How to Export Google Search Console Data to JSON

Three ways to get Google Search Console data as JSON: the Search Analytics API, the BigQuery bulk export and an MCP server, with the limits that shape the file.

TL;DR

Google Search Console has no JSON export button. The Search Analytics API returns JSON natively, with 1,000 rows by default and a 25,000 row maximum per request. The BigQuery bulk export gives you a daily feed you can query, and an MCP server returns the same rows to an AI assistant. All three share Google’s 16-month window and anonymized queries.

You want your Search Console data in JSON because something downstream reads JSON: a script, a dashboard, a language model, or a data warehouse. The interface gives you CSV. The API gives you JSON, but only after you set up OAuth, handle pagination and respect quotas that are easy to miss.

This guide covers the three routes that work, the fields you get back, and the mistakes that make an export look complete when it is not.

There is no JSON export button in Search Console

The Performance report offers a CSV export, and some reports download a zip of CSVs. That is the end of the built-in options. Anything that produces JSON is calling the API on your behalf, or reading the BigQuery export.

That distinction matters when you compare files. A CSV export from the interface and a JSON response from the API can show different row counts for the same date range. The interface applies its own view, the API applies its own row cap, and both leave out anonymized queries. Neither file is “the complete data”. They are two views of a dataset Google limits on purpose.

Route 1: the Search Analytics API

The Search Analytics endpoint is the direct route. You send a POST request to the property’s endpoint, with a JSON body that describes the date range, the dimensions and the row limit.

curl -s -X POST \
  "https://searchconsole.googleapis.com/webmasters/v3/sites/sc-domain%3Aexample.com/searchAnalytics/query" \
  -H "Authorization: Bearer $ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "startDate": "2026-10-01",
    "endDate": "2026-10-31",
    "dimensions": ["query", "page"],
    "rowLimit": 25000,
    "startRow": 0,
    "type": "web",
    "dataState": "all"
  }'

The property identifier must be URL encoded. sc-domain:example.com becomes sc-domain%3Aexample.com, and a URL-prefix property keeps its trailing slash. If you get a 403, check the property string before you check the token.

The response is JSON with a rows array. Each row has a keys array that follows the order of your dimensions, plus four metrics:

{
  "rows": [
    {
      "keys": ["search console api", "https://example.com/blog/api-guide"],
      "clicks": 12,
      "impressions": 840,
      "ctr": 0.0143,
      "position": 18.4
    }
  ],
  "responseAggregationType": "byPage"
}

Note what is not there. There is no date field unless you ask for the date dimension. There is no query field unless you ask for query. The ctr is a fraction, not a percentage. The position is an average, so a row can sit at 8.5 even though no single impression happened at that position.

Pagination and the row limit

The default rowLimit is 1,000. The maximum is 25,000 per request. If your property has more rows than that for the date range and dimensions you chose, you need startRow:

  1. Send the first request with startRow: 0 and rowLimit: 25000.
  2. If the response has 25,000 rows, send the next request with startRow: 25000.
  3. Stop when a response returns fewer rows than you requested, or zero rows.
  4. Save each page before you request the next one.

Two details break naive loops. The API can return fewer rows than the property contains even when more data exists, because of internal limits and anonymization. And a paginated export is not a snapshot: if you run it across midnight, the last page may come from a slightly different dataset than the first. For monthly reporting, run the export after the data settles, or use dataState: "final".

The 16-month window applies here. A request for a date range older than that returns nothing, not an error you can fix. If you need a longer history, you have to store the data yourself from today, or start the BigQuery export and let it accumulate.

Quotas that shape the design

Google publishes quotas per site, per user and per project. The numbers that matter for a JSON export:

ResourceLimit
Search Analytics1,200 queries per minute per site
Search Analytics1,200 queries per minute per user
Search Analytics40,000 queries per minute per project
URL Inspection2,000 inspections per day per site
URL Inspection600 inspections per minute per site

There is also a load quota measured over 10 minutes and one day. Queries that group or filter by both page and query are the most expensive, and long date ranges cost more than short ones. If you hit quotaExceeded, wait 15 minutes before retrying. If a single query still fails after that wait, you are over the daily load quota, and the fix is to split the date range or remove a dimension.

For a normal site, the limits are generous. For a large site with millions of query and page combinations, the load quota is the real constraint, and a daily incremental export beats a monthly full one.

Route 2: the BigQuery bulk export

The bulk export writes your Search Console data to BigQuery every day. You enable it in Search Console settings and point it at a Google Cloud project and dataset. From there you query with SQL and export the result as JSON, CSV or whatever the next tool needs.

