Google Ads Automation: What to Automate, What You Decide

See which Google Ads jobs a store can hand to Smart Bidding, rules, scripts or AI agents, and which decisions, such as targets and budgets, stay with a person.

An online store should automate the repeatable work in Google Ads. Let Smart Bidding set the bid in each auction, and let rules, scripts and the API run checks, alerts, reports and exports. The decisions that set what the account learns from and spends on stay with a person. These are conversion goals, targets, campaign structure, budget moves beyond small steps and switching products off. Give AI assistants read access first. Any tool that makes changes should show the current value and wait for your yes.

Automation in Google Ads comes in layers that stack up like a ladder. The layers differ in two ways: who decides what should change, and who presses the button that changes it. This guide climbs from Google’s built-in automation up to AI agents. For each layer, you’ll see the job it does well in a store, the limits Google documents for it, and the decision that should stay with a person.

What are the layers of Google Ads automation?

From the bottom up, there are six:

Google’s built-in automation. Smart Bidding sets a bid in every auction. You can set recommendations to apply on their own, and Ads Advisor suggests changes in a chat inside the account.
Rules. Automated rules change a status, a budget or a bid when a condition you set is met.
Scripts. Google Ads scripts are JavaScript code that runs inside the account on a schedule and can read, create and change items in it.
The API and exports. The Google Ads API lets your own software read the account and write to it. Exports copy the data into a spreadsheet or BigQuery.
AI assistants. An MCP server connects an assistant such as Claude or ChatGPT to your account data, so you can ask questions in plain language.
AI agents with write access. The same assistants, given the right to make changes, turn a finding into an edit in the live account.

Layer	Who decides	Who presses the button	What it does well in a store	Where it stops
Google’s built-in automation	Google’s models	Google; for recommendations, you or the auto-apply setting	Bids in every auction across thousands of products	Optimises for whatever conversions and values you send
Automated rules	You, in advance	Google Ads, on a timetable	Dated promotions, alerts, simple caps	One condition on one window of data
Scripts	Your code	The script	Account-wide checks, reports, edits from outside data	30 minutes per run; someone has to maintain the code
API and exports	Your software	Your software, or nobody if it only reads	Product-level history and your own tools	A developer token; the API change logs reach back 30 and 90 days
AI assistants	You, after reading the answer	Nobody, if the tools only read data	Questions in plain language	Your data goes to the model; numbers need checking
AI agents with write access	The model proposes	The agent, ideally after your confirmation	Turning a finding into a change quickly	A wrong change lands in the live account

For a store, the columns that matter most are who decides and who presses the button. Each layer up takes a person out of one more step, so each one needs a stricter check before it writes anything. To choose between the lower layers for a given job, read rules vs scripts vs Smart Bidding.

Google Ads Editor sits beside the ladder. It imports bulk changes from a CSV file. So a script or an AI tool can prepare a change, and a person can review it before it reaches the account.

What does Google already automate in a store account?

According to Google Ads Help (About Smart Bidding), Smart Bidding strategies use Google AI to optimise for conversions or conversion value in every auction. Google calls this auction-time bidding. The four strategies are Target CPA, Target ROAS, Maximize conversions and Maximize conversion value. For Performance Max, Google’s developer guide (Create a Performance Max campaign) lists only two: Maximize conversions and Maximize conversion value, each with an optional target.

That makes bidding the most automated decision in a typical store account. Performance Max runs in 141 of 146 stores in our data and takes a median 95.7% of their ad budget (GetProfit data, June 2025 – June 2026). For most of that money, Google sets every bid.

A person still controls the input: which conversions and values the model learns from, the target, the budget, and which products go into which campaign. How Smart Bidding works explains how the model uses that input and what no bidding change can fix.

Auto-applied recommendations are the second piece. Google Ads Help (About applying recommendations automatically) describes a setting that applies the recommendation types you choose on a regular basis, at account level only. The list includes adjusting your CPA and ROAS targets and adding broad match keywords, but leaves out raising the budget. Applied recommendations show in the History tab and in change history.

For a store, the types that move a target or widen keywords change what the account buys. Those should stay with a person. To see what Google applied without you and decide which types to switch off, follow how to audit auto-applied recommendations.

The third piece is Ads Advisor. Google announced it on 12 November 2025 (Google’s AI advisors: agentic tools to drive impact and insights) as an agent inside Google Ads that uses Gemini models. Google said it would reach all English-language accounts globally in early December 2025.

Ads Advisor analyses performance, suggests keywords and assets, and helps find the cause of problems such as policy disapprovals. After you review and approve the changes, it applies them itself. Ads Advisor for an online store covers what it can take over in a catalogue store.

Google’s recommendations work at the level of campaigns and settings, and they rest on auction signals and campaign statistics. A store also needs the product view: which products earn, which only spend, and how that has changed over time. The two views answer different questions, and a store needs both before it hands decisions to a machine.

Automated rules: if-then changes on a timetable

Google Ads Help (Set up automated rules) describes rules as automatic account changes based on settings and conditions you choose. A rule can change ad status, budgets and bids, or send you an email. In Performance Max, rules work on asset groups: go to Rules in the Tools menu, select the plus button and choose ”+ Asset group rules”.

