Robert Galasso
10 essays

Writing

Each essay covers something I have built or run and is written so someone else can follow the method.

  1. Curating training data for diffusion models: what the research shows

    What peer-reviewed research says about curating training data for diffusion models: duplicates and memorization, filtering against repetition, synthetic captions, and small hand-picked fine-tuning sets.

  2. Do AI Overviews reduce clicks? What the independent research shows

    What Pew, the Reuters Institute, Columbia's Tow Center, and peer-reviewed studies found about AI Overviews and clicks, how accurate AI search citations are, and what an SEO program should change.

  3. Why language models hallucinate, and what the research says reduces it

    What current research says about why LLMs hallucinate, why retrieval doesn't stop it, how to detect confabulation, and how to build AI systems that say when they don't know.

  4. Generative engine optimization research: what holds up in practice

    A working read of the 2025-2026 arXiv research on generative engine optimization: the critical survey, the measurement papers, the brand-stature data, and what a practitioner should change.

  5. Generative engine optimization: how brands get cited by AI

    GEO is how brands earn citations in ChatGPT, Gemini, Perplexity, and AI Overviews. The on-page patterns and entity signals that work, plus how I measure it.

  6. How to get a Google Knowledge Panel when you share your name

    How to get a Google Knowledge Panel when someone famous shares your name: the entity SEO playbook I'm running, from disambiguatingDescription to Wikidata.

  7. How to measure AI visibility using confidence intervals

    How to measure AI visibility in ChatGPT and other LLMs: repeat sampling, Wilson confidence intervals, and alerts that fire only when intervals separate.

  8. How to train your own FLUX checkpoint: lessons from 12 versions

    What it takes to finetune FLUX.1 on your own GPUs: dataset curation, captioning, eval suites, and versioning from 12 shipped Nepotism releases.

  9. Lessons from building an agency's internal AI platform solo

    Build vs buy, fail-closed AI design, numbers that admit uncertainty, and the adoption fight: what shipping an agency's internal AI platform solo taught me.

  10. Technical SEO for AI search: what still matters in 2026

    What technical SEO means now that LLMs crawl your site: server-rendered HTML, entity signals, citation-ready structure, and some llms.txt skepticism.