Query Site Analytics

This article shows how to read a site's analytics with the Semantic Model API. A semantic model describes one analytics subject area (such as site traffic, revenue, etc.) and defines the measures, dimensions, and parameters you can query.

Prerequisites

1. The app/caller has the Site Analytics – read permission scope (SCOPE.DC-ANALYTICS-AND-REPORTS.READ-SITE-ANALYTICS).
2. A site context is available (the request is authorized against a specific site).

Required APIs

• Semantic Model API: REST
• Site Properties API (for the site's time zone): REST

Base path: https://www.wixapis.com/analytics/semantic-model/v3

Decision flow

Always follow List → Get → Query. You cannot construct a valid query without first discovering the model's field names from Get Semantic Model.

1. List Semantic Models — discover which subject areas exist and their IDs.
2. Get Semantic Model — inspect a model's measures, dimensions, and parameters to find the exact name values to query and their supported filters/sorting.
3. Query Semantic Model Data — request specific field names for a time interval, with optional filters, sorting, paging, and formatting.

Run each step as its own separate API call, and stop to read the result before starting the next step. The inputs to each step do not exist until the previous step returns:

• You cannot set semanticModelId until List returns it. Never fabricate or guess a model ID — it must be a GUID copied verbatim from a List Semantic Models result. A plausible-looking GUID that List did not return will fail (SEMANTIC_MODEL_NOT_FOUND).
• You cannot set fields, filters[].field, or sort.fieldName until you have read the schema from Get.

Choosing the model and the field names is a reasoning step you perform by reading the returned JSON — not something to automate with string matching. Do not chain List, Get, and Query into a single script/execution.

❌ Don't do this — one execution that lists, gets, and queries in a single pass, picking fields by pattern-matching their names:

This guesses field names, ignores each field's dependencies (so those fields come back silently empty), and never lets you actually verify the model or schema.

✅ Do this — three separate calls, reading each result before composing the next:

1. Call List. Read the returned models; pick the id whose subject area matches the request.
2. Call Get with that id. Read measures/dimensions/parameters; choose the exact names you need and note each field's dependencies.
3. Call Query with the field names you chose.

Before you begin (sharp edges)

