Sanity Best Practices

title: GROQ Query Maintenance & Best Practices description: Guidelines for GROQ queries, type safety, performance optimization, and syntax highlighting.

GROQ Query Maintenance & Best Practices

Constructing queries before execution

Start with the smallest query that returns the requested result. Use known document IDs, types, and fields, and use query parameters for filter values:

*[_id == $id][0]{_id, title, "launchCode": metadata.launchCode}

Before executing queries with reference lookups in filters, mixed OR conditions, computed sorts, or deep pagination, review Performance Rules. A small returned result does not establish that the query is cheap to execute.

Projection keys and attribute traversal

Plain fields can use {_id, title}. Give nested fields and computed expressions an explicit quoted output key:

*[_type == "article"][0...10]{
  _id,
  "displayName": title,
  "slug": slug.current,
  "launchCode": metadata["launchCode"]
}

{displayName: title} is invalid: quote the alias. {slug.current} and {metadata["launchCode"]} also need explicit keys. After a dot, use an attribute name, such as .title; do not write .(title).

Use functions that GROQ actually provides

Do not assume JavaScript or SQL functions exist in GROQ. Check the GROQ functions reference for unfamiliar operations.

• To join an array into a string, use array::join(tags, ", "). There is no built-in string::join().

• There is no built-in keys(), object::keys(), or array::keys(). To report the properties present in an object, retrieve that object and inspect the returned JSON outside GROQ. Do not retry by guessing another namespace. To learn declared fields instead, read the schema.

• For a conditional value, use select(featured => title, "Other"). A standalone featured => title is not a value expression. Conditional projection branches use objects, such as featured => {title}.

• To count matching documents, use count(*[_type == "article"]).

To inspect the fields present on a document, fetch it by ID:

*[_id == $id][0]

Read the property names from the returned object outside GROQ. This does not require a key-enumeration function in the query.

Check the complete expression

Before submission, check that filters and slices close with ], projections with }, and function calls with ). For example:

*[_type == "article" && featured == true][0...10]{title}

count(*[_type == "article" && featured == true])

After a syntax rejection, use the reported position and error to inspect the relevant expression, correct it, and retry. Do not resend the unchanged query or change a function namespace without checking that the replacement exists.

Use this contents list to jump to the query concern you need to solve.

Table of Contents

• Constructing queries before execution

• Query definition and imports

• Query fragments

• Expansion patterns

• Maintenance workflow

• Common patterns

• Performance rules

• API version best practices

1. Query Definition & Imports

The defineQuery Function

In application code, wrap GROQ queries in defineQuery for TypeGen support. The import location depends on your framework:

// Framework-agnostic (Angular, Remix, SvelteKit, Astro, vanilla)
import { defineQuery } from "groq";

// Next.js (re-exported for convenience)
import { defineQuery } from "next-sanity";

Syntax Highlighting

For VS Code syntax highlighting, either:

1. Use the groq tagged template (recommended): groq\...``

2. Or prefix with /* groq */ comment when using defineQuery

import { defineQuery } from "groq";

// ✅ Option A: groq tag (provides highlighting automatically)
import groq from "groq";
const QUERY = defineQuery(groq`*[_type == "post"]`);

// ✅ Option B: Comment prefix (for plain template literals)
const QUERY = defineQuery(/* groq */ `*[_type == "post"]`);

// ✅ Also valid: Just defineQuery (TypeGen works, but no editor highlighting)
const QUERY = defineQuery(`*[_type == "post"]`);

2. Query Fragments

Use string interpolation to reuse query logic and keep queries maintainable.

// src/sanity/fragments/image.ts
export const imageFragment = /* groq */ `
  asset->{
    _id,
    url,
    metadata { lqip, dimensions }
  },
  alt
`;

// src/sanity/queries/post.ts
import { defineQuery } from "groq";
import { imageFragment } from "../fragments/image";

export const POST_QUERY = defineQuery(/* groq */ `
  *[_type == "post"][0] {
    title,
    mainImage {
      ${imageFragment}
    }
  }
`);

3. Expansion Patterns (Page Builder)

When building a Page Builder query, expand all potential component types.

