Wissenswertes

Key Takeaways: Tran ran a nice experiment on PDF extraction, especially for tables, using different models/libraries In short: LlamaParse wins on speed and accuracy. It’s the fastest overall and produces the cleanest output, but it requires sending PDFs to LlamaCloud. Marker is the best local option. It’s faster than Docling and handles simple tables well, but it merges columns on dense layouts. Docling is the slowest of the three and prone to hallucinating values on dense ta

Key Takeaways: Wills analyzed citations in ChatGPT of 100k+ prompts including 145 industries and 1.595 personas to draw conclusions on Citation factors: 1. SEO is the foundation. OpenAI is using search data today and building their own index. As that matures, the connection between search authority and LLM visibility deepens. Traditional SEO principles are not obsolete, they're the starting point for LLM visibility too. 2. Persona is the measurement unit. The #1 airline for a

Key Takeaways: Konitzny analyzed one prompt in 10 different markets in ChatGPT, Perplexity and AI Mode ""I need a juicer that produces little waste and is not very loud. Price doesn't matter."" Retrieval via Peec AI 10 Markets: US, DE, FR, NL, UK, IT, DK, AUT Listicles: biggest share on average, but massive variance by market → dominate the US. In Germany and the UK, they account for roughly half that share (US: 53.9% / DE: 25.9% / UK: 25.7%) Reddit: almost entirely a US phen

Key Takeaways: Michalik created 3 listicles for a made up matcha tea that got listed in most major AI chats Claude and AIO cited the fake Matcha quite frequently ChatGPT and AI Mode also cited the fake product from time to time Gemini did the best job in not mentioning the fake product Now all chats caught up, still, way to easy to game in the short run A while ago Fishkin reported that only in <1% of the runs of the same prompt the KI chat returns the same list of recommen

Key Takeaways: DEJAN/Dan Petrovic analyzed how ChatGPT retrieves websites for grounding and outlined it on an example of his page using the Web Search tool in the Asisstants API. How it works in a nutshell: Initially retrieves small data object from web search results (title, description, 1-3 sentences, retrieval ID) Then, it can look at certain windows/passages and even follow links It uses an open() call to access the page at a certain row and click() to initiate the same

Key Takeaways : Dan Petrovic tested the use of Schema Structured Data mark-up on a webpage of his asking ChatGPT 5 to summarize it There was no evidence of Schema even reaching the model The answers with and without Schema were quite similar ChatGPT itself replied multiple times that the browsing tool does not provide any Schema structured data just plain text from the raw HTML Inspired by a previous experiment that found impact of Schema on former ChatGPT models -...

Key Takeaways : Google Experiments are exposed to the public with states 'Experiment', 'Control', 'Treatment', 'Launch' 🤖 AI Focus: The...