BERT SEO: What Changed and What Didn’t

BERT SEO refers to the practice of optimising content for Google’s BERT algorithm, a natural language processing model Google rolled out in 2019 that changed how the search engine interprets the meaning behind queries, not just the keywords within them. Where earlier algorithms matched words, BERT reads context, understanding how the position of a word in a sentence changes its meaning entirely.

The practical implication is straightforward: content written to satisfy a keyword pattern without genuinely answering the question behind it became less competitive overnight. BERT didn’t punish keyword-stuffed pages so much as it rewarded pages that were actually useful, and that distinction matters more than most SEO commentary acknowledges.

What BERT Actually Does Inside Google’s Algorithm

BERT stands for Bidirectional Encoder Representations from Transformers. Google published the underlying research in 2018 and began applying it to search results in late 2019, initially in English before expanding to other languages. The “bidirectional” part is what separates it from earlier language models: it reads a sentence in both directions simultaneously, which means it understands how words relate to each other across the whole sentence rather than just left-to-right.

Consider a query like “can you get medicine for someone at the pharmacy.” Before BERT, Google might have matched that query to pages about pharmacy services in general, because the individual words pointed there. BERT understands that the word “for” changes the meaning entirely. The user wants to know about picking up someone else’s prescription, not just about pharmacies. That’s a different question, and it deserves a different answer.

Google’s own announcement at the time stated that BERT affected around one in ten searches in English, which sounds modest until you consider the volume of searches Google processes. The queries most affected were long-tail, conversational, and question-based, the kind of searches that have grown as voice search and mobile use have normalised more natural phrasing.

BERT also applies to featured snippets, which is commercially significant. If your content earns a featured snippet, it appears above the standard organic results. BERT’s ability to match nuanced queries to precise answers made featured snippet eligibility more about genuinely answering a question than about formatting tricks alone.

Why Most “Optimise for BERT” Advice Misses the Point

When BERT launched, the SEO industry produced a wave of tactical guidance: write conversationally, use natural language, avoid keyword stuffing. All of that is correct. None of it is new. The advice existed before BERT and it will exist after whatever comes next. BERT didn’t introduce a new set of tactics. It made the existing fundamentals harder to game.

I’ve spent a long time in agency environments watching content teams chase algorithm updates as if each one required a new playbook. When I was running performance at iProspect, we’d see the same pattern after every major Google update: a flurry of reactive content audits, tactical pivots, and client calls asking what had changed. The honest answer, more often than not, was that the clients whose content was already genuinely useful barely moved. The ones who’d been gaming thin content or keyword density took the hit.

BERT is a clearer version of that pattern. The update didn’t reward “natural language content” as a new content type. It rewarded content that answered real questions with real depth. If you were already doing that, BERT was largely irrelevant to your rankings. If you weren’t, BERT exposed the gap.

The tactical framing also misses something structural. BERT is one of many signals Google uses. It doesn’t override domain authority, technical health, or link quality. A page with strong BERT-friendly content on a weak domain will still lose to a mediocre page on a strong domain in most competitive categories. Understanding where BERT fits within the broader ranking system matters more than treating it as a standalone lever. For a fuller picture of how these signals interact, the complete SEO strategy hub covers the full ranking picture in more depth.

The Queries BERT Changed Most, and What That Means for Your Content

Not all queries were affected equally. BERT had the most visible impact on three categories: long-tail queries with specific phrasing, queries containing prepositions and function words, and conversational queries that read like spoken questions rather than typed keyword strings.

Function words, words like “for”, “to”, “without”, “not”, carry meaning that earlier keyword-matching systems largely ignored. A query like “flights to London without a layover” and “flights to London with a layover” would have returned similar results under older systems because the core keywords matched. BERT understands that “without” and “with” produce fundamentally different user needs.

For content strategy, this has a specific implication. If you’re targeting long-tail queries, the precision of your language matters. Writing a page that broadly covers a topic isn’t enough if the user’s question is narrow. The content needs to answer the specific question, including the function words that define its meaning.

