This is the operating playbook for AdsPulse, version 1.0 of the October 2026 beta. It is not a product presentation: the announcement explains what a Google Ads MCP is and why we built one. This is how we use it every day, written down so you can do the same without going through our trial and error first.
Four numbers to frame it: 16 use cases, 4 families of work, 38 Google Ads managers behind the scenes, 0 changes without confirmation. The playbook is split into four chapters, one per family, each with four use cases, and for each one the prompt to copy, what you get back and what to watch out for. This page is the map.
How it works: you ask in plain language, AdsPulse works on the account
AdsPulse connects your AI assistant, Claude for example, to your Google Ads accounts. It reads the data, interprets it and, when needed, prepares the changes. Three rules explain everything else.
- Every request names its account. No account stays selected between one request and the next. Give a name or an ID: AdsPulse always tells you which account it is working on, and you can follow several clients in parallel without mixing their data.
- Reads are free. Reports, search terms, change history, Quality Score, auction insights. All read-only: no analysis touches the account, so you can run one on a client that arrived this morning.
- Writes go through you. Every change arrives first as a preview. You see what would change, you approve, and only then is it applied. Removals and exclusions are flagged as permanent.
The first ten minutes
- Connect. Add AdsPulse as an MCP connector to your assistant and authenticate with your personal API key.
- Find the accounts. Ask: “List my Google Ads accounts with their MCC and ID”. From here on every request names one of these.
- Give the context. Describe the client: objective, target CPA, what they do not want to sell. The numbers are read in that context, not in a vacuum.
- Start from the overview. “Give me the overview of [account] for the last 30 days against the previous period”. It is the first read that tells you where to look.
Five specialists behind every answer
The answers do not come from a single generic assistant. Behind every request works the right specialist for the topic:
| Specialist | What it handles |
|---|---|
| Simba | Reports, breakdowns and trends |
| Nemo | Keywords, competitors, market |
| Elsa | Budgets, bids, negatives |
| Aladdin | Shopping and Performance Max |
| Genie | Custom GAQL queries |
The map: sixteen use cases, four families
Eleven use cases are read-only: you can try them today on any account with no risk. Five write to the account, always after your confirmation. Every row links to the chapter and to the exact spot where the prompt lives.
| # | Use case | Cadence | Type |
|---|---|---|---|
| Routine and control · chapter 1 | |||
| 01 | Multi-account morning report | Every day | Read |
| 02 | Search terms and negatives | Every day | Write with confirmation |
| 03 | Pacing and budget | Every week | Read |
| 04 | What changed? | When needed | Read |
| Audit and diagnosis · chapter 2 | |||
| 05 | Full account audit | Every month | Read |
| 06 | Tracking check | Every month | Read |
| 07 | Landing page audit | Every month | Read |
| 08 | PMax and Shopping X-ray | Every two weeks | Read |
| Optimisation and execution · chapter 3 | |||
| 09 | RSA and sitelinks from the landing page | When needed | Write with confirmation |
| 10 | New ad group or new campaign | When needed | Write with confirmation |
| 11 | Bids and budgets | Every week | Write with confirmation |
| 12 | Google recommendations | Every week | Write with confirmation |
| Research and new business · chapter 4 | |||
| 13 | Keyword research and forecast | When needed | Read |
| 14 | Competitors and impression share | Every week | Read |
| 15 | Pitch for a prospect | When needed | Read |
| 16 | Any question, in GAQL | When needed | Read |
If it is your first time. Start with 01, 02 and 04: they are the ones that save time every day and take two minutes to try. All three are in the first chapter.
The typical week: a routine that holds up even with twenty accounts
One way to spread the sixteen use cases across the week of a freelancer or a small agency. Adapt it: the cadence matters, not the day.
- Every morning · know where to look01 Multi-account morning report, 02 Search terms and negatives.
- Monday · set up the week03 Pacing and budget, 11 Bids and budgets.
- Wednesday · clean up and decide12 Google recommendations, 08 PMax and Shopping X-ray.
- Friday · look outside the account14 Competitors and impression share.
- Once a month · the deep check05 Full account audit, 06 Tracking check, 07 Landing page audit.
- When needed · on request04 What changed?, 09 RSA and sitelinks from the landing page, 10 New ad group or new campaign, 13 Keyword research and forecast, 15 Pitch for a prospect, 16 Any question, in GAQL.
If you manage many accounts. Write the request once and ask it for every account under your MCC. Each call names its own account, so the analyses can run in parallel without mixing the data.
The rules: seven habits that make the difference
They come from the daily use of AdsPulse on the accounts Konvtrack manages. Almost every disappointing answer comes from one of these.
- Always name the account. Name and ID. Check that the answer repeats it before you read the numbers.
- Give a period, a metric and a threshold. “Last 14 days, more than €10 of spend, zero conversions” beats “the terms that are not working”.
- Ask for the cause, not just the number. Have performance cross-referenced with the change history: the figure on its own explains nothing.
- Do not judge incomplete days. With offline or affiliate conversions the latest days fill in later. Wait for them to settle.
- Read the whole preview. It is the only moment when correcting costs nothing. If you change a detail, ask for a new preview.
- One removal at a time. Deletions and exclusions are permanent. Do not bundle them with other changes.
- New landing page, new ad group. Do not overwrite ads that have history: run the new one alongside, compare, then pause.
Anatomy of a good prompt
A prompt that works has five parts, and the one that does not work has almost always skipped one of them.
On [account], over the last 14 days, find the search terms with more than €10 of spend and zero conversions. Group them by theme in a table and propose the negatives, without applying them.
| Part | The question it answers | In the example |
|---|---|---|
| Account | Where | On [account] |
| Period | When | over the last 14 days |
| Metric and threshold | What counts | more than €10 of spend and zero conversions |
| Format | How you want it | group them by theme in a table |
| Action | How far to go | propose the negatives, without applying them |
The four chapters
- Routine and control: what you do every day, in ten lines of chat.
- Audit and diagnosis: understanding what is wrong before touching anything.
- Optimisation and execution: from the decision to the change, with the preview in between.
- Research and new business: the numbers you need before you spend.
Missing a use case you run every day? Tell us: this playbook grows with feedback from the beta. AdsPulse is in beta access at adspulse.dev; if you would rather see how we apply it to an account, the free audit starts there.