• interval is required on every query. There is no way to query without a date range — omitting it fails.
• start/end are absolute UTC instants — not wall-clock in interval.timezone. The trailing Z is respected as a real point in time; interval.timezone does not reinterpret it, it aligns the range to local day boundaries (and materially changes the result — the same UTC window returned different totals under UTC vs Asia/Jerusalem vs America/Los_Angeles in testing). So to capture a local calendar range, send the UTC instants that equal local midnight in the site's time zone — not ...T00:00:00.000Z. Example — all of January 2026 for a America/New_York site (EST, UTC−5): start: "2026-01-01T05:00:00.000Z", end: "2026-02-01T05:00:00.000Z". Using ...T00:00:00.000Z would actually start the range at 7 pm on Dec 31 local and misalign every day boundary.
• interval is start-inclusive, end-exclusive ([start, end)). start is included, end is not — set end to the local-midnight instant of the day after the last day you want (e.g. for all of January, end is Feb 1 local midnight, in UTC).
• Convert local midnight → UTC per date, and watch DST. Compute each boundary as local-midnight-in-UTC using the site's offset for that specific date — the offset changes with daylight saving. America/New_York is UTC−4 in summer (June local midnight = 04:00:00.000Z) but UTC−5 in winter (January local midnight = 05:00:00.000Z), so a range spanning a DST switch has different offsets at its two ends. Don't hardcode one offset across a range.
• Set interval.timezone to the site's time zone to match the Wix dashboard. Analytics in the Wix business manager are bucketed by the site's time zone. If interval.timezone is omitted it defaults to UTC, so day boundaries shift and your numbers won't match what the owner sees in the dashboard (and the same goes for using a different time zone). Get the site's IANA time zone from properties.timeZone via Get Site Properties (GET https://www.wixapis.com/site-properties/v4/properties) and pass it through.
• Field names must come from Get Semantic Model. The fields, filters[].field, and sort.fieldName values must exactly match a name returned by the model schema (e.g. traffic.sessions_count). Do not guess field names.
• The field-name prefix is NOT the model slug. Do not build names as <slug>.<field>. A model's fields often use a different prefix — e.g. the model with slug crm-people-subscribers exposes its measure as people.contacts_count, not people_subscribers.contacts_count. Copy the exact name strings from Get Semantic Model; inferring the prefix from the slug produces a field the model doesn't have.
• Read the COMPLETE schema — never sample or truncate it. When inspecting a model in Step 2, read the full measures and dimensions lists; do not cap them with .slice(0, N) or otherwise return only the first few. Lists are often long and alphabetical, so a cutoff silently hides exactly the field you need (e.g. on traffic, a .slice(0, 25) drops traffic.referrer_category_name, traffic.referrer_source_name, and traffic.visitor_type). Choosing from a partial list re-introduces guessing — the agent assumes a missing field doesn't exist or fabricates a name. The same applies to a field's full dependencies array and to the model list from List Semantic Models.
• A wrong field name can fail loudly OR fail silently — assume neither. A fields/filters[].field/sort.fieldName value that isn't a valid model field either (a) rejects the query with a 4XX error and a self-explanatory string code (e.g. fieldIsInvalid), sometimes listing the model's available field names, or (b) is silently dropped — the query returns 200 and that field just doesn't appear in results[].fields (same as a missing dependency). In testing, an unknown field in fields was silently omitted, not errored. Never rely on an error to catch a bad field. (See Handling wrong fields below.)
• Field dependencies. A field returns data only if at least one of the field names in its dependencies array is also included in the same query; otherwise it's silently omitted from results (no error). For example, a measure may only return data when a specific dimension is also requested.
• Sorting a nullable measure — set sort.nullsLast: true. nullsLast defaults to false and only affects descending (DESC) order. If a measure can return null and you sort it DESC, the null rows sort first — so a "top N" query surfaces nulls before your real values. Set nullsLast: true to push nulls to the end.
• Result cap: 1,000 rows per query. results is capped at 1,000 rows — paginate with paging.offset for larger datasets.
• Formatting is opt-in. Set formattingEnabled: true to also receive a human-readable formattedValue per cell (e.g. 1500 → "$1,500.00" or 1.5K). Raw typed values are always returned.
• Totals are opt-in. Set totalsIncluded: true to get a totals row summing numeric fields across the full (unpaginated) result set.
• Unique fields are not additive. For unique measures (e.g. unique visitors), query the exact time range you want in a single request — never sum values from separate date-range queries. Uniques are deduplicated within each queried range, so adding per-range results double-counts anyone who appears in more than one range and overstates the true total.

Step 1: List semantic models

Response:

Pick the model whose subject area matches the request, and keep its id.

Step 2: Get the model schema

Response shape:

Each Field (in measures, dimensions, and parameters):

• Measures are quantitative fields you aggregate (revenue, page views, order count).
• Dimensions are categorical fields you group by (traffic source, country, product name).
• Parameters are optional inputs that customize query behavior (currency, date granularity).

Step 3: Query the model data

POST /semantic-models/query-data (body requires Content-Type: application/json).

Required body fields: semanticModelId, interval, fields.

Request fields

Response shape

Each cell in results[].fields is a typed value — one of numericValue, stringValue, booleanValue, timestampValue, arrayValue, or objectValue — plus formattedValue when formattingEnabled is true. totals is present only when totalsIncluded is true.

Time zone (match the Wix dashboard)

Analytics shown in the Wix business manager are aggregated by the site's time zone. To return numbers that match what the site owner sees, pass that time zone in interval.timezone on every query. When timezone is omitted, the API defaults to UTC — which shifts day boundaries and produces totals that don't line up with the dashboard (the same applies to any non-site time zone).

Two independent things both matter, and both use the time zone:

1. interval.timezone aligns aggregation to the site's local day boundaries. It is not cosmetic — the same start/end return different totals under different time zones.
2. start/end themselves are absolute UTC instants. To line them up with local calendar days you must send local-midnight-in-UTC — i.e. add the site's UTC offset for that date (DST-aware). For America/New_York: local midnight is 05:00:00.000Z in winter (UTC−5) and 04:00:00.000Z in summer (UTC−4). Sending ...T00:00:00.000Z with a non-UTC timezone starts the range partway through the previous local day and skews the result.