Three things to know before you enable it:

  • It starts on the day you turn it on. There is no backfill for the previous 16 months.
  • It is a daily feed, not a live feed. Yesterday’s data arrives later.
  • BigQuery storage and query costs apply, though the export itself is free.

The bulk export is the right route if you want a growing history beyond 16 months, if you need to join Search Console data with GA4 or CRM data, or if your site is large enough that API pagination is painful. It is the wrong route if you need one file today.

Route 3: an MCP server

An MCP server exposes the same API to an AI assistant as a set of tools. You ask a question in Claude, ChatGPT or Cursor, the server runs the query, and the model receives JSON rows it can reason over.

The advantage is that you skip the OAuth setup and the pagination code. The disadvantage is that the model can summarise the rows before you save them. Ask for the raw rows and a file path when you need an archive, not a paragraph.

A useful prompt is specific about all four parameters: property, date range, dimensions and row limit. “Export the last 28 days of queries and pages for gscmagic.com as JSON, 25,000 rows per page, and tell me if the result was truncated” gives the model a check it can perform. “Give me my Search Console data” does not.

GSC Magic’s analytics tool returns JSON rows and reports whether the result was truncated, so the assistant can tell you when a page cap shaped the answer. The Google Search Console API guide covers the underlying endpoints if you want to build the call yourself.

Which route fits which job

JobBest routeWhy
One monthly file for a reportAPIDirect JSON, no warehouse setup
History beyond 16 monthsBigQuery exportIt accumulates daily from the day you enable it
Ad hoc questions in an AI assistantMCP serverNo code, same rows, same limits
Joining search data with revenueBigQuery exportSQL joins in one place
A quick look at one queryInterfaceFaster than any export

Five mistakes that make a JSON export useless

  1. Summing query rows and calling it the property total. Anonymized queries and row caps make the sum smaller than the total. Use the total from a query without the query dimension.
  2. Forgetting the date dimension. Without date in the dimensions, the rows are aggregates for the whole range, and you cannot compare periods.
  3. Comparing a filtered export with an unfiltered one. A country filter changes both the rows and the totals. Keep the filter in the filename.
  4. Treating position as a rank. It is an average across impressions, weighted by where the page appeared. A page with one impression at position 3 and one at 90 has an average near 46.
  5. Storing only the flattened table. Keep the raw JSON. When a number looks wrong three months later, the raw response is the only way to check what Google actually returned.

When JSON is the wrong format

If the next stop is a spreadsheet, CSV is easier. If the next stop is a chart in a BI tool, the tool probably has a native connector. If the next stop is a single number for a client update, you do not need an export at all.

JSON earns its place when the data feeds code, a model or an archive. It preserves types and nesting, and it survives schema changes better than a hand-built CSV. For everything else, the simpler format wins.

Frequently asked questions

Is there a JSON export button in Google Search Console?

No. The interface exports CSV or a zip of CSV files, depending on the report. JSON comes from the Search Analytics API, from a tool that wraps it, or from a query you run on the BigQuery bulk export.

How many rows can one JSON request return?

The Search Analytics API returns 1,000 rows by default and accepts a rowLimit up to 25,000 per request. You page with startRow, and a response with fewer rows than you asked for usually means you reached the end.

Why is the JSON missing some queries?

Search Console withholds anonymized queries from query-level results, and row caps or filters can truncate the rest. The sum of your query rows will rarely equal the property total. Compare like for like, with the same dimensions and date range.

Can I get more than 16 months of data as JSON?

No. Search Console keeps about 16 months of performance data, and no export route can return days outside that window. The BigQuery bulk export starts collecting on the day you enable it, so it does not backfill older history either.

Do I need an API key for the Search Console API?

No. The API uses OAuth 2.0 access tokens, or a service account that has been added as a user on the property. There is no simple API key you can paste into a URL.

Can an MCP server return Search Console data as JSON?

Yes. An MCP server calls the same API and returns structured rows to the client. You still face the same row caps, quotas and 16-month window, because the limits belong to Google, not to the connector.

Do this with GSC Magic

GSC Magic returns Search Console rows as JSON to Claude, ChatGPT and Cursor, with the truncation flag included so you know when a row cap shaped the answer. Ask for a date range, dimensions and a limit, and save the result without writing OAuth code.

Start free or read how the Search Console API works before you build your own client.

, Founder of GSC Magic

Łukasz builds GSC Magic, the MCP server that gives AI agents access to Google Search Console, Bing Webmaster Tools and GA4. He writes about using Search Console data with Claude, ChatGPT and coding agents, and about the SEO workflows those tools make possible.

lukaszblania.comgithub.com/lumizone

Give your agent your search data.

GSC Magic connects Claude, ChatGPT and Cursor to Google Search Console, Bing and GA4. Free plan, 7-day Pro trial, no card.

Start free