AI-Driven Google Ads Analysis Workflow

Overview

A structured workflow for using AI tools — primarily Gemini and Claude — to analyze Google Ads campaign and search term reports, generate actionable recommendations, and track performance over time. This process was demonstrated live during a weekly performance review using the Bluepoint ATM account as a test case, revealing critical campaign flaws that manual review had missed.

This workflow is intended to be run weekly by account managers and the performance marketing team as a complement to (not a replacement for) hands-on campaign management.

Why Gemini and Claude Instead of ChatGPT

Different AI tools have different strengths:

ChatGPT sometimes fails to open URLs depending on a site's robots.txt settings. Gemini is more reliable for file uploads and link-based research. Claude tends to produce higher-quality marketing reasoning.

"When I really want to do internet stuff, like where I'm going to look at links or scrape something, I use Gemini. And if you're asking marketing questions, oftentimes Claude's better." — Mark Hope

Step-by-Step Workflow

1. Start a Dedicated Chat

Create a new chat for each client or topic. Do not mix accounts or unrelated campaigns in the same conversation.

• Keeping chats focused prevents the AI from conflating context across unrelated accounts.
• A focused chat is also easier to search and return to later.

Opening prompt pattern:

"I want to talk about Google Ads for [Client Name]. I'm going to upload some documents. Wait for my questions."

This prevents the AI from generating premature analysis before all data is loaded.

2. Download and Upload Reports

From Google Ads, download the relevant reports as CSV files for the desired date range (typically last 30 days):

• Campaign Report — overall spend, impressions, clicks, conversions, impression share
• Search Term Report — actual queries triggering ads, especially useful for identifying PMax waste

Drag and drop the CSV files directly into the Gemini or Claude chat window. You can also paste a screenshot using a screen snipping tool if a quick visual check is sufficient.

3. Provide Campaign Context

AI tools do not inherently know the business model behind a campaign. You must tell it:

• Campaign type: B2B lead gen, B2C e-commerce, brand awareness, etc.
• Known issues or recent changes: e.g., "We just switched the landing page from the homepage to a dedicated landing page."
• What to disregard: e.g., for B2B lead gen, instruct it to ignore conversion value (which is set to $1 as a placeholder) and focus on CPA and lead volume instead.

Example clarifying prompt:

"Disregard conversion value. This is a B2B lead gen campaign."

Without this context, the AI will flag a $1 conversion value as a critical error — which is technically correct but misleading for lead gen accounts.

4. Ask Targeted Questions

Once reports are uploaded and context is set, ask specific questions rather than requesting a generic summary:

• "Evaluate the impression share loss. What's causing it?"
• "Which search terms in the PMax campaign are irrelevant and should be negated?"
• "The cashless campaign has a 61% conversion rate. Should we increase its budget?"
• "What if I increased the bid from $1 to $3? What would you expect to happen?"
• "We changed the landing page last week. Are you seeing any trends that suggest it's improving?"

The AI can also be used to compare periods, predict outcomes, and prioritize action items.

5. Maintain Chat History for Weekly Tracking

Return to the same chat each week. The accumulated context makes the AI progressively more useful:

• It can compare this week's data to last week's.
• It can confirm whether previously recommended changes had the expected effect.
• It builds a running record of decisions and outcomes.

Weekly return prompt pattern:

"It's been a week. Here are the updated reports. Last week you recommended [X]. Can you tell if it worked?"

What the AI Can Catch

The Bluepoint ATM demonstration surfaced three issues that had not been explicitly flagged before:

See ^[1] for the full account context and remediation actions.

Tips and Best Practices

• One chat per client or topic. Mixing accounts pollutes the context window and makes searches harder.
• Save important outputs. Copy key recommendations into a Google Doc — chat search is not always reliable.
• Use transcripts too. Fathom call transcripts can be pasted into a chat to ask specific questions: "What did Melissa say about the landing page?"
• Upload PDFs and large documents. The AI can summarize or answer questions about any document you drag in.
• Don't over-engineer prompts. For Google Ads analysis, plain-language questions work well. The AI already understands campaign structure.

Related

• ^[1]
• ^[2]
• ^[3]
• ^[4]
• ^[5]