Substack Search: Hybrid, Not Semantic | Generated by AI

Home 2026.07

Question: Is Substack search good? For example, can I search Yin Wang’s Substack for “时间” and get semantic matches, not only exact keyword matches?

Answer:

Yes, Substack search is much better than a simple grep "时间" style search, but it is not a full semantic search engine like modern AI search.

For a publication like Yin Wang Substack, Substack provides search for posts/publications/topics through its search bar. (Substack Support)

Your observation is basically correct:

This is closer to hybrid search:

query = "时间"

score =
    keyword_match_score
  + title_similarity
  + content_similarity
  + popularity/relevance signals

rather than:

grep -r "时间" posts/

A pure keyword engine would miss:

"人生为什么有限"
"不要浪费生命"
"过去与未来"
"意识和存在"

because they may not contain the character “时间”, but semantically relate to time.

However, Substack search is not as powerful as building your own RAG system:

Yin Wang articles
        |
        v
crawl markdown/html
        |
        v
chunk documents
        |
        v
embedding model
        |
        v
vector database
        |
        v
query: "王垠关于时间的观点"
        |
        v
semantic retrieval

Then you can ask:

"Find all places where Yin Wang discusses time,
including related ideas like life, existence, change, and consciousness."

A local implementation would be simple:

from sentence_transformers import SentenceTransformer
import faiss

model = SentenceTransformer("BAAI/bge-large-en-v1.5")

docs = load_yinwang_articles()

embeddings = model.encode(docs)

index = faiss.IndexFlatIP(embeddings.shape[1])
index.add(embeddings)

query = "王垠关于时间是什么的思考"

q = model.encode([query])

scores, ids = index.search(q, 10)

for i in ids[0]:
    print(docs[i])

This would probably outperform Substack search for deep philosophical research.

For someone like you who already has:

a very natural project is:

personal knowledge engine

Markdown
   |
   v
chunking
   |
   v
embedding
   |
   v
FAISS / Qdrant
   |
   v
LLM answer with citations

Basically a personal “AI second brain”.

References:

Back Donate