Get the site's time zone from Get Site Properties:

The IANA time zone string is returned in properties.timeZone:

Use that value as interval.timezone in Query Semantic Model Data. The Site Properties timeZone reflects the site's primary business address and requires the SITE_SETTINGS.VIEW permission.

Pagination

To retrieve more than 1,000 rows, page through with offset:

1. Request page 1 with paging: { limit: 1000, offset: 0 }.
2. Increment offset by the page size until pagingMetadata.count is less than the requested limit.

Handling wrong fields

A field can be "wrong" in three ways — guard against each:

1. Rejected (4XX). The query fails with a self-explanatory string code (e.g. fieldIsInvalid) and, when provided, the model's valid field names. Recovery: read them, pick the correct name, and re-query — do not retry the rejected name or invent a new one. If no field list is returned, fall back to Get Semantic Model (Step 2) and re-read the schema.

2. Silently omitted (200). The query succeeds but a requested field is absent from every results[].fields entry (and from totals). This happens for unknown field names and for fields whose dependencies weren't included — no error is raised. So after every query, verify that each name in fields actually appears in the results. If one is missing: re-check its exact name via Get Semantic Model, add any required dependencies, and re-query. Treat a silently-missing field as a failure, not as empty data.

3. Valid but semantically wrong. The name exists, so you get 200 and real-looking numbers — but it answers the wrong question (e.g. using traffic.page_url instead of traffic.page_url_from for per-page traffic, or a field whose description says "Do not use it ever"). This is the most dangerous case because nothing looks broken. Prevent it in Step 2: choose fields by reading each field's description, honor "do not use" notes, and never pick by name pattern alone. (This is why List → Get → Query are separate steps — see Decision flow.)

Cases 1 and 2 are caught after the query by validating the response; case 3 is prevented before the query by reading field descriptions when you choose fields.

Answer from the site's own models — don't ask or call external sources

Analytics questions are answered from the site's semantic models, discovered via List Semantic Models. This includes data that originates from a third party — e.g. Google Search / Search Console metrics (impressions, clicks, CTR, average position, top queries and pages) — which Wix ingests and often enriches with its own site data. So: do not ask the user whether a source is connected, and do not call the external provider (e.g. Google) directly — find the matching model and query it. If a query returns no data for the requested period, report that there's no data (the source may not be connected) rather than asking the user to connect first.

Common Errors

Error codes are self-explanatory strings (e.g. fieldIsInvalid) — read the code and the fields it returns rather than parsing a fixed body shape. Note: not every invalid field errors; some are silently omitted from a 200 response, so also validate the response (see Handling wrong fields).

Silent gap (no error): a requested field returns no data because none of its dependencies were included in the query — add a dependency field and re-query.

Best Practices

1. Always run List → Get → Query as three separate calls; read each result before composing the next, and never hardcode or pattern-match field names — read them from Get Semantic Model.
2. Pass the site's time zone (properties.timeZone from Get Site Properties) in interval.timezone and set start/end to local-midnight-in-UTC for that zone (DST-aware) so results match the Wix dashboard. start/end are absolute instants, not wall-clock.
3. Include a field's dependencies in the query, or expect that field to be silently dropped.
4. After each query, validate the response — confirm every requested field appears in results[].fields. A wrong field either errors (4XX, e.g. fieldIsInvalid) or is silently dropped from a 200; a missing field means an unknown name or a missing dependency. Fix and re-query rather than trusting the partial result. Choose fields by reading their description (honoring "do not use" notes), never by name pattern.
5. Use formattingEnabled: true for anything shown directly to a user; keep raw values for calculations.
6. Use totalsIncluded: true to get period totals alongside a paged breakdown in a single call.
7. Keep fields minimal (projection) and paginate large result sets with offset.

Related Documentation

• Semantic Model API: Introduction
• Semantic Model API: Sample Flows
• List Semantic Models
• Get Semantic Model
• Query Semantic Model Data
• Get Site Properties (source of the site's timeZone)