Brand promotion: this article introduces Guangsuan's Google Ads management service for B2B teams.
A B2B paid-media owner reaches month-end with one task: hand finance a Google Ads file that can sit next to CRM records. The file opens fine, yet spend and qualified leads still do not line up. In most cases the mismatch was fixed before download, when the team chose a format, a row shape, and a date rule without writing them down. This article answers one practical question: how do you get JSON you can actually reconcile?
Start from the month-end question
Write one sentence before opening Google Ads. Example: which campaigns spent money in October for sales-accepted demo requests, and how much did each spend? That sentence decides the period, the row detail, and which conversion action travels forward.
A total by campaign answers spend. A split by day or device answers timing. Search terms placed beside the bought keyword answer intent. These are different row shapes. Pick the shape the question needs instead of pulling every column at once.
Report downloads do not produce JSON
Google Ads does not provide JSON in its report downloads. The download menu lists spreadsheet and document formats plus XML and Google Sheets, and XML is not JSON. XML serves readers built for XML; it will not satisfy a pipeline waiting for JSON.
You may convert a CSV or XML download into JSON on your own computer. Treat that script as part of reporting: keep it versioned, name an owner, and test it on a small slice you have checked by hand. For field background behind this guide, see Guangsuan's guide to exporting Google Ads data to JSON.
What each JSON path is for
There are two realistic paths when JSON is required, and they fit different jobs.
For automation, call the REST form of the Google Ads API. The API supports gRPC and REST interfaces, and the REST interface returns JSON representations of the same resources. You write the request in GAQL, and the service returns only the fields named in the query, such as campaign number and name with clicks, conversions, and cost in micros. Access requires a developer token and an authenticated client, with tiered daily limits beginning with test-account use. Keep tokens in a secret store.
For repeated CRM matching, stage the data first and emit JSON downstream. A BigQuery transfer or reporting connector can hold history and produce JSON for the next join. That adds a project to maintain, so it suits a monthly routine better than a single file. For a one-off request, local CSV-to-JSON conversion is usually faster.
A local fixture covers a third need: shape before access. The fixture mirrors GAQL field names and prints valid JSON, so a developer can build parsing before credentials exist. The numbers inside are invented teaching data, not account data:
import json
rows = [{"campaign.id": "550201", "campaign.name": "Demo-Brand", "metrics.clicks": 180, "metrics.conversions": 7, "metrics.cost_micros": 12400000}]
print(json.dumps(rows, indent=2))Fields and date rules to write down first
Keep the customer number in the file. Keep campaign number next to campaign name because names are edited while numbers remain stable. Keep each split that makes a row unique, such as date, device, and conversion action where used. A campaign total and a campaign-by-day file cannot share one join key. Export the grain you will actually analyse, keep the raw download unchanged, and work in a copy.
Exports are dated in the account time zone, which may not match your calendar or your CRM timestamps. A record saved late on the last evening can belong to different months on each side without either side being wrong. Write the zone on the file. Use the segment control to divide by day, week, or month rather than stitching separate pulls together.
Google Ads can report the same conversion under different timing rules. One rule places the event where it happened; the other places it where the click happened. A September click that becomes an October signup therefore falls in different months. Neither rule is wrong, but they must not be mixed in one comparison. Before building the view, note the basis, which actions are primary, the counting method, and the attribution model. Only primary actions feed the Conversions column used for bidding; secondary micro-steps stay separate. Cost arrives in micros in account currency, so handle that unit once in a noted place.
When CRM disagrees, inspect the file before moving bids
Limit the first check to one primary action, one campaign group, and one closed period. Align basis, range, and zone on both extracts. Keep ad-sourced records apart from organic and referral entries. Review counting behavior and repeated uploads before calling a remainder unexplained. Leave any leftover gap marked as open instead of forcing agreement.
As a hypothetical teaching illustration with invented figures, not a client outcome: a file holds 39 ad-tracked contacts in a month; sales accepts 13 as qualified under a written definition, 9 dated inside that month and 4 dated in the next month, with 2 reaching order. Each number describes a separate stage.
The export only shows what Google was able to record; sales-confirmed numbers live in your own system. When few contacts meet the written qualified mark, review query intent and landing-page fit before adjusting bids. A clean structure helps: clear names, one intent per campaign, and tracking that matches how leads truly arrive.
Where Guangsuan fits
Guangsuan's Google Ads management service covers account structure, search terms, landing pages, conversion checks and reporting, with scope and fees set in a contract and no fixed enquiry promise. Details are on the Guangsuan Google Ads management service page. For a first review, bring the action list with primary flags plus one annotated extract that shows basis and zone.
FAQ
Can Google Ads download JSON directly? No. Use the REST API path or a warehouse path.
Does a file that parses as JSON reconcile cleanly? No. Parsing says nothing about row grain, units, date rule, or which actions were summed.
Should a missing term be read as zero? Not yet. Very quiet rows can drop from a filtered view, so confirm filters and grain first.