Conversational queries matter here too. As voice search and mobile-first behaviour have shifted how people phrase searches, the gap between typed keyword queries and natural spoken questions has narrowed. A user searching by voice is more likely to ask “what’s the best way to track SEO performance without spending a lot” than to type “SEO performance tracking tools.” BERT handles the former with more precision than older systems could. Content that answers the full question, not just the keyword cluster, performs better in those results.

The practical output here is a content brief that starts with the actual question, not the keyword. What is the user trying to accomplish? What do they need to know to accomplish it? What would a genuinely useful answer look like? Those questions produce better content than a keyword density target does.

How to Write Content That BERT Rewards Without Overcomplicating It

The practical guidance here is less tactical than most SEO content suggests, and that’s intentional. BERT rewards clarity of thought. If your thinking is clear, your writing will be clear, and clear writing answers questions precisely. That’s the whole loop.

A few principles that hold up in practice:

Answer the question directly, early. BERT is particularly relevant to featured snippets, where Google extracts a specific answer from a page. If your answer is buried in paragraph six after three paragraphs of preamble, the algorithm has to work harder to find it. Front-loading the direct answer, then expanding on it, is both good writing and good BERT optimisation. These two things are not in tension.

Write for the question, not the keyword. This sounds obvious but the execution is different. A keyword like “BERT SEO” tells you the topic. The question behind it might be “has BERT changed how I should write content” or “what does BERT mean for my existing pages” or “how do I know if BERT affected my rankings.” Those are different questions with different answers. Identifying which question your target audience is actually asking, and writing to that question, produces content that BERT can match to the right queries.

Use complete sentences and complete thoughts. Fragmented bullet-point content that strips out connective language loses the contextual signals BERT uses. This doesn’t mean avoiding bullets entirely. It means not reducing your content to keyword fragments dressed up as structure. If a bullet point can’t stand alone as a coherent thought, it’s probably not serving the reader or the algorithm.

Cover the topic with appropriate depth. BERT can identify whether content genuinely addresses a topic or merely mentions the relevant terms. Thin content that hits keywords without building understanding performs worse under BERT than it did under earlier systems. Depth doesn’t mean length for its own sake. It means covering the question thoroughly enough that a reader leaves with what they came for.

I’ve reviewed a lot of content audits over the years, both internally and as part of client strategy work. The pages that consistently underperform share a common characteristic: they were written to a brief that started with a keyword and ended with a word count target. The brief never asked what the reader actually needed. BERT didn’t create that problem, but it made it more expensive to ignore.

BERT, E-E-A-T, and the Relationship Between Them

BERT and E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) are separate systems that interact in practice. BERT handles language understanding. E-E-A-T is Google’s quality framework for evaluating whether a source is credible enough to rank for a given query. They’re not the same thing, but they pull in the same direction.

Where they connect is in what they both reward: content written by someone who genuinely knows the subject. A subject matter expert writing naturally about their field produces content that satisfies BERT’s language understanding requirements and E-E-A-T’s credibility requirements simultaneously. They’re not separate optimisation tasks. They’re the same task approached from different angles.

The implication for content production is that who writes the content matters. A content generalist writing from a brief produces different output than a practitioner writing from experience. The practitioner uses different language, covers different nuances, and answers questions that the brief didn’t anticipate because they know which questions matter. BERT is better at recognising that difference than earlier systems were.

I’ve judged the Effie Awards, and one thing that’s consistent across the entries that perform well is that the strategy behind the work is visibly grounded in real understanding of the market, the customer, and the business problem. The entries that don’t perform well often have polished presentation but shallow thinking underneath. BERT is doing something structurally similar in search: it’s getting better at distinguishing surface polish from genuine depth.

What BERT Means for Technical SEO and Site Architecture

BERT is primarily a content-layer development, not a technical one. It doesn’t change how Google crawls or indexes pages. It changes how Google interprets the content on those pages once they’re indexed. Technical SEO fundamentals, site speed, crawlability, structured data, mobile performance, remain relevant and unaffected by BERT directly.

