Twenty-two thousand checks on every page.
Informed decisions, in the team’s hands.

Hey, I've worked in SEO for more than 20 years, and for the last 15 I've kept a record of everything that proved to matter.

SEO used to be simpler. Today there is far more to check: more pages, more signals, and AI assistants that answer in place of the results page.

I used to work this way: crawl, export, analyze about 300 elements, prioritize, write the tickets by hand, then follow up with the developers. That's how most of the field works, and a growing part of it now adds AI.

Good solutions are emerging. I started building mine 2 years ago, and I went through everything that can go wrong with AI: misunderstandings, hallucinations, answers that sound smart and aren't correct. I worked on exactly that: every answer gets cross-referenced against its sources, and several AI agents check each other. It has run in production at enterprise level for 1 year, and I improve it almost every day. This is not vibe coding. Every part of it was built on purpose, over thousands of hours, and proven in production.

By hand, that work is segmented and takes weeks. My system checks more than 22,000 elements on every page, across the whole site at once, and I get it done in a couple of days.

Everything I've kept over the years goes into one record: Google's patents read in full, every update it announces and what its own people say, the case studies and the best practices. More than 22,000 elements, and counting.

Every finding answers to the record, and 28 sources that disagree get compared, element by element. Take a 5,000 page site, and hold every page against those 22,000 elements. 110,000,000 checks. Yes, you heard that right. And every one of those checks is verified a second time: when 2 sources disagree, it comes back to me to review by hand. And it's scalable: the same checks run on a few pages, or on millions of URLs.

Most of that does not apply to the website in front of me. The system keeps what matters for the website: no noise, straight to the source, page by page. Knowing what applies, and what goes first, is the work. From there, it is a chain. The reviews, with the source behind each finding. The strategy, from what the reviews show and what the business is after. The priorities, worked out with the team. Then the tickets and the briefs, to the developers and the writers, and a dashboard for each role.

This system isn't here to replace SEOs. It makes the work more efficient, gives one uniform view from every tool, and the work stays hands-on.

The dashboard has 19 sections, from search performance and technical health to content, AI visibility and competitors.

Then I stay with the work. Jira, Asana, ClickUp, Monday, Trello, Linear, Notion: any tracker reachable through MCP can be connected, so I know where each ticket stands without interrupting the team.

In the end, it recommends, and a human decides. And I'm still building. What comes next could revolutionize part of the industry. Stay tuned. In the meantime, the 2 other videos go deeper, on the technical side and on content. Thank you for watching.

Hey, this is video 2 of 3, the technical side: technical SEO, GEO and AEO.

Every URL gets crawled 3 ways, and when the crawlers disagree it gets fetched again, so nothing ships on a single source.

I compare what the server sends with what the browser builds, because anything that only exists after scripts run can go missing for a crawler.

Each page gets checked against a fixed list of named checks, in 21 groups, and every finding answers to a record of more than 22,000 elements.

That record holds granted patents read in full, sworn testimony on NavBoost, and every update Google announces, compared with the website's own traffic.

That record keeps growing, every announcement, core update and spam update goes in as it lands.

For GEO, I first check that the AI crawlers can reach the page at all, then I measure the website's share of the answer on a fixed set of prompts, with a control. For AEO, I check whether the answer about the website is correct, because being described wrongly is worse than absence. And I say what cannot be answered: no one can tie an AI mention without a click to a sale today.

28 sources get compared, element by element. Take a 5,000 page site, and hold every page against those 22,000 elements. That's 110,000,000 checks. Every one of them is verified a second time, and when 2 sources disagree, it comes back to me to review by hand. And it's scalable: the same checks run on a few pages, or on millions of URLs.

What reaches the developers is a ticket with more than 40 fields filled in: the business objective first, the expected result, and how we confirm the fix.

And I stay hands-on: I work closely with the dev team through the release, and I follow up on every ticket. Jira, Asana, ClickUp, Monday, Trello, Linear, Notion: any tracker reachable through MCP can be connected, so I know where each ticket stands without interrupting the team. After release I check the fix on the live site, and a ticket only closes once the result is measured.

2 examples from real work: one pass that read 14,935 pages across 28 sites, and a full year in production on a site with more than 500,000 indexed URLs.

In the end, it recommends, and a human decides. That was the technical side. The overview and the content side are in the 2 other videos. Thank you for watching.

Hey, this is video 3 of 3, the content side: what to write first, and how I work with the writers.

Content starts with reading everything: the full text of every page, on the website and on the competitors that actually rank.

Every page goes on one board with its type, its word count, its markup and what it still needs. It's scalable: the same board holds a few pages or millions of URLs.

The gap gets measured at the level of the claim: what the pages that rank say, and the website doesn't say yet.

Then everything goes into one ranked queue, one score fed by traffic decay and index status, re-scored every week.

A new idea only moves up if it beats what's already on the list.

The writers see only what they need. The view is completely customizable: I adapt it to their workflow and their methodology, so nobody gets overwhelmed.

Each brief is built from what the ranking pages really contain, measured with the same parser I use on the website's own pages, and I read it before a writer does.

Thin pages get enriched, I add what's missing before anything else.

Instead of analyzing what I think people want, I go directly to what they are sharing, on forums and on various websites. I built my own tool to extract those conversations from Reddit: in one study, 6,126 posts and 8,633 comments, and it turned up a product finding.

For the AI assistants, each page gets a direct answer up top, tables and lists, a dated update and an FAQ.

The content checks keep up with every core update and spam update, exactly like the technical checks.

And I stay hands-on with the writers, the same way I do with the dev team: every page has its status on the board, so nobody has to be chased.

Every change is logged with its date. Then I wait 8 weeks before I read the result, the time a change needs to show up in search, so nothing gets judged too early.

In the end, it recommends, and a human decides. That was the content side. The overview and the technical side are in the 2 other videos. Thank you for watching.