Google: keywords in URLs are an incredibly small ranking signal

Google has posted the first video in a series it calls the Non-Canonical Cabinet, with Gary Illyes and Cherry Prommawin of Google’s Search Relations team going through how Search handles URLs. Search Engine Roundtable’s write-up of October 14 quotes Illyes calling keywords in a URL or a domain name an “incredibly small signal” for ranking. Illyes also said words in a path should be separated with hyphens rather than underscores, and that Google can detect spamming in URLs.

Google has been saying all of this for years, and the repetition is the useful part. Search Engine Roundtable’s post lists more than a dozen of its own earlier stories on the same two questions, going back to Matt Cutts, who then ran Google’s webspam team, arguing for dashes over underscores. A position restated that consistently is unlikely to invert next quarter, so a URL strategy built around keyword placement rests on a signal Google has called negligible for years. Ranking claims attached to URL paths keep appearing anyway, including the recent argument that path structure changes how often AI assistants cite a page, which arrived without a test behind it.

The hyphen advice is about parsing, not keyword weight. Google’s URL structure documentation recommends hyphens because underscores are commonly used to mark words that belong together, as in a function name like format_date. On a URL that does not exist yet, picking the right separator costs nothing. On a URL that already ranks, swapping separators rarely pays for itself, because renaming paths means redirects, and the gain on the other side is a signal Google has just called tiny.

The spam point is the one to pass on to clients who still want keywords in the path. Illyes made a point of saying Google can detect spamming in URLs, and the video asks whether a path like best-grilled-chicken-cheap-delicious-amaze-amaze-amaze makes a page number one forever. A path built that way chases a signal Illyes has just called tiny, and advertises the attempt while doing it.

Most of the rest of the video is about what can go in a URL at all rather than what ranks. It covers internationalized domain names and Punycode, including why a hostname beginning with xn-- is ordinary encoding and not a typo. It also covers emoji, with two answers worth knowing: an all-emoji domain cannot exist, and emoji in a URL path are percent-encoded, which turns a short path into a string nobody can read once it is shared or pasted into a ticket.

What to do

• Use hyphens between words in new URL paths, keep the path short and readable, then leave it alone.
• Do not rewrite live paths to add keywords or to swap underscores for hyphens. The ranking gain is close to zero and the redirect work is real.
• Keep emoji and other non-ASCII characters out of paths that people will copy, share, or read in logs. Punycode exists for hostnames that need non-ASCII characters, not for decoration.
• When a client asks for a keyword-stuffed domain or folder name, Google’s answer is on record: it does almost nothing, and Google can see what it is.