, with the same depth, for adjacent audiences, and watching whether they pull citations at a similar rate. If they do, the formula generalizes. If they don't, that one page has something the others won't easily replicate, and that is worth understanding too.
Reply to this email with one of:
Your current AI citation count, if you've ever checked it (most people never have, and the webmaster console that reports it is free)
A recommendation question in your space that you think you should be the cited answer for, but aren't yet
Whether you've ever found AI crawlers blocked on your own site, and how you noticed
What resonated? What did I get wrong? Hit reply: I read everything and I'm building this with you and with your input.
P.S. The toolkit, in the order I'd use it: Visibility Audit → SEO Strategy Playbook → How to Get Cited by AI → Title & Meta Audit → Sovereign Idea Workflow → Business Model Canvas. All free.
P.P.S. Know someone who's losing traffic to AI summaries and feeling helpless about it? This is the issue that reframes it. Forward this email; they can subscribe at thesovereigntechnologist.com.
That’s all for this week.
See you next Thursday.
