APProjects/uspto-trademark-gazette-word-marks
USPTO Trademark Official Gazette — word marks, frozen snapshot (no longer updated) This dataset is a frozen snapshot and is no longer updated. The last Gazette issue it contains is 2026-09-01. The project that maintained it (TM Watch) was discontinued on 2026-09-04, and the USPTO bulk source it was rebuilt from now requires an ID.me-verified Open Data Portal key we cannot obtain. The rows below are exactly as published by the USPTO and remain public domain (CC0) — treat them as… See the full description on the dataset page: https://huggingface.co/datasets/APProjects/uspto-trademark-gazette-word-marks.
USPTO Trademark Official Gazette — word marks, frozen snapshot (no longer updated)
This dataset is a frozen snapshot and is no longer updated. The last Gazette issue it contains is 2026-09-01. The project that maintained it (TM Watch) was discontinued on 2026-09-04, and the USPTO bulk source it was rebuilt from now requires an ID.me-verified Open Data Portal key we cannot obtain. The rows below are exactly as published by the USPTO and remain public domain (CC0) — treat them as a historical extract, not a live feed, and check TSDR for the current status of any serial number.
Every word mark published for opposition or registered in the USPTO Trademark Official Gazette over a ~120-day window ending 2026-09-01, parsed from the weekly ST.96 XML issues by a pipeline that has since been retired. 361,260 rows across 18 weekly issues (2026-05-05 → 2026-09-01). Newest issue: 2026-09-01. Design-only marks are excluded (no text to match on).
This was the index behind TM Watch — a free, in-browser similarity check (edit distance + phonetic + variant forms, benchmarked on real §2(d) pairs; matcher and benchmark are open source at github.com/APVentureEngine/trademark-watch).
Quickstart
from datasets import load_dataset
ds = load_dataset("APProjects/uspto-trademark-gazette-word-marks", split="train")
print(ds[0])import pandas as pd
df = pd.read_csv("https://huggingface.co/datasets/APProjects/"
"uspto-trademark-gazette-word-marks/resolve/main/data/gazette_word_marks.csv")
df[df["event"] == "published"].head()Schema
Permanent per-issue copies
This file covers a ~120-day window ending 2026-09-01. If you need a citable, immutable copy of a single Gazette issue, the same records are published as one CSV per issue (never rewritten) at apventureengine.github.io/trademark-watch/data/ — public domain (CC0), no login, with a manifest.json listing every issue.
Why a rolling window
A mark published for opposition can be opposed (or an extension of time requested) for 30 days from its Gazette date, so this file was built as a rolling recent window. It no longer rolls: nothing has been added since 2026-09-01, and every opposition window in it has closed. It is useful as a historical extract of what published in that period, not as a watch list. The full historical bulk data was formerly on bulkdata.uspto.gov (decommissioned 2026-06) and now requires an authenticated USPTO Open Data Portal key; the Gazette issues themselves are public.
Not legal advice
Similarity flags are for human review; nothing here is an opinion on likelihood of confusion. Source data is a US government work (public domain).
