Most SEO keyword targeting operates on a false premise: that the keyword a user types is the query Google actually ranks content against. It isn’t. There’s a rewriting layer between user input and ranking, and if your optimization doesn’t account for it, you’re solving the wrong problem.
Google has published research on query rewriting, synonym expansion, and query reformulation going back well over a decade. The practical implications for on-page SEO are underappreciated — and they’ve gotten more consequential as neural matching and MUM-influenced understanding have been folded into core ranking.
What the Query Interpretation Layer Actually Does
Before Google scores a single document, it runs the raw query through several transformation steps. They run in parallel, feed into each other, and the results get blended — but conceptually there are four things happening:
- Synonym expansion: “Cheap flights NYC to London” may be interpreted as also covering “affordable airfare New York Heathrow” and dozens of other variants simultaneously. Google’s research on synonym systems shows this isn’t simple one-to-one lookup — it’s embedding-space proximity, which means contextually similar terms get pulled in even when they share no root word.
- Entity resolution: Ambiguous terms get resolved to specific entities in the Knowledge Graph. “Jaguar” becomes either the car brand or the animal based on surrounding query context. Your page gets scored against the resolved entity, not the raw string.
- Intent classification: The query gets bucketed — informational, navigational, transactional, or local — and that bucket determines which ranking signals get weighted more heavily. A transactional bucket weights conversion-signal pages differently than an informational one.
- Query reformulation: In some cases, Google rewrites the query outright. If the original query produces low-confidence results, it relaxes or tightens constraints — sometimes silently, sometimes surfaced as “results for [X]” suggestions.
Your content gets evaluated against the post-interpretation query, not the pre-interpretation one. That’s the gap most keyword optimization ignores.
Why Exact-Match Keyword Targeting Is Less Useful Than It Looks
Exact keyword match in your title tag, H1, and body copy still matters — but its marginal value has eroded in a specific way.
If Google’s interpretation layer expands your target query to include 40 semantically related variants, and your page only optimizes for the surface form of one of them, you’re leaving the other 39 unaddressed. A competitor page that never uses your exact keyword phrase but covers the concept space more completely will often outrank you on the rewritten query — which is the query that actually determines ranking.
This is different from “write for users not search engines” advice, which is vague to the point of being useless. The concrete mechanism is: Google’s query-side representation is richer than your keyword, so your document-side representation needs to be richer too. Matching the surface string is necessary but not sufficient.
Where this plays out most clearly is in informational queries with strong entity components. A page optimized for “how to treat iron deficiency” that also covers ferritin levels, hepcidin, dietary absorption inhibitors, and the distinction between iron deficiency and iron deficiency anemia will score higher on the rewritten query than a page that has the exact phrase stuffed into every other paragraph but treats the topic at surface depth. The interpretation layer is looking for conceptual coverage, not keyword repetition.
Entity Resolution Is the Part Most Sites Get Wrong
When Google resolves an ambiguous query to a Knowledge Graph entity, it’s not just deciding what the query means — it’s also deciding which entities are relevant to the answer. That’s where your entity definition problem becomes a ranking problem.
If your brand or content isn’t clearly associated with the resolved entity in Google’s entity graph, your page may not even enter the candidate set for ranking on that query variant, regardless of your keyword targeting. You’re not being outranked — you’re being excluded upstream, before scoring even begins.
We’ve run into this when working on sites with ambiguous brand names — names that share string overlap with other entities in the Knowledge Graph. The disambiguation step resolves against whichever entity has stronger graph signal, and if that’s not you, you lose before the race starts. Fixing it requires building entity associations externally (structured mentions on authoritative third-party pages, schema that explicitly references your Knowledge Graph ID) — not better on-page optimization.
How to Optimize for the Rewritten Query, Not Just the Raw Keyword
The practical shift isn’t huge, but it is different from standard keyword-centric SEO.
1. Map the semantic neighborhood, not just the keyword
Before writing, use the top-10 results for your target query to identify which related concepts, entities, and sub-topics appear consistently across ranking pages. These aren’t just “related keywords” — they’re signals about what the interpretation layer considers part of the query’s concept space. If seven of the top ten results cover sub-topic X and your page doesn’t, you’re missing something the rewritten query is asking for.
SERP-based content gap analysis helps here, but the underlying logic is simpler: you’re reverse-engineering what Google’s query interpretation decided the query is actually about.
2. Use schema to anchor your entity associations explicitly
Schema markup — @type, sameAs pointing to Wikidata or Wikipedia, about relationships — tells the interpretation layer which entities your content is associated with before it has to infer from prose. This is one of the few places schema actually influences ranking upstream, not just rich snippet eligibility.
Getting your sameAs references right matters. Pointing to a Wikipedia disambiguation page instead of the specific entity page is a common error that weakens the signal.
3. Separate your keyword from your intent target
Know which intent bucket your query resolves to, and optimize the page’s conversion architecture for that bucket, not just its content. An informational-bucket page with heavy transactional CTAs creates a mismatch that shows up in user signals, which feeds back into ranking. The interpretation layer’s intent classification should inform your page design decisions, not just your content decisions.
4. Optimize for the query Google infers, not the query you want
Check what Google actually returns for your target query before you write. If the SERP is showing results that look different from what you’re planning to produce — different format, different depth, different entity focus — that’s the interpretation layer telling you what it thinks the query means. You can fight that interpretation, but rarely successfully. Better to understand it first and decide whether this is even the right query to target.
One Genuine Caveat
Query interpretation behavior isn’t uniform across query types. Head terms with high commercial value tend to have more aggressive synonym expansion and entity resolution applied. Long-tail queries with very specific phrasing often get interpreted more literally. Optimizing a long-tail informational post doesn’t require the same semantic breadth investment as a high-competition head term — and treating it like it does wastes effort. Know which you’re dealing with before you decide how far to expand your coverage.
Also: keyword presence still matters. It gets you into consideration; conceptual coverage and entity alignment determine whether you actually rank. Neither replaces the other.
For your most important target queries — do you actually know what Google’s interpretation layer is resolving them to? Pull the SERP, map the entities that appear across the top results, and compare that set against what your page explicitly covers. The gap there is almost always more diagnostic than your keyword density report.