Context decides what your words mean

On October 25 Google announced BERT — short for Bidirectional Encoder Representations from Transformers — and did not undersell it, calling it the biggest leap forward in five years and one of the biggest in the history of Search. Stripped of the acronym, BERT is a machine-learning technique for natural language processing whose job is to help Google understand what a search query means more the way a person would. The defining idea is right there in the first word: it is bidirectional. Where older approaches read a query in one direction — left to right, or right to left — and tended to weigh words more or less in sequence, BERT looks at the words before and after a given word simultaneously, using both sides as context to settle what that word means here. Google says it affects roughly one in ten searches, for now English-language queries in the US with more languages to come, and that it applies to featured snippets as well as ordinary results. It is a cousin of RankBrain — another machine-learning system aimed at understanding language — but it does not replace RankBrain or Google’s other methods; depending on the query, Google may use BERT alone, in combination, or not at all. One curiosity worth noting is that the update barely registered on the usual rank-tracking tools, which surprised an SEO community braced for a five-year-sized shake-up. The reason is instructive: those tools mostly track shorter keyword queries, while BERT does its work on the longer, more conversational searches where the relationships between words actually carry the meaning. The quiet surface hides a real shift underneath — Google got materially better at reading language, by reading words in context instead of in isolation.

The most concrete way to feel what BERT changes is to look at the words it rescued, because they are the ones that used to disappear. Before BERT, Google leaned on the obvious content words in a query — the nouns and verbs that look like keywords — and tended to underweight or skip the small connective words around them: prepositions like “to” and “for,” and, most consequentially, negations like “no.” The trouble is that those small words frequently carry the entire meaning. Google’s own illustration is the query “2019 brazil traveler to usa need a visa.” The little word “to” fixes the direction of travel, but the old systems could miss it and return results about US citizens heading to Brazil — the exact opposite of what was asked. Another is “parking on a hill with no curb,” where Google admits it once put too much importance on “curb” and ignored “no,” which reverses the entire request. Bidirectional reading is the fix. Because BERT considers the words on both sides of a term at once, it can register that “to” links Brazil and the USA in a particular direction, or that “no” negates “curb,” rather than treating each token as an isolated thing to match against a page. Human language is layered and the sense of a word leans on its neighbours; bidirectional processing is the engine finally modelling that dependency instead of flattening it. The practical upshot is that the connective tissue of a sentence — the part that was effectively invisible to the engine before — now counts toward what Google thinks you meant.

Underneath the examples sits a principle that is the real takeaway, and it is one about language itself: a word does not carry a fixed meaning that an engine can read off in isolation — its meaning is settled by the words around it. “Bass” is a fish or a frequency; “charge” is a fee, an accusation, or an electrical current; “rose” is a flower, a colour, or the past tense of rise. None of these resolves until the context resolves it, and that is exactly what BERT is built to do: read the surrounding language and let it disambiguate. This reframes what an engine is doing when it reads your content. It is not scanning for keyword tokens and counting matches; increasingly, it is reading sentences and inferring meaning from how the words relate. That has a direct implication for how you should write, and it runs opposite to an old habit. For years, a strand of SEO writing consisted of “keyword-ese” — repeating target phrases, clipping sentences down to their key terms, and stripping out the ordinary connective language a person would use — all of it a concession to an engine that could not read context and had to be fed keywords. BERT is the point at which that concession stops paying. An engine that reads context understands clear, complete, naturally written language better than a keyword-stuffed approximation of it, because the context is precisely what it now uses to determine meaning. Writing so that meaning lives in the language, where a context-reading engine can find it, rather than in a scatter of isolated keywords, is the entity-and-language discipline the AC Group has worked by for {years} years.

Translate the principle into how you actually write, and resist the urge to look for a BERT “tactic,” because there is not one — you cannot optimize for BERT any more than for RankBrain, and Google’s own guidance is simply to write for users. Concretely, start by dropping keyword shorthand: where you once might have compressed a heading or sentence to its target phrase, write the full, natural sentence a person would actually say, since the small connecting words you used to delete are exactly the ones the engine can now read and use. Answer the question your page addresses directly and without hedging, so that the context around your key terms makes their intended meaning unmistakable — you control the neighbours that pin down whether your “charge” is a fee or a current. Aim your content at the real intent behind the search, with particular attention to the longer, more conversational, more specific queries where BERT operates and where a page that genuinely answers a precise need can now rise above the broad, vague pages that used to crowd it out. And keep your structure clean — clear headings, direct answers, no filler or padding — so nothing dilutes or muddies the meaning you are conveying. Notice that none of these are tricks aimed at the algorithm; they are just the marks of writing clearly for a reader, whether that reader is a person or a machine that reads like one. That is the whole shift: you are no longer writing around an engine’s blind spots, you are writing for one that reads context the way your audience does. Making your meaning unmistakable, carried by clear language rather than isolated keywords, is the entity-and-language discipline the AC Group has worked by for {years} years.