There’s a lot of hype right now around running your Google Ads with AI. Every week there’s a new post promising you can just chat with an assistant and watch it optimize your account.

Under the hype is a question worth sitting with: how much do you want an AI reaching into your account, and what should stay your call even after it’s done the work?

So we ran a small test. We asked four AI tools to take a plain-language request and turn it into a real, previewable rule staged in the account, not just a description you go build yourself. Then we watched where each one stopped, and where it handed the decision back to us. Here’s what we found.

The test

Same prompt for every tool: “Build a rule that pauses any keyword with over $50 spend and zero conversions in the last 30 days, and apply it to my account so I can review it before it goes live.”

Two things we watched for: does it get the rule right, and can it put it in the account?

Claude (without the Optmyzr MCP)

With no MCP connected, Claude hit the wall. It said straight away that it can’t touch the account, so it can only describe the rule, not put it anywhere, and handed over the exact Rule Engine config to build ourselves and run in preview.

That’s the rule we asked for. But Claude didn’t stop there. It focused on tightening the rule so it wouldn’t pause keywords we’d want to keep, which is the kind of second opinion you want before running anything:

• Conversion lag: shift the window to 31 days ago through 3 days ago so recent conversions aren’t counted as zero.
• Brand terms: exclude them.
• Flat $50: tie the threshold to a multiple of tCPA, since $50 means different things at a $15 vs. $80 target.
• Volume floor: add Clicks > 20 so you’re not calling a keyword dead on ten clicks.
• Assist keywords: use all-conversions or conversion value, not just last-click.

Sharp thinking, and better than a flat “here’s your rule.” But you’re still the one building all of it by hand. So we connected Claude to the Optmyzr MCP and asked again.

Connecting Claude to the Optmyzr MCP

Same question, MCP connected: can we build this rule in my account now?

This time the answer was yes.

Claude checked what the MCP exposed, found a preview_keyword_rule capability that builds an automated keyword rule with pause actions and cost/conversion conditions, and pulled our accounts to confirm which one to target before building anything.

Then it built the rule. Not a config to paste in, the actual rule, live in the account and staged with nothing paused yet. It also ran the rule against the account and flagged a match that looked like it might be intentional rather than wasted spend, telling us to confirm before applying. And it offered to rebuild the rule as Report Only if we wanted to watch what it caught for a few weeks before pausing anything.

The link drops you straight into the Rule Engine with the strategy already built.

The conditions are all there (Cost > 50, Conversions == 0, Clicks >= 20, last 30 days ending 3 days ago → Pause Keyword), tagged as AI-generated.

The preview screen shows the flagged keywords with full spend, conversion, and click data, plus an Apply Suggestions button. You review it, and you’re the one who clicks apply.

That’s the difference the MCP makes. Every tool could describe the rule. Only this one built it into the account for you, staged it against real data as a preview, and left the final click to you.

Gemini

Gemini landed where unconnected Claude did. It was upfront that it can’t act on the account, but wrote the rule out for us to set up in Google Ads’ native automated rules (Tools > Bulk Actions > Rules) and pointed to the preview step.

We asked what it would take to type a prompt and land on a ready-to-preview rule in the account itself. Not possible with the standard web app, Gemini said. You’d have to build the connective tissue yourself:

• a Chrome extension that reads your prompt and creates the rule via the Google Ads API
• a Retool/Make.com dashboard wired to do the same

No standard AI can deep-link into a private account natively, so short of building one of those, the best it offered was a bookmark to the Rules page.

Next, we asked directly about MCP. Gemini confirmed MCP is the mechanism that would close the gap, and that both Google’s Ads MCP and Optmyzr’s MCP are built for this kind of task.

It was also clear about its own limits:

• The standard Gemini web app can’t connect to custom MCP servers.
• It can’t connect to Google’s own Ads MCP either.
• Using either one means working outside the chat window, through the Gemini CLI, Vertex AI, or a developer tool like Cursor.

It also noted a difference between MCPs.

Google’s Ads MCP is oriented toward reading data and simple campaign changes, and doesn’t natively create an automated rule schedule out of the box.

Optmyzr’s MCP, by contrast, it described as connecting directly to the Rule Engine.

For a next move, Gemini offered two paths: connect Optmyzr through an MCP-compatible client like the Claude Desktop app, or take a Google Ads Script it offered to write on the spot and paste into the account yourself.

ChatGPT (with and without Optmyzr MCP)

Without an MCP, ChatGPT did what Gemini and unconnected Claude did: told us it couldn’t touch the account, then handed over the rule settings to build ourselves.

Connect it to the Optmyzr MCP and it clears the same wall Claude did.

It built the strategy directly in the account, returned a working preview link, and made no changes, recommending we run it as a report-and-label strategy first before pausing anything.

Same lesson as everywhere else in this piece: the MCP is what turns a rule you describe into a rule that’s built and waiting for your approval.

