June 9, 2026
Ctrl+F meets main character syndrome
The Evolution of 'More Like This'
Search’s ‘find similar’ makeover has commenters fighting over old-school word matching vs spooky AI vibes
TLDR: The article says “find similar” search is moving from simple word matching to AI-style meaning matching, especially when users start from an existing article, product, or ticket. Commenters are split between people who want exact results they can trust and people who want smarter matches that catch the same idea in different words.
This was supposed to be a calm explainer about how the humble “more like this” button evolved from matching obvious shared words to using modern AI-style similarity. Instead, the community turned it into a full-on food fight over whether search should be literal, clever, cheap, or all three. The article’s basic point is simple: older systems found similar items by spotting the same important words, while newer ones try to understand meaning even when the wording changes. Great in theory — and absolute catnip for commenters.
The loudest camp basically yelled, “If I searched for an error code, don’t give me its emotional cousin.” They argued that for things like product IDs, legal text, bug reports, and exact phrases, old-school word matching is still the dependable adult in the room. Another faction fired back that people don’t always describe the same problem the same way, and that modern tools are better at catching the idea instead of just the exact wording. Cue the classic comments-section melodrama: one side calling the new approach magical, the other calling it a very expensive way to be vaguely wrong.
And yes, the jokes wrote themselves. People compared AI search to a friend who hears “memory leak” and replies, “Have you considered vibes?” Others pitched the ideal solution as a messy but lovable combo platter: use exact matching first, then let the smarter system sort the near-misses. In other words, the hottest take wasn’t “old bad, new good” — it was “stop pretending one button can read minds.”
Key Points
- •The article defines More Like This as search that starts from an existing document and aims to find similar documents.
- •It contrasts classic lexical MLT, based on term extraction and full-text ranking, with modern embedding-based similarity search using vectors.
- •The glossary introduces related search concepts including embeddings, KNN, ANN, RAG, hybrid search, and reranking.
- •Classic lexical MLT used mechanisms such as TF-IDF, BM25, stopwords, field boosts, and document-frequency thresholds, often exposed through parameters like `min_term_freq` and `max_query_terms`.
- •Lexical MLT remains effective for exact identifiers and stable wording such as error codes, SKUs, stack traces, and legal text, but can miss synonyms, paraphrases, and cross-lingual matches.