Pascal Landau

I'll release a series of articles on the #PHP development on #Docker in the upcoming days. It'll be a continuation of the tutorial series I started ~6 years ago via github.com/paslandau/dock… You can either follow me, this 🧵or subscribe to the newsletter to not miss anything 👇

Am I the only one thinking that this can get horribly wrong? When they talk about "meta directives", this likely means meta robots noindex/nofollow (not robots.txt). Since this requires the page to be crawled first, this sounds like a potentially huge waste of crawl budget.

blog.cloudflare.com/ai-labyrinth/ This is the coolest thing I've ever seen.

The agenda for the 2025 Tech SEO Summit is final since a few days. We are hyped for the deep technical SEO depths our speakers will dive into with you. We also expect a fierce competition for the best tech SEO tip at the conference. 🏆 #techseosummit

Sucht euch einfach einen Bahnhof eurer Wahl, und lasst euch Statistiken anzeigen. Wir sammeln per default für alle Stationen 🇩🇪 Deutschlandweit Daten, ihr könnt aber auch andere Stationen dem Tracking hinzufügen. Weitere Features sind in Arbeit 🎉🥂 >>> trainboard.de.cool/dashboardsearc…

@dejanseo I was rather thinking about code ;)

@PascalLandau Happy to share. Currently trying albert-xxlarge-v2 on top level categories on 4 different machines. Different hyperparameters on each.

Creating The World's Best eCommerce Classifier That's the goal. To have a machine learning model able to automatically classify input text into one or more of the 5,595 categorical classes in Google Product Taxonomy* Training Data & Technical Details Already in place is a synthetic data generation pipeline for each category using Gemma-2-2b-instruct. As I type this there are 233 samples generated. Cute, but I aim to have have ~1,000 training samples per class, totaling 5,000,000 training samples of the following types: [ ]0:[0:11:"Search Query" ]1:[0:21:"Title Tag" ]2:[0:31:"Meta Description" ]3:[0:51:"Category Page Description" ]4:[0:61:"Sentence" ]5:[0:71:"Paragraph" ]6:[0:81:"Page" ]7:[0:91:"Article" ]8:[0:101:"Brochure" ]9:[0:111:"Product Description" ]10:[0:121:"Product Review" ]11:[0:131:"Blog Post" ]12:[0:141:"Customer Review" ]13:[0:151:"Home Page" ]14:[0:161:"FAQ Content" ]15:[0:171:"How-To Guides" ]16:[0:181:"Comparison Content" ]17:[0:191:"Social Media Posts" ]18:[0:201:"Taglines or Slogans" ]19:[0:211:"Product Specifications" ] The above variety is there to provide input diversity and enable model to generalise well on a wide range of input. As my base, I'll work with deberta-v3-large, a model pre-trained by @microsoftai and as a fallback I have Gemma-2-9b, either fine-tuned or as a LoRA adapter. If I don't succeed at 5000 classes, plan B is to work with top-level categories only. Currently considering a custom loss function which penalises using a scalable approach which issues softer penalties to incorrect predictions in the same taxonomy path. In other words if the model makes an incorrect final category, but predicts a direct parent category the penalty will be mild. Likewise if it's a grandparent category the penalty be more severe but not as great if it was a great-grandparent category. Complete penalty applies for making a prediction on a completely unrelated taxonomy tree. Thinking about allowing softmax to represent prediction probabilities on inference and allow end-user to decide what to do with that information as classification down to deep categories can be very subjective. Having a top choice and a few runner-ups might be nice. Hope this works. I'll post my progress from time to time. *google.com/basepages/prod…

📢 We're looking for a new Head of SEO at ABOUT YOU More details and some insights on our philosophy with regards to data & our SEO data warehouse over at LinkedIn 👉 linkedin.com/posts/pascalla… If you're team "😍" please reach out 👋

Ecommerce SEO: The Keys for Success Now & Beyond - My presentation from #SERPConf2024 👀 going through: 1. Maximize Your PDPs optimization efforts 2. Facilitate Google access and understanding of your product related content with SD, Merchant Center Feed and Image Optimization 3. Grow your brand authority with informational content investment … and more covered in the presentation: speakerdeck.com/aleyda/ecommer…

@nikrangerseo @darth_na @thetafferboy @lilyraynyc Can you elaborate on how you build the model? I was thinking of something like position, visual position (in pixels), serp features on the serp, serp features of the snippet - but that would require crawling the serp (only feasible for the most important keywords)

@darth_na @thetafferboy @lilyraynyc Creating CTR models, namely a distribution curve from GSC data to look at the expected CTR against rounded average position, then will test this against actual CTR to find patterns, anomalies - to segment and prioritise for experiments.

Navboost uses clickstream data - it’s embedded within the code. It’s a boosting mechanism that considers user engagement, how long the user spends on the page, etc. This is why CTR optimization is so important. @nikrangerseo #AhrefsEvolve

@nikrangerseo @thetafferboy @darth_na @lilyraynyc @aleyda Same here - would love to see it as well!

