Search for “why does my TV look strange?” and Google can return pages about the “soap opera effect”, even though your query never used those words. That leap from words to concepts is the job of a group of AI systems Google lists in its ranking systems guide, and neural matching is one of the most important.
Neural matching is often confused with RankBrain and BERT. All three help Google understand language, yet Google describes them differently and they arrived at different times. This article explains each one using Google's own descriptions and contemporary coverage, then turns that into practical advice for keyword-focused content in 2026.
Key Takeaways
- Neural matching is an AI system Google uses to understand representations of concepts in queries and pages and match them to each other.
- Google's Danny Sullivan discussed it publicly in September 2018, describing it as “super synonyms” and saying it affected 30% of queries at the time (per Search Engine Journal).
- RankBrain (revealed October 2015) helps relate words to concepts; BERT (from October 2019) understands how word combinations change meaning.
- None of these systems can be “optimised for” directly. They reward clear, complete coverage of a topic rather than repeated exact-match keywords.
- Neural matching has no single launch date in Google's documentation and is not a named update.
What Google says neural matching is
“Neural matching is an AI system that Google uses to understand representations of concepts in queries and pages”
— Google Search Central, A guide to Google Search ranking systems
In other words, rather than looking only for the same words, Google represents the ideas in a query and in a page, then checks how well those ideas match. When Google first talked about it, Search Engine Journal reported Danny Sullivan calling it an “AI method to better connect words to concepts” and “super synonyms, in a way”, affecting 30% of queries. That figure dates from September 2018; Google has not published an updated one.
Neural matching vs RankBrain vs BERT
In March 2019, Sullivan explained the difference between neural matching and RankBrain in a series of posts from Google's Search Liaison account. Search Engine Journal summarised it as RankBrain helping Google relate pages to concepts, and neural matching helping relate words to searches. Google's current guide describes them slightly differently, which is a reminder that these are evolving systems, not fixed algorithms.
| System | First public mention | Google's current description (paraphrased) | Where it helps most |
|---|---|---|---|
| RankBrain | October 2015 (Bloomberg interview) | Helps understand how words relate to concepts, so pages can rank without the exact query words | Unfamiliar or new queries |
| Neural matching | September 2018 (Google 20th anniversary event) | Understands representations of concepts in queries and pages and matches them | Vague or descriptive queries |
| BERT | 25 October 2019 (US English) | Understands how combinations of words express different meanings and intent | Queries where small words (to, for, not) change meaning |
| Passage ranking | Live February 2021 (US English) | Identifies individual sections of a page to judge relevance | Long pages that answer a niche question in one section |
Google's ranking systems guide quotes RankBrain's role directly:
“RankBrain is an AI system that helps us understand how words are related to concepts.”
— Google Search Central, A guide to Google Search ranking systems
For more history see our RankBrain explainer and the BERT update article. All of these descend from the shift Google made with Hummingbird in 2013 towards understanding meaning rather than matching strings.
Why neural matching is not an “update”
You will not find neural matching on the Search Status Dashboard. Like most systems in the guide, it runs continuously and is improved over time. If your rankings changed and someone says “neural matching did it”, ask for evidence — the change is far more likely to relate to a confirmed core update or to competitors' content. Our ranking systems guide explainer lists which systems have had named updates and which have not.
What it means for keyword-focused content
Neural matching changes the economics of keyword targeting. Exact-match repetition adds little, because Google can already connect “TV looks strange” to “soap opera effect”. What helps is covering the underlying concept thoroughly and naturally:
- Use the vocabulary real people use, including the problem description and the technical term.
- Answer the follow-up questions a searcher would reasonably have.
- Group closely related keyword variants on one page instead of making near-duplicate pages for each.
- Use headings that describe concepts clearly so sections can be understood on their own.
Our semantic SEO guide and entity SEO guide go deeper on structuring topics this way, and search intent shifts during core updates explains why the dominant meaning of a query can change.
A step-by-step process for concept-led pages
- Collect the query family. Gather the head term, synonyms, problem-style phrasings and questions from Search Console and keyword tools.
- Cluster by intent, not wording. If two phrasings return the same results, they usually belong on one page.
- Map concepts. List the sub-topics, entities and terms an expert would expect to see.
- Write for the reader. Explain the concept plainly first, then go deeper.
- Check coverage gaps against the pages that currently rank — not to copy them, but to see what questions they answer.
- Measure by query group, not a single keyword, in Search Console.
Old vs new keyword habits
| Old habit | Why it no longer helps | Better approach |
|---|---|---|
| One page per keyword variant | Google connects variants to the same concept; near-duplicates compete | One strong page per intent |
| Fixed keyword density targets | Concept matching does not need repetition | Natural language and complete coverage |
| Ignoring problem-style queries | Searchers describe symptoms, not terms | Include both the symptom and the technical name |
| Exact-match anchor text everywhere | Looks manipulative and adds little | Descriptive, varied anchors |
Common mistakes
- Claiming a tool can “optimise for neural matching”. Google has published no such signals.
- Quoting the 2018 “30% of queries” figure as current.
- Confusing neural matching with MUM. Google says MUM is not used for general ranking.
- Stuffing synonyms into a page to “help” the system.
- Creating separate thin pages for every phrasing of the same question.
Related Guides
- Google's Original Content Systems and the 2019 Original Reporting Change
- Google's Site Diversity System: Why One Site Rarely Gets More Than Two Results
- PageRank and Link Analysis After a Decade of Link Updates
Frequently Asked Questions
Is neural matching the same as RankBrain?
No. Both relate words to concepts, but Google lists them as separate systems. In 2019 Google described RankBrain as relating pages to concepts and neural matching as relating words to searches.
When did Google start using neural matching?
Google discussed it publicly in September 2018 and said it had been in use for some months before that.
Can I optimise for neural matching?
Not directly. Write clearly about the full concept and searcher intent; that is what the system is designed to understand.
Is neural matching a ranking factor?
Google lists it as a ranking system, not a factor you can score well on. It helps Google decide which pages are relevant to a query in the first place.
Do keywords still matter?
Yes. Words remain the clearest signal of what a page is about. Repetition just matters less than clarity and completeness.
Conclusion
Neural matching means Google reads for meaning. Content that explains a concept clearly — in the words real people use — benefits from it; content built around keyword repetition gains nothing. If you want your content planned around concepts and intent, our content writing service can help, or request a quote.