Where technical and content considerations intersect is in structured data. Schema markup helps Google understand the structure and context of your content. It doesn’t replace BERT’s language understanding, but it provides explicit signals that complement it. For FAQ content, HowTo content, and review content, structured data can increase the likelihood of rich result features that BERT-matched content might otherwise miss.

Site architecture matters in a related way. If your content is well-organised, with clear topic clusters and logical internal linking, Google’s ability to understand what each page is about improves. BERT can interpret language at the page level, but site architecture provides the broader context of what a site covers and how deeply. A well-structured site with genuinely useful content compounds both signals.

Tools like Semrush’s website metrics can help you identify which pages are underperforming relative to their topic relevance, which is a useful diagnostic when you’re trying to understand whether content quality or technical factors are the primary constraint. The two are often conflated in post-BERT audits when they deserve separate analysis.

How to Audit Your Existing Content for BERT Alignment

An audit framed around BERT alignment is really an audit of content quality and intent matching. The BERT framing is useful because it focuses attention on the right questions, but the output is a standard content improvement plan.

Start with pages that lost ranking positions around or after October 2019, when BERT rolled out in English. That’s not a definitive signal, other updates occurred in that period, but it’s a reasonable starting point for identifying pages that may have been over-indexed on keyword matching rather than genuine intent satisfaction.

For each underperforming page, ask three questions: What is the user’s actual question? Does this page answer it directly and early? Does the depth of coverage match the complexity of the question? If the answer to any of those is no, you have a content improvement opportunity that sits within BERT’s reward structure.

Also look at your long-tail query performance in Search Console. If you’re ranking for head terms but not for the long-tail variants around the same topic, that’s a signal that your content is matching keywords but not the fuller range of questions users are asking. Expanding coverage of specific question variants, using natural language rather than forced keyword insertion, typically improves performance across that query cluster.

A domain-level overview can be useful context here. Moz’s domain overview reporting gives you a baseline for understanding where your authority sits relative to the queries you’re targeting, which matters because BERT-aligned content still needs domain authority to compete in high-volume categories.

One thing I’d caution against is running a BERT audit as a standalone exercise disconnected from your broader SEO strategy. I’ve seen agencies sell BERT audits as a distinct service, and the problem is that the recommendations are almost always the same as good content strategy recommendations. Framing it as BERT-specific creates the impression that a discrete fix exists, when the real work is ongoing content quality improvement. The SEO strategy hub covers how content quality fits within the full picture of sustainable organic performance.

BERT and the Future of Language Understanding in Search

BERT was a significant step, but it wasn’t the last one. Google has continued developing its language understanding capabilities, with MUM (Multitask Unified Model) announced in 2021 as a more powerful model capable of processing information across text, images, and eventually other modalities. The direction of travel is clear: Google is getting better at understanding what users mean, not just what they type.

The strategic implication is that optimising for specific algorithm iterations is a diminishing returns exercise. The underlying principle, write content that genuinely answers real questions with real depth, is stable across algorithm generations. BERT rewarded it. MUM rewards it more. Whatever comes after MUM will reward it further.

This is the part of SEO that I find most commercially interesting. The tactics change. The principle doesn’t. Marketing that starts with a real understanding of what the audience needs and builds something genuinely useful tends to survive algorithm updates better than marketing engineered around the current rules. That’s not a philosophical position. It’s an observation from watching enough content audits and ranking fluctuations to see the pattern clearly.

The SEO industry has a tendency to over-index on the new thing, BERT, then MUM, then whatever Google announces next, and under-index on the boring fundamentals that compound over time. Understanding how to explain the value of SEO in business terms, rather than algorithm terms, is a more durable skill than tracking every model update.

The marketers who’ve done best in organic search over the past decade aren’t the ones who moved fastest on each update. They’re the ones who built content programmes grounded in genuine audience understanding and maintained them consistently. BERT didn’t change that. It confirmed it.

About the Author

Keith Lacy is a marketing strategist and former agency CEO with 20+ years of experience across agency leadership, performance marketing, and commercial strategy. He writes The Marketing Juice to cut through the noise and share what works.

Frequently Asked Questions