GAQL: Google Ads Query Language, Now Written by AI

GAQL — Google Ads Query Language — is the language every Google Ads API report is written in. One statement — SELECT fields FROM one resource, with optional WHERE, ORDER BY and LIMIT — describes exactly which columns, rows and date range come back. The Google Ads UI builds these queries behind its screens; the API makes you, or your AI, write them. And that is the honest 2026 update: you no longer need to write GAQL. You need to read enough of it to check what an AI wrote for you.

How a GAQL query is built

A representative query:

SELECT
  search_term_view.search_term,
  metrics.cost_micros,
  metrics.conversions
FROM search_term_view
WHERE segments.date DURING LAST_30_DAYS
ORDER BY metrics.cost_micros DESC
LIMIT 50

Four parts, four jobs. FROM names exactly one resource — the table-like view the query runs against: campaign, ad_group, keyword_view, search_term_view, change_event. SELECT names fields of that resource, plus metrics.* (the numbers) and segments.* (the slicers, such as date or device). WHERE filters — and carries the date range. ORDER BY and LIMIT shape the output. Selecting a segment splits rows by it: add segments.device and each campaign comes back as one row per device.

Money needs one reflex: the API returns cost in micros — millionths of the currency unit. A cost_micros of 12,500,000 is 12.50 in account currency. Every sane tool divides before showing you, and an undivided micros column is the first thing to suspect when a number looks absurdly large.

GAQL vs SQL: the differences that trip people up

• No JOINs. One resource per query; relationships are pre-baked into the resources themselves, which is why views like keyword_view exist at all.
• No GROUP BY. Aggregation is implicit in the resource; slicing happens by selecting segments, not by grouping.
• No SELECT-star. Every field is named explicitly — verbose, but self-documenting.
• Dates are special. DURING LAST_30_DAYS, or BETWEEN two ISO dates — the range lives in WHERE, and getting it wrong quietly answers a different question than you asked.
• Zeros need asking for. Whether entities with no impressions appear depends on the resource and filters — a "missing" keyword is usually a filtered keyword, not a deleted one.

Five useful queries, written as plain-English prompts

With a connected account, prompts do the writing — a Google Ads MCP server exposes the query tool and the AI fills in the GAQL:

1. "Search terms with spend and zero conversions, last 30 days, worst first" — the waste finder, straight into search terms report territory.
2. "Campaign spend, conversions and conversion value by week, last eight weeks" — the trend baseline.
3. "Keywords whose Quality Score components are below average, with their spend" — the diagnostic pull.
4. "Everything that changed in this account in the last 14 days" — change_event, the audit view.
5. "Conversion rate by device per campaign, this quarter" — the segment slice.

More where these came from in the prompt library.

Four checks on a query an AI wrote

An AI writes GAQL fluently and fast — and like any fluent writer, it can be confidently wrong in quiet ways. Four checks catch nearly everything:

• The date range. Is WHERE segments.date the period you asked about? A defaulted or shifted range is the commonest silent error.
• Micros. Confirm the division happened; a figure a million-fold off is an undivided cost_micros.
• The resource. Search terms come from search_term_view, not keyword_view — the answer changes with the FROM.
• Zeros and filters. Before believing an entity vanished, ask what filtered it.

The queries are no longer the skill. The verification reflex is — and it takes ten seconds per query, which is a fair price for acting on numbers you trust.