@thetafferboy @darth_na @lilyraynyc Of course, I am a fan of you and your work would be happy to share. I’ll upload it to slide share, send it to @aleyda as I’ve promised and share it with you too 🥰

@rdohms @magalu Haha I first read this as "10 out of 10" and was wondering why they are so excited about the estimate 😁

Deliveries in Brazil. @magalu says: your package is on a truck! And good news, the delivery estimate is 10/10!! Sadly, the message was sent on 11/10... So I'm not sure the news is that good 😂

Are you looking for a new SEO job? There are quite a few 🔥 open positions in the Free #SEOFOMO Job Board 👀👇 Check them out: hub.seofomo.co/seofomo-jobs/

SEO updates you ✨NEED✨ to know [30 Sep]: ⚠️ ​Google has updated its spam policy with several notable changes, including talking about sites using "extensive automation to product content", likely meaning human-tweaked AI content. via @Marie_Haynes 📊 ​Some recent recoveries after the August core update are being reversed. The volatility continues as some sites that were hit particularly hard by last year's HCU are now seeing their small gains disappear again. via @abbysuegleason 🎠 ​Google is testing a 'from small businesses' carousel. This is being tested on mobile and is likely connected to the 'small business' attribute in Google Business Profile. via @rustybrick 🧠 ​Cloudflare launches Speed Brain, a speculative model that can reportedly reduce LCP by up to 75%. It does this by prefetching the most likely next pages' content and is now available for all Cloudflare users at no extra cost. 📉 ​Forbes Advisor has recently lost rankings for over 1.7 million queries. This may be the result of a manual penalty and related to the huge parasite SEO problem Forbes has had recently. via @glenngabe 🏷️ ​Google is testing 'For You' and 'Preferred Source' labels. This appears to be another step towards the personalisation of the main SERP that we've seen previously in Google Discover and Google News. 🚫 ​WP Engine has been banned by WordPress. If you're using their plugins, tools or hosting, you may want to look for an alternative solution even though the ban is being temporarily lifted. ✨ ​Google now highlights content creators and their expertise in Knowledge Panels. This label may be a significant boost in credibility for content creators as trusted online sources. via @jasonmbarnard 🗳️ ​SearchPilot confirms no benefit to using JSON-LD or Microdata for your markup. Despite Google listing JSON-LD as best practice, there appears to be no advantage or disadvantage either way. via @rida_a95 🤖 ​Cloudflare's AI Audit will allow sites to charge AI bots for scraping content. This is a step in the right direction of offering some compensation for content creators whose work is being "stolen" by LLMs and AI-powered search. 💰 ​Google adds support for sale pricing and priceType structured data. This should make it easier for users to specify sale prices and directly compare them to the full price, original listing. 🔍 ​Google removes the cache: search operator. Google had previously announced this in March 2024 and has since reacted by including links directly to the Internet Archive (which I covered two weeks ago) via @martinibuster 💌 If you like these kind of tldr SEO updates and don't want to miss them, I send them out every Monday along with SEO deeps and a deep dive podcast - Just Google: Core Updates newsletter Links to everything in the comments ⬇

Google Cache is now fully dead - Google disabled it this morning, you can now longer do a cache:URL operator - seroundtable.com/google-cache-d…

Note to self: Keep this in mind whenever someone claims to have figured out how to recover from an algo update 🙃

👀 The judge in the DOJ vs. Google trial has ruled the Google violated antitrust laws. I'm no lawyer, but I'm super interested in this case. I thought my day was done, but instead, let's dig in and see what's interesting in this 286 page document. (David has brought me wine🍷. We could be here for a while...)

I 'hacked' Chrome and used its shopping intent classifier to build an eCommerce image optimisation tool for my team. Details in the article: dejanmarketing.com/product-image-…

@brodieseo @googlesearchc I was wondering the same thing. @googlesearchc could we get an official statement that the GMC setting is indeed equivalent to the structured data and that hence the warning can be safely ignored? A note in the corresponding docs developers.google.com/search/docs/ap… would be great as well

@googlesearchc Interesting new addition. A quirk in GSC reporting is that when using GMC for returns policy info, it’s flagged as missing in GSC ‘merchant listing’ reporting (because not added with structured data). Will this report need to change with the new Organization markup method?

Today we're adding support for return policies in Organization markup, making it easier to define a return policy for your entire business, instead of having to specify a separate return policy for each individual product you sell. Learn more at developers.google.com/search/blog/20…

@g33konaut @JohnMu @methode Gotcha. One last question for the sake of completeness: Once all URLs are removed from the index, is it then appropriate to block them from being crawled again via a robots.txt? FYI: The vast majority (99.99..%) will not have any links pointing to them.

Tech SEO question for @JohnMu / @methode / @g33konaut: What's the fastest way to get rid of a large (1mio+) nr of URLs in the index for the purpose of "removing low quality pages"? URLs are already in the index but can be blocked entirely with a simple robots.txt rule BUT (1/3)