Optmyzr Sidekick

Every tool up to now has been working around the same limit: a generic AI doesn’t know anything about your account. It can describe a rule in the abstract, but it can’t see your campaigns, your spend, or which keywords are truly burning money, so it can’t build the rule against your data. That’s the gap.

The MCP closes it by feeding your account context to Claude or ChatGPT.

Sidekick never has that gap. It’s Optmyzr’s AI, sitting inside the platform with your account already in front of it. Nothing to connect, nothing to feed it.

It’s already where the rule needs to go, looking at the data the rule needs to run against.

So we typed the same prompt straight into the chat.

It built the reviewable rule in place, ran it against the account, and handed back the keywords that matched, each one with its cost and conversions and marked “Will be Paused.”

The “Open rule preview” link takes you into the Rule Engine to see the full strategy and apply it.

Nothing goes live until you click Apply. Same review-before-it-goes-live safety as the MCP setups, without any of the setup.

For Optmyzr users, this changes what building a rule takes.

No digging through the Rule Engine, no setting up conditions one field at a time, no remembering how to write a lookback window. You describe the rule you want in plain English, and Sidekick builds it, previews it against your real data, and waits for your go-ahead. Need a custom rule? Just type it. Sidekick handles the build; you still make the call on what gets paused.

The other three tools can get to this point once you connect them. Sidekick starts here.

Build, preview, and apply a PPC rule with any AI assistant, using Optmyzr

We gave the same job to Claude, Gemini, ChatGPT, and Optmyzr’s own Sidekick: “Build a rule that pauses any keyword with over $50 spend and zero conversions in the last 30 days, and stage it for review before it goes live.” Simple enough to type in one line, annoying enough to build by hand.

What we wanted to see was who could truly put the rule in the account. The answers ranged from “here’s a config, go build it yourself” to “done, here’s the preview, click apply.” That’s the whole trick: build, preview, apply.

Any AI can attempt the first step. Getting to a real preview staged against your live data, with an apply button waiting for you, takes context only Optmyzr gives it. The Optmyzr MCP feeds that context to Claude or ChatGPT, and Sidekick has it built in.

Start your 14-day free trial and connect it with your favorite AI assistant today!

FAQs

1. Can ChatGPT or Claude make changes to my Google Ads account?

On their own, no. An unconnected ChatGPT or Claude can suggest changes but can’t touch your account. Once you connect them to the Optmyzr MCP, they can stage real changes, a Rule Engine strategy or new positive and negative keywords, but everything comes back as a preview card first.

You review it and click Apply yourself. The AI can’t push anything live, because applying is triggered from the Optmyzr UI, not by the assistant. It’s the same review-and-approve step you’d get working inside Optmyzr directly.

2. Can AI build a Google Ads rule for me automatically?

It builds the rule, it doesn’t fire it. With the Optmyzr MCP connected, Claude or ChatGPT turns a plain-language prompt into a full Rule Engine strategy, runs it against your real data, and returns the result as a preview. Nothing is applied until you approve it. And it’s not limited to Google, the MCP works across Google, Microsoft, Amazon, Yahoo, Meta, and LinkedIn Ads.

3. Does the AI run in the background and act on its own?

No. It acts when you prompt it, in that conversation. It doesn’t monitor your account or make changes on its own between sessions. For something running on a schedule, that’s Optmyzr’s Rule Engine once you’ve applied the rule.

4. How do I connect ChatGPT or Claude to my Google Ads data?

Through the Optmyzr MCP, using OAuth (recommended for Claude and ChatGPT, no API key needed). In Claude, go to Settings → Connectors → Add custom connector, paste https://tools.optmyzr.com/OptmyzrMcp, then connect and approve access.

In ChatGPT, create an app with the same URL and set authentication to OAuth. Every other MCP client (Claude Code, Copilot Studio, Langdock, Cursor) connects with an API key from your Optmyzr settings.

5. Optmyzr MCP vs. Google’s Ads MCP: what’s the difference?

Google’s Ads MCP gives AI access to your Google Ads data through raw GAQL queries: useful for reporting, limited for optimization, and it expects you to know the syntax. The Optmyzr MCP connects the AI to Optmyzr’s analysis engine (the same one that powers Sidekick), so what comes back is already processed, with links straight into Optmyzr, instead of raw data the AI has to interpret.

6. Is my account data safe? Does it go to Anthropic or OpenAI?

Optmyzr’s MCP server doesn’t send your data to Anthropic. If you connect a Claude client, any data passing through it is governed by your own agreement with Anthropic, not Optmyzr, and the same applies to whichever client you pick. Once data reaches that client, the handling belongs to its provider under your terms with them, so review your AI client’s data policy before connecting it to live account data.

The MCP only exposes accounts your Optmyzr instance already manages, and Optmyzr keeps server-side logs of queries and metadata (tools called, response times) for security, not your full conversations.