How to Use n8n with ._Template 184 Automacao Google Ads Relatorio Ia

If you run Google Ads for clients or your own product, you already know where the hours go: not in optimizing bids, but in pulling numbers into a spreadsheet, writing the same "here's what happened this week" summary, and formatting it so a non-technical stakeholder actually reads it. Template 184 — Automação Google Ads Relatório IA — is an n8n workflow that removes that entire task. It pulls your campaign metrics on a schedule, hands them to an AI model that writes the narrative, and delivers a finished report to email, Slack, or a doc. This guide shows exactly how it works and how to run it.

The problem: reporting is manual, repetitive, and late

Google Ads reporting fails in three predictable ways. First, it's manual — someone opens the Google Ads UI, exports CSVs, and copy-pastes into a template every Monday. Second, it's inconsistent — the person writing the summary this week frames the data differently than last week, so trends get lost. Third, it's late — by the time the report reaches the client, the spend decisions it should have informed are already made.

The raw data isn't the bottleneck. Google Ads exposes everything through its API: impressions, clicks, CTR, CPC, conversions, cost per conversion, and campaign-level breakdowns. The bottleneck is the glue — the repetitive work of fetching that data on a schedule and turning rows of numbers into a paragraph a human can act on. That glue is exactly what n8n plus an LLM automates.

The solution: a scheduled n8n pipeline that writes itself

Template 184 chains four responsibilities into one workflow: trigger on a schedule → fetch Google Ads metrics → generate an AI narrative → deliver the report. Each stage is a node you can inspect and edit, so nothing is a black box.

The AI step is what makes this different from a plain data export. Instead of dumping a table, the workflow feeds the metrics to a language model with a prompt like "You are a Google Ads analyst. Summarize this week's performance, flag any campaign where cost-per-conversion rose more than 20%, and recommend one action." The model returns a report that reads like an analyst wrote it — because, functionally, one did. The founder or ops lead gets a decision-ready summary, not homework.

Step-by-step: building the workflow in n8n

Here is the node-by-node structure. If you're importing Template 184, these are the pieces you'll be configuring; if you're building from scratch, follow the same order.

1. Schedule Trigger node. Set the interval to how often you report — Weeks with a Trigger at Day = Monday and Trigger at Hour = 8 is the common choice. This replaces the human who "remembers" to run reports.

2. Google Ads request. You have two options. The cleanest is the native Google Ads node using an OAuth2 credential; select the operation to run a report and query. If your n8n version predates full Google Ads coverage, use an HTTP Request node pointed at https://googleads.googleapis.com/v17/customers/{customerId}/googleAds:searchStream, with a POST body containing a GAQL query. A typical query:

SELECT campaign.name, metrics.impressions, metrics.clicks, metrics.cost_micros, metrics.conversions, metrics.average_cpc FROM campaign WHERE segments.date DURING LAST_7_DAYS

Remember that Google Ads returns cost in micros (1,000,000 = one currency unit), so divide by 1,000,000 downstream.

3. Set / Code node to shape the data. Drop a Code node in to normalize the response: convert cost_micros to dollars, compute CTR (clicks / impressions) and cost-per-conversion, and build a compact JSON array. Keeping this transformation explicit means the AI prompt receives clean, labeled numbers instead of raw API noise — which sharply improves the quality of the summary.

4. AI node for the narrative. Use the OpenAI / LLM node (or the built-in AI Agent node). Set the system message to define the analyst persona and the exact report structure you want — headline, per-campaign notes, one recommended action. Pass the cleaned metrics JSON in the user message via an expression like {{ $json.campaigns }}. Set temperature low (0.2–0.4) so the tone stays factual and repeatable across weeks.

5. Delivery node. Route the generated text to wherever your audience lives: the Gmail / Send Email node for clients, the Slack node for internal ops channels, or the Google Docs / Notion node to append to a running report log. Many teams wire two delivery nodes in parallel — email for the client, Slack for the team.

Once connected, run the workflow manually once (click Execute Workflow) to verify the full chain, then activate it so the Schedule Trigger takes over.

Benefits: what you actually get back

Time. A report that took 45–90 minutes to assemble now costs zero human minutes. For an agency managing ten accounts, that's a full workday reclaimed every week.

Consistency. Because the AI works from a fixed prompt and a fixed data shape, every report follows the same structure. Stakeholders learn where to look, and week-over-week trends become obvious instead of buried.

Speed to insight. Reports arrive at 8 a.m. Monday, automatically, before anyone asks. The recommendation the AI flags — "pause Campaign X, cost-per-conversion up 34%" — reaches a decision-maker while it still matters.

Scale. Duplicate the workflow per account, or loop over multiple customer IDs in a single run. The marginal cost of the eleventh report is the same as the first: nothing.

Common pitfalls and how to avoid them

OAuth and the developer token. The Google Ads API needs more than a standard OAuth credential — it requires an approved developer token and, for the HTTP approach, a login-customer-id header when accessing accounts under a manager (MCC). A "PERMISSION_DENIED" error almost always means the token isn't approved or the header is missing, not that your query is wrong.

Micros and null conversions. Forgetting to divide cost_micros produces reports claiming you spent millions. And campaigns with zero conversions return null, which breaks cost-per-conversion math — guard for it in the Code node (conversions > 0 ? cost / conversions : 0).

Feeding the AI raw data. If you skip the shaping step and pass the full API response to the LLM, you'll get vaguer summaries and burn far more tokens. Clean, labeled input is the single biggest lever on output quality.

AI hallucination on numbers. Instruct the model explicitly to only use figures present in the input and never invent metrics. Keeping temperature low and including the exact numbers in the prompt keeps the narrative anchored to reality.

Silent failures. An API rate limit or expired token can make the workflow fail quietly. Add an Error Trigger workflow or a fallback Slack alert so a broken report notifies you instead of just not arriving.

Template 184 turns Google Ads reporting from a weekly chore into an invisible background process. Configure it once, and the reports write and deliver themselves — leaving you to act on the insights instead of assembling them.