Query templates: authority attaches to the query format, not only to the topic.
A single query can exist in thousands of variations. Mapping that pattern is the second route to topical authority, alongside covering entities and their attributes.
Two routes to topical authority
There are two ways to build topical authority. The first covers every entity in a class through the attributes they share. The second covers every variation of a single query pattern. The strongest approach combines both.
The first route is the one described on our page about topical authority: entities of the same type share the same set of attributes. “Calories” is an attribute of every entity in the class “food”, just as “symptom” is an attribute of every entity in the class “disease”. Covering a whole class through its shared attributes signals completeness.
The second route does not require the pages to be topically related at all. Authority attaches to the format of the query, not to the subject. This is why one source can rank for unrelated topics if it consistently satisfies the same shape of question.
| Approach | What is covered | Where authority attaches | Risk |
|---|---|---|---|
| Entity–attribute | All members of a class, all shared attributes | To the topic | Depth without breadth of format |
| Query template | All variations of one query pattern | To the query format | Breadth without topical depth |
| Hybrid | Every entity in the class, across all attributes, in every variation of the pattern | To both | Demands the most discipline and data |
What is a query template?
A query template is a pattern in which one or more parts vary while the structure stays the same. For example taxi [city] fares or how much does [service] cost in [city].
The pattern usually breaks down into three positions: a contextual noun, a central entity and a synonym for the function or intent. Once those positions are listed, you have a map of every variation a single content system could cover.
| Component | Position | Variations |
|---|---|---|
| Central entity | Start of the query | taxi, cab |
| Contextual entity | Middle of the query | Belgrade, Niš, Novi Sad, Kragujevac, Šabac, Subotica… |
| Intent synonym | End of the query | fares, price, how much does it cost, tariff, calculator |
What matters is that the predicates attached to each pattern are not always the same. “Fares” naturally brings tariff, base fare, distance and waiting time. “How to book” brings phone, app and arrival time. A difference in predicates is the signal that these are separate pages rather than one.
When does a query deserve its own page?
The decision rests on four metrics: query demand, the presence of a different entity, low similarity to existing queries, and the existence of a pattern. When three of the four are satisfied, the variation deserves its own URL. When they are not, it belongs as a section on an existing page.
This is a more precise version of the rule “one dominant micro-intent per URL”. Without a threshold and a similarity criterion, that rule is not operational — under it, any nuance can be justified as a new page, which is exactly what bloats a site.
| Metric | Question | Example from the taxi industry |
|---|---|---|
| Demand | Is there enough search volume to justify maintaining a page? | “taxi Belgrade fares” has it; “taxi Belgrade fares on Sundays” does not |
| Different entity | Does the variation introduce an entity the existing page lacks? | Each city is a separate entity with its own tariffs |
| Low similarity | Is it distinct enough from queries you already cover? | “fares” and “price per kilometre” are too close to justify two pages |
| Pattern | Is there a pattern that repeats across variations? | “taxi [city] fares” exists for twenty-three cities |
The metrics are not weighed one at a time. A variation that clears demand but fails on similarity and entity stays a section, however often it is searched.
| Situation | Decision | Reasoning |
|---|---|---|
| Different predicates, same entity | New page | The intent differs and the content does not overlap |
| Similar predicates, different entity | New page | The data belongs to the entity and must not be mixed |
| Similar predicates, same entity | A section on an existing page | Two pages would cannibalise each other |
| Demand below the threshold | A sentence or a table row | A separate URL costs crawl budget and dilutes internal links |
The threshold is not a universal number. It is set per project, before writing, and recorded in the topical map together with the source of the demand estimate. A threshold decided after writing is not a threshold — it is a justification.
Query similarity is not string comparison
Two queries are not similar because they share most of their characters, but because they share the term carrying the greatest weight. Similarity is weighted by entity importance and demand, not by string overlap.
Research on end-to-end query term weighting shows that language models do not treat every word equally. In the query “Nike running shoes”, the term “Nike” carries more weight for relevance than “running” or “shoes”. That has a direct consequence for the page decision.
| Query pair | Heaviest term | Decision |
|---|---|---|
| taxi Belgrade fares / taxi Niš fares | the city name | Two pages — the heaviest term changes |
| taxi Belgrade fares / taxi Belgrade price per km | Belgrade | One page — the heaviest term is the same |
| taxi Belgrade / taxi Belgrade airport | Belgrade, then airport | Two pages — the second introduces a new high-weight entity |
The practical consequence: before deciding on a page, establish which term in the query carries the most weight. If that term changes between variations, they are probably separate pages. If it stays the same and only the modifiers change, they are probably sections.
The same applies to the document itself. If the heaviest term is a city name, it should also be the subject of the claims on the page — in the heading, in the opening sentence and in the structured data. A page whose heaviest term does not match the heaviest term of the query is solving a different problem from the one the reader has.
Limit: Google does not publish how it calculates term weight or where the similarity threshold sits. This is a model of reasoning derived from published research on term weighting and from the patent on detecting query-specific duplicate documents, not a documented algorithm.
Microsemantics: word order changes relevance
Two sentences can state the same fact and still differ in relevance for the same query. The difference comes from which term occupies the subject position.
Compare:
Financial independence is achieved by families with the help of financial advisors.
Financial advisors help families achieve financial independence.
Same facts, same entities. But if the query is “financial advisor”, the second sentence has that term as its subject and matches the intent better. If the query is “financial independence”, the first works better.