Best Practice: Use a pageFields fragment or similar strategy to keep the main query clean.

const pageBuilderExpansion = /* groq */ `
  pageBuilder[] {
    ...,
    _type == "hero" => {
      ...,
      cta[] { link, label }
    },
    _type == "gallery" => {
      images[] { ${imageFragment} }
    }
  }
`;

4. Maintenance Workflow

When you add a new field or component to the Schema:

1. Update the Query: Add the new field/expansion to the relevant GROQ query immediately.

2. Run TypeGen: If you have typegen.enabled: true in sanity.cli.ts, types regenerate automatically during sanity dev/sanity build. Otherwise, run npm run typegen manually.

3. Verify: Ensure the new field is available in the generated types.

5. Common Patterns

Ordering

// Single field
*[_type == "post"] | order(publishedAt desc)

// Multiple fields (tiebreaker)
*[_type == "post"] | order(featured desc, publishedAt desc)

// ⚠️ Order BEFORE slice, not after!
*[_type == "post"] | order(publishedAt desc)[0...10]  // ✅ Correct
*[_type == "post"][0...10] | order(publishedAt desc)  // ❌ Wrong order

Slice Notation

*[_type == "post"][0]       // Single document (object, not array)
*[_type == "post"][0...5]   // First 5 (exclusive) ← Most common

Slice bounds must be constant numbers — $params aren't allowed. For dynamic pagination, validate the numbers in application code and interpolate them directly into the query string:

const start = Number.isInteger(page) && page >= 0 ? page * pageSize : 0
const end = start + pageSize

const query = `*[_type == "post"] | order(publishedAt desc)[${start}...${end}]`

Default Values with coalesce()

*[_type == "page"]{
  "title": coalesce(seoTitle, title, "Untitled"),
  "image": coalesce(ogImage, mainImage, defaultImage)
}

Conditionals with select()

*[_type == "product"]{
  title,
  "badge": select(
    stock == 0 => "Out of Stock",
    stock < 5 => "Low Stock",
    "In Stock"
  )
}

Aggregation with count()

// Total count
count(*[_type == "post" && defined(slug.current)])

// Count per document
*[_type == "category"]{
  title,
  "postCount": count(*[_type == "post" && references(^._id)])
}

Reverse References

*[_type == "author"]{
  name,
  "posts": *[_type == "post" && references(^._id)]{ title, slug }
}

Array Filtering

*[_type == "movie"]{
  title,
  "mainCast": castMembers[role == "lead"]->{name}
}

// Match a known category ID (see Avoid Joins in Filters for dynamic lookups)
*[_type == "post" && $categoryId in categories[]._ref]

Special Variables

// ^ = parent document (in nested queries)
*[_type == "author"]{
  name,
  "posts": *[_type == "post" && author._ref == ^._id]
}

// @ = current item (in array operations)
*[_type == "post"]{
  "tagCount": count(tags[@ != null])
}

6. Performance Rules

Optimizable vs Non-Optimizable Filters

GROQ uses indexes for optimizable filters. Other expressions require evaluating candidate documents. An indexed _type constraint can narrow those candidates; without an indexed restriction, the query may scan the dataset.

Reduce the candidates for non-optimizable filters:

// Add selective indexed constraints; textual order does not force execution order
*[_type == "product" && defined(salePrice) && salePrice < displayPrice]

Avoid Joins in Filters

Reference resolution (->) in filters is expensive. Use _ref instead:

// ❌ Slow: Resolves reference for every document
*[_type == "post" && author->name == "Bob Woodward"]

// ✅ Fast: Direct _ref comparison
*[_type == "post" && author._ref == "author-bob-woodward-id"]

When you need dynamic lookups (don't know the IDs upfront), fetch all matching IDs and pass them as a parameter:

// Two-step approach:
// 1. Get every matching author ID
*[_type == "author" && name == $name]._id

// 2. Pass every returned ID as $authorIds in the main query
*[_type == "post" && author._ref in $authorIds]

Using [0] in the lookup would silently drop other authors with the same name. Discover the actual reference target types from the schema. Keep the same dataset, API version, and perspective in both requests. If the ID set needs pagination, collect all pages rather than silently truncating it; two requests can observe intervening content changes. A nested lookup is another option, but verify its timings rather than assuming it is faster.

