Live against the Google Ads API — no sample data anywhere

Build the campaign. Then let AI argue with it.

Connect a Google Ads account, then tell the AI what you want to accomplish. It analyses, researches keywords, writes ads, and proposes changes — and every change waits for your approval before it reaches your account.

Runs on your own Claude or ChatGPT account. AI usage bills to you; your ad spend never routes through us.

The builder, playable

Click through the eight steps. Remove a keyword, pin a headline, approve the AI's recommendation, publish it — everything below responds.

nothing here touches a real account
manageads.com/campaigns/new/reviewtry it
Review & publishStep 8

Change preview — exactly what will be written

  • CREATECampaign US | NonBrand | Plumbing | Sales · Maximize conversions · $120/daypaused
  • APPLY3 locations · 1 language · 84 h/week schedule
  • CREATEAd group Emergency
  • ADD+4 keywords
  • ADD+12 negative keywords
  • CREATE1 responsive search ad
◆ AI recommendation · needs review
Recommendation
Move 3 keywords to exact match before publishing.
Reason
They overlap your brand campaign — “emergency plumber near me” among them — so the two campaigns would bid against each other.
Potential action
Change match type on 3 keywords, or exclude them here.
Publishes to Google Ads
6 recorded operations — a rejection halfway stays recoverable
The evidence

Numbers you can argue with.

“Always live” is the wrong target — Google's own conversion data lags, and constant polling burns the API quota that approved changes depend on. So: a scheduled background sync, a rate-limited “Refresh now”, a just-in-time refresh of any object immediately before a change is written to it, and a visible “last synced” on every screen showing performance.

Window
Cost per conversion, after an approved bid change
Daily · 14 days either side of the change
change applied day 0
$50$42$34$26day 014d0+14d
Illustrative shape, not a customer result — hover the plot, or tab to it and use the arrow keys, for a day's value. Your own numbers replace it once an account is connected.
Where an audit finds problems
On-demand audit · select a finding type
Keyword overlap18Low Quality Score12Broken final URL7Near-duplicate ad copy4
Select a bar

Each finding links to the object it was found on, with the evidence that triggered it.

Illustrative counts, not a customer result — the shape a first audit tends to take. Your own account replaces them once it is connected.
MCP server

Run your ads account from the assistant you already talk to.

Generate a personal MCP token in Settings, add it to Claude Desktop, claude.ai, Claude Code or ChatGPT, and ask in plain English. Read and analyse tools answer live. Change tools still create a pending proposed change inside ManageAds — MCP is not a second, less-supervised way into your account.

Claude DesktopClaude Codeclaude.aiChatGPT
manageads · mcpread: live · write: proposals only

Read & analyse — answered liveChange — filed for approval, never applied

The build

Every field Google demands, in the order a human would ask for them.

The wizard is opinionated on purpose: it refuses to publish a campaign that would serve everywhere, spend on Search Partners you didn't choose, or run without a conversion goal.

Guided campaign builder

Conversion goal, objective, bidding, targeting, budget, ad groups, ads, review. Leave mid-way; the draft resumes exactly where you left it.

8 steps · resumable

Live Keyword Planner

Real search volume, competition and bid estimates from Google, scoped to the languages and locations this campaign actually targets.

KeywordPlanIdeaService

Responsive search ads, pinned

Headlines and descriptions with real pinning, a live Ad Strength meter, and a best-effort policy screen before anything reaches Google.

15 headlines · 4 descriptions

Assets that outlive the campaign

Sitelinks, callouts and structured snippets are created once at account level and attach to anything you build afterwards.

account-level, reusable

Audit alerts & monitors

Audits run when you ask for them. What stands watch between them is the alert count — broken links, near-duplicate ad copy, keyword and negative-keyword opportunities — plus performance monitors you define yourself.

All · Critical · Errors · Recommendations

Publish that survives failure

Publishing runs as six recorded steps. If Google rejects step four, the first three stay done and a retry picks up where it stopped.

durable progress

The gate

AI proposes. A person approves. Nothing else touches the account.

Every AI-originated change — from chat, an on-demand audit, an alert, or a recommendation — becomes a proposed change first. You see the change preview, approve or reject it, and only then does it reach the Google Ads API. The AI never makes significant account changes on its own.

01

Analyse

Campaign structure, performance, relevance and opportunities — from chat, an on-demand audit, or a triggered alert.

02

Recommend

Recommendation, reason, and potential action — stated plainly, with the change preview attached.

03 · the gate

Review & approve

Approve or reject. Owner, Admin and Manager can act; Viewers see everything and change nothing.

04

Apply

Validated, sent to the Google Ads API, and kept in Recommendation and Audit History with who approved it.

Proof

Where the customer quotes go.

Three slots, sized and styled. Drop in real quotes with names, roles and the account size behind them — an unattributed quote is worth less than an empty box.

Quote slot 01

Agency, 20–50 accounts. Ideally about the approval gate.

Quote slot 02

In-house team. Ideally about time-to-launch.

Quote slot 03

Anyone using the MCP integration daily.

Get started

Connect an account. Build one campaign. Judge it on that.

Built for internal teams: connect a Google Ads account, connect your own Claude or ChatGPT key, and work. Credentials are encrypted at rest, never displayed again, and never written to logs in plaintext.