How Google's BERT algorithm affects your website traffic

When Google began applying transformer-based language understanding to search, a lot of SEO advice became obsolete overnight. The advice that replaced it was mostly worse, because it treated a comprehension improvement as though it were a ranking penalty.

What actually changed

Search moved from matching strings to modelling meaning. A query like "can you get medicine for someone pharmacy" now resolves correctly, where previously the preposition was effectively discarded and the result set was wrong in a way nobody could diagnose.

The practical consequence: pages that answered a real question in plain language started performing better, and pages assembled around keyword density started performing worse. Not because they were penalised, but because they were never good answers and the system finally noticed.

What to do about existing content

Audit for intent, not volume

Take your top fifty pages and, for each, write the question a visitor was asking when they arrived. If you cannot state it in one sentence, the page has no clear job and rewriting it will do more than any technical change.

Structure for extraction

Clear headings, direct answers in the first paragraph under each, and lists where the content is genuinely a list. This is not a trick; it is the same structure that helps a human skim.

Stop writing for the middle

Content aimed at a broad, undifferentiated audience now competes with content written for a specific one, and loses. Depth on a narrow question consistently outperforms breadth on a wide one.

What has not changed

Site speed, crawlability, internal linking and information architecture still determine whether good content is ever found. Language understanding raised the ceiling on what content can achieve; it did nothing about the floor.

The most reliable SEO strategy remains the least fashionable one: work out what your audience is actually trying to do, and be the most useful page on the internet for that.