Mixed OR Filters Can Still Require Joins

// Resolves brands while testing product candidates
*[_type == "product" && (
  name match $search || brand->name match $search
)]{_id, name, "brand": brand->name} | order(name asc)

The indexed name match branch does not make the whole OR optimizable: a product can qualify solely through its brand. Resolve matching brand IDs first, using the schema's target type (here brand):

*[_type == "brand" && name match $search]._id

Pass that complete result as $brandIds, keeping $search unchanged (for example, "Acme*"):

*[_type == "product" && (
  name match $search || brand._ref in $brandIds
)] | order(name asc) {
  _id,
  name,
  "brand": brand->name
}

An empty $brandIds array still permits direct product-name matches. Keep match semantics, both OR branches, ordering, and returned fields intact. Dereferencing in the projection is appropriate when the caller needs the related value; the expensive pattern here is dereferencing to decide which documents qualify.

Small Results Do Not Prove Cheap Execution

A final slice limits returned results, but may still leave substantial filtering or sorting work. count() can also be expensive when its filter requires joins or other per-document evaluation. Add pagination when the task permits it; do not add a limit that changes a request for all results.

Before claiming an improvement, compare results and timings on representative data with the same parameters, API version, and perspective. For a two-step rewrite, include both requests in the timing. Check multiple matching reference targets, no matching targets, and missing references. Local GROQ evaluation can check semantics, but cannot establish Content Lake performance.

Merge Repeated Reference Resolutions

Each -> is a subquery. Don't repeat it:

// ❌ Slow: Two separate subqueries
*[_type == "category"]{
  "parentTitle": parent->title,
  "parentSlug": parent->slug.current
}

// ✅ Fast: Single subquery, merged
*[_type == "category"]{
  ...(parent->{ "parentTitle": title, "parentSlug": slug.current })
}

Cursor-Based Pagination (Not Deep Slicing)

Deep slices are slow because all skipped docs must be sorted first.

// ❌ Slow: Must sort and skip 10,000 docs
*[_type == "article"] | order(_id)[10000...10020]

// ✅ Fast: Cursor-based, only fetches 20
*[_type == "article" && _id > $lastId] | order(_id)[0...20]

For custom sort orders, include the sort field in the cursor:

// Compound cursor: publishedAt + _id for deterministic pagination
*[_type == "article" && (
  publishedAt < $lastDate || 
  (publishedAt == $lastDate && _id > $lastId)
)] | order(publishedAt desc, _id)[0...20]

Always Project Fields

Always use projections to return only the fields your application needs. Fetching entire documents wastes bandwidth and processing time.

// ❌ Returns ALL fields including unused ones, metadata, revisions
*[_type == "post"]

// ✅ Only fetch what the component needs
*[_type == "post"]{
  _id,
  title,
  "slug": slug.current,
  publishedAt,
  excerpt
}

Apply projections at every level, including nested references:

*[_type == "post"]{
  title,
  author->{ name, "avatar": image.asset->url },
  categories[]->{ title, "slug": slug.current }
}

Use conditional projections for different contexts:

*[_type == "post"]{
  title,
  slug,
  // Only include body for single post view
  $includeBody == true => { body }
}

Don't Filter/Sort on Projected Values

Computed attributes can't use indexes:

// ❌ Not optimizable (computed attribute)
*[_type == "person"]{
  "fullName": firstName + " " + lastName
} | order(fullName)

// ✅ Optimizable (original attribute)
*[_type == "person"] | order(firstName, lastName)

Quick Checklist

7. API Version Best Practices

Always use dated versions (YYYY-MM-DD) for consistent behavior:

const client = createClient({
  apiVersion: '2026-02-01', // Use current date for new projects
})

• New projects: Use current date (e.g., 2026-02-01)

• Existing projects: Keep current version unless you need new features

• Dated versions lock behavior; v1 or vX may change unexpectedly