Rasa SEO Strategist

14/07/2026

It's official: Google can detect AI-generated content at scale 🧨 🧨 🧨

And they just published the research explaining how. Huge kudos to Chris Long for surfacing this - it's a big one for SEO.

Google’s system uses a specialized two-stage architecture:

🔍 Classifier ΨA
​(The Bot-Net Hunter): Instead of looking at a single post, it analyzes infrastructure signals (IPs, device IDs, and inorganic behavior) to identify "Generation Clusters" - coordinated groups of accounts likely using the same automation script.

🕸️ Classifier ΨC
​(The Synthetic Rater): This uses "AI to catch AI." It leverages LLMs (like Gemini 2.0 Flash) specialized with LoRA (Low-Rank Adaptation) and S-BERT to identify "Generative Artifacts" and repetitive, templated narratives at scale.

Google’s own paper calls it a "highly accurate defense" with a

22/06/2026

Everyone's racing to make their website "LLM-friendly." Google's advice? Don't 🧨

On the latest Search Off the Record podcast, the Search team was blunt: don't convert your site to Markdown to help AI understand it.

Crawlers solved HTML decades ago. They're not confused by it. And HTML carries context Markdown can't - headers, nav, internal links - the signals that show how your page fits your whole site.

The bigger trap is building a separate "LLM version" of every page:
💎 two copies to keep in sync
💎 when the AI version breaks, no human reports it
💎 bots keep indexing the broken one

So is Markdown dead? Far from it. It's the best way to write — clean and distraction-free. Just draft in Markdown and publish in HTML.

The one exception: developer docs an AI agent needs to parse. A Markdown version of your shoe catalog? Makes no sense.

For now, getting discovered still runs through plain old HTML. Don't rebuild your site for a standard that doesn't exist yet.

Seeing pressure to ship LLM-specific pages or holding the line on HTML?

18/06/2026

Thankful for the people who made the work worth it. 💪

17/06/2026

Huge kudos to Kaitlin McMichael for the clearest take on this I've read: I don't think SEO A/B testing is as straightforward as people make it sound, and the standard A/B tools most teams already use don't actually work for it 🫳 🎤

Here's why: those tools split users into two groups. But search bots don't take cookies, so Googlebot just sees the page flip between versions every crawl and learns nothing.

The fix is to split pages, not users:
💎 Grab a group of similar pages (same template, comparable organic traffic)
💎 Apply your change to half, leave the other half as an untouched control
💎 Compare organic visits over ~3 weeks

That control group is the whole point. It's the difference between "traffic went up" and "this change drove the lift, not seasonality or a Google update."

Two rules before you start:
💎 Need scale - aim for 50+ pages and 1,000+ daily organic visits per group
💎 Skip before/after "tests" - no control, no real result

Learn how SEO testing can help you validate ideas, accelerate organic growth, and make data-driven decisions with confidence. This practical guide covers SEO split testing, common pitfalls, implementation options, and how to apply testing principles to both SEO and AI search visibility.

15/06/2026

Words from the people I've had the pleasure to work with. 🚀

08/06/2026

Let's talk if SEO is driving visits, but not revenue. 🤝

08/06/2026

Google finally gives you a window into AI search visibility 🔍

Google just launched dedicated Search Generative AI performance reports in Search Console - a standalone view of how your pages show up inside AI Overviews, AI Mode, and generative features in Discover.

What you get:
💎 Impressions broken down by pages, countries, devices, and dates
💎 Separate reports for Search and Discover

The catch: it's impressions only. No clicks, no CTR, no average position, no query data - though Google says more metrics are coming over time. So it answers "how visible am I in AI search?" but not yet "how much traffic is it actually sending me?" 🙃

⚠️ Not yet available in Lithuania.

27/05/2026

Is your SEO agency billing you for fixing meta descriptions and alt tags that their audit tool found? Most of those don't matter 🔥

Credit to Adam Heitzman, whose Search Engine Land piece nailed this: "fix everything" is the wrong SEO strategy.

Audit tools flag a missing H1 on a dead page with the same red icon as a noindex on your homepage. There's no column for "this actually matters."

Usually safe to ignore:
📉 Meta description tweaks on low-value pages
📉 Missing alt tags on non-strategic content
📉 Core Web Vitals micro-optimizations once you've already hit "good"
📉 Redirect chains that don't touch link equity or UX
📉 Thin, zero-traffic posts on topics nobody searches

What actually moves revenue:
📈 Refreshing pages already sitting on page one (highest-ROI work there is)
📈 Lifting rankings stuck in positions 11–30 - they're one upgrade from breaking through
📈 Building out topic clusters that signal real expertise
📈 Fixing true technical blockers: indexation failures, broken canonicals, crawlability

Focus on the work that moves the needle - not the work that raises your Ahrefs Health Score 🫳 🎤

What did your audit flag this month that you can safely ignore?

14/05/2026

Pastaruoju metu su komanda naršėme po AI paieškos ir GEO tyrimus iš visų įmanomų kampų, bandėme suprasti, kas iš tikrųjų veikia.

Kartu su Adomas Šulcas galų gale viską aprašėme. Straipsnyje - dabartinė GEO situacija, tyrimų apžvalga ir principai, į kuriuos verta atkreipti dėmesį.

https://versloidejos.lt/reklama-ir-marketingas/dirbtinio-intelekto-paieskos-optimizavimas-geo-principai-praktikos-mitai/

We've been digging into AI search and GEO research from every angle - trying to separate what actually works from what's just noise.

The result is a thorough look at the current state of GEO: the research, the findings, and the principles worth paying attention to.

Published for the Lithuanian SEO community - but Google Translate will do a solid job if you're curious.

Link in the comments 👇

11/05/2026

AI citations = 120% more clicks 😶‍🌫️ Most brands are leaving them on the table.

Cyrus Shepard - Zyppy study found brands cited in Google's AI overviews get 120% more organic clicks per impression and a 41% increase in paid clicks compared with uncited brands.

He analyzed 54 published experiments, studies, and patents on AI citations across ChatGPT, Gemini, and Perplexity.

Then scored every ranking factor by:
💎 Repeatability across studies
💎 Strength of evidence
💎 Support from patents and official docs

The top three?
🗒️ URL accessibility (9.5): if AI bots can't crawl your page, they can't cite it - and new user agents make this easier to break than ever.
🗒️ Search rank (9.4): 38% of AI Overview citations come from Google's top 10 results, so traditional SEO is still the foundation.
🗒️ Fan-out rank (9.3): AI engines silently run related queries to ground their answers - rank for the whole cluster, not just one keyword.

The biggest surprise: LLMs.txt scored a 2, with no credible evidence it influences AI citations.