The rule that follows: the heaviest term of the query should sit in the subject position of the claim on your page. A claim is treated as a subject–predicate–object triple, and microsemantics is the alignment of that triple with the term weighting of the query.
| Target query | The sentence that fits better |
|---|---|
| taxi Belgrade price | “A taxi in Belgrade charges a base fare of 320 dinars and 105 dinars per kilometre.” |
| taxi price per kilometre | “The price per kilometre in Belgrade is 105 dinars by day and 135 at night.” |
| how much is a ride to the airport | “A ride to the airport of 18 kilometres costs about 2,210 dinars.” |
On a single page this looks like hair-splitting. On a pattern with three thousand variations, the same detail is multiplied three thousand times and stops being a detail.
The core and outer sections of a map
A topical map has two parts. The core is predominantly commercial and carries the most important entities. The outer section is predominantly informational and bridges the central entity to the rest of the map. Internal links always flow from the outer section towards the core.
This is a more precise division than the hub-and-spoke model, because it describes not only the direction of links but what each part is for. Every click and impression the outer section earns passes signal towards the commercial pages in the core.
| Part of the map | Role | Example |
|---|---|---|
| Core | Commercial, receives signal | City fare pages, city hubs, the calculator |
| Outer section | Informational, sends signal | Guides on how to book, what affects the price, how to recognise a reliable service |
One rule when linking: do not repeat the same anchor text more than three times in the main content. Anchor diversity preserves contextual uniqueness and stops the links from reading as templated.
A counterweight: dividing a map into core and outer does not mean it must have many pages. There are projects where an entire map fits on one URL and outperforms competitors using dozens of pages for the same coverage. Page count is a consequence of the threshold, not a goal.
Cost of retrieval as a frame
The cost of ranking a source must not exceed the cost of not ranking it. This is the frame that turns technical SEO and semantics into one problem rather than two.
A search engine does not select the best possible answer at any cost. It weighs quality against the cost of processing. When a source is expensive to handle — slow, unclearly segmented, full of URLs that contribute nothing — and the quality does not justify that cost, the simpler outcome is not to rank it.
The practical consequence is that every page without its own intent raises the cost of the whole site. Removing surplus URLs is therefore not just housekeeping; it increases the amount of signal per remaining document.
Within that logic, topical authority can be expressed as a ratio: ((historical data × topical coverage) ÷ cost of retrieval) × the right visual annotations, where visual semantics is a multiplier rather than an afterthought. The formula is not a unit of measurement but a way of seeing why coverage alone does not help when the cost of processing is too high, and why page layout is not a separate subject from architecture. As a multiplier, a wrong annotation cancels everything else: neither the lowest processing cost nor the broadest coverage helps if the page’s main content is not what the query asks for.
Paul Haahr of Google put it plainly at a conference: if a result is wrong for an individual query, they will not fix it; if that query belongs to a cluster, they will. That is the most direct confirmation that the work pays off at the level of the pattern, not the individual page.
Limit: “cost of retrieval” is an analytical frame derived from public patents, statements by Google engineers and industry analysis. Google does not publish a formula and does not confirm this name as one of its ranking systems. We use it as a way of thinking, not as a documented fact.
What our two projects show
We run two projects in the same industry using the same methodology, but with a different number of pattern variations. The comparison is useful precisely because the outcome is not the same.
| Project | Variations of “[city] fares” | AI share, fare pages | AI share, whole site |
|---|---|---|---|
| taxi.co.rs | 22 fare pages with impressions (of 23 cities) | 30.9% | 14.2% |
| leskovac.taxi | 1 city | 6.9% | 5.2% |
On both sites, pages carrying a concrete fare outperform the average of their own site. But the ratio to that average is 2.2× on the site with 22 variations and 1.3× on the site with one. This is consistent with the idea that variations of the same pattern reinforce one another — but it is not proof. See the control case for the full list of limitations, including a second explanation we consider stronger.
How this is applied
- List the positions in the pattern. Contextual entity, central entity, intent synonym. Write down every value that genuinely exists in your industry.
- Set the demand threshold before writing, and record the source of the estimate.
- Check the predicates per variation. If the predicates differ, that is a new page. If they match and the entity is the same, that is a section.
- Link the variations towards the commercial core, so informational pages pass signal.
- Remove URLs that fail the threshold and redirect them to the page that covers their intent.
- Measure by group of pages, not by individual URL. The pattern only becomes visible at group level.
Frequently asked questions
Does a query template replace a topical map?
No. They are two complementary routes. A topical map covers the entities and attributes of one subject; a query template covers the variations of one query pattern. The strongest system uses both at once.
What demand threshold should I set?
There is no universal number. It depends on the industry, the competition and your capacity to maintain the page. What matters is that the threshold is defined before writing and recorded together with the source of the estimate.
Do more pages always mean more authority?
No. A page without its own intent consumes crawl budget and dilutes internal links, reducing the signal per remaining document. Ten complete pages often outperform a hundred templated ones.
Is “cost of retrieval” an official Google term?
It is not. It is an analytical frame derived from patents, statements by Google engineers and industry analysis. Google does not publish a formula and does not use that name for any of its systems.
How many metrics must be met for a query to get its own page?
Three of the four: demand, a different entity, low similarity and the existence of a pattern. A variation that clears only demand, while failing on similarity and entity, stays a section on an existing page.
How is the similarity of two queries calculated?
Not by string comparison but by term weight. If the heaviest term changes between variations, they are probably separate pages; if it stays the same, they are probably sections.