Two details from Google’s examples matter for a store (Common ways to use automated rules). A rule runs within 2 hours after its conditions are triggered, so a rule set for 9 a.m. may run any time until 11 a.m. Conversions may take up to 30 days to arrive, so Google recommends long data ranges, such as the previous seven or 30 days, for rules based on conversions.

Why a ROAS threshold rule fires in ordinary months

A rule checks one condition on one window of data, and a store’s ROAS moves a lot on its own. We looked at 110 stores with at least eight months of history (GetProfit data, June 2025 – June 2026). In a typical month, ROAS sits 16% away from the store’s own median. The best month’s ROAS is 3.1× the worst month’s (both figures are medians).

Example store, not client data.

A tableware shop spends 40,000 a month and gets 180,000 in revenue, a ROAS of 4.5. A month 16% below that gives a ROAS of 4.5 × 0.84 = 3.78, which is ordinary for this store. A rule that cuts the budget by 20% whenever the 30-day ROAS drops below 4.0 fires in that month. If it fires twice, the budget falls from 40,000 to 32,000 and then to 25,600, while the account itself works as usual.

The portal flags a negative trend when ROAS falls by 15% or revenue by 10% two months in a row. The second month separates a trend from an ordinary dip. A rule that reads one window sees only the dip.

Rules do their best work when the condition is a date or a hard limit. Good examples are switching promotional ads on and off for a sale and pausing what clearly underperforms over a long window. A rule can also send an email when spend crosses a line.

Scripts: code that runs inside your account

According to Google Ads Help (Using scripts to make automated changes), scripts are JavaScript code you enter in the account. They can create, edit and remove items, use outside data such as inventory to pause or resume keywords, and write reports to a spreadsheet. A manager account can run one script across its client accounts. You can preview a script’s results before running it, and a script needs your authorisation before it makes changes.

For Shopping, scripts manage existing campaigns: product group hierarchies, product group bids and reports. They can’t create Shopping campaigns, set campaign-level Shopping properties or link Merchant Center accounts (Shopping campaigns).

Google’s Limits page sets how far scripts stretch in a large catalogue:

Limit	Value	What it means for a store
Run time	30 minutes; up to 60 for manager account scripts that process accounts in parallel with a callback	A script that walks a large catalogue item by item may not finish
On timeout	Changes made before the cut-off stay applied	An unfinished run leaves the account half changed
Iterator	50,000 results by default, adjustable with `withLimit()`	Large lists need batching
IDs in one selector	10,000	Long ID lists need batching too
Log output	Truncated at 100 KB	A long log loses its end
Manager scripts in parallel	Up to 50 accounts	More accounts than that need several runs
Authorised scripts	250 per account; beyond that, an older script is deauthorised until it is reauthorised the next time it is opened	Old scripts pile up unnoticed

Reports sit outside these entity limits. A script usually does one task, and when Google updates something, someone has to fix the code. Google Ads scripts for online stores lists the scripts worth running.

Why a “pause on zero sales” script loses future sales

Take a script that pauses products with spend and no sales. In our study of 1.4 million products, we tested a fixed threshold of this kind on 514,602 products with spend over 13 months. The decision to switch a product off was right 88.1% of the time at $5+ of spend and only 66.2% at $500+. And 35.4% of products in the study’s “loser” status converted in later months, with a median of 2 months.

So a script should flag such products and shift budget away from them, and a person should decide whether to pause them. The exclusion rule in the portal’s methodology waits for at least 50 clicks and zero conversions over 360 days. A script for spend without sales shows how to build that check.

The API and exports: when do you need your own history?

The API lets outside software do what scripts do inside the account, and more. It needs a developer token, and the token’s access level sets how many operations it can run (Access levels and permissible use):

Access level	Operations a day against live accounts
Explorer	Up to 2,880
Basic	Up to 15,000
Standard	No operation limit

The API has a single OAuth scope, https://www.googleapis.com/auth/adwords (OAuth 2.0 internals for Google Ads API). The same access covers reading and making changes, so even a tool that only reads asks for broad permissions on Google’s screen. For a hard limit, connect the tool through a Google account that holds the Read only role in your Google Ads account (how to grant read-only access). The tool then acts with that user’s rights, and Google rejects any change it attempts.

For automation, the key API fact is how far back each change log goes:

Where	How far back	What you get
Change history in the Google Ads interface	2 years	Who changed what, shown against performance
`change_event` in the API	Past 30 days, up to 10,000 rows per query	Old and new values of each change, and its source
`change_status` in the API	Past 90 days, up to 10,000 rows per query	Which resources changed; only the latest change for each

The figures come from Google’s pages About change history, Change Event and Change Status. Each change in change_event also has a field for its source. According to the API reference (ChangeClientType), that field records whether a change came from the web interface, automated rules, scripts, recommendations, recommendations you subscribed to, or the API. Google Ads Editor has a value in that list too, but the API doesn’t return those changes.

So an automated audit through the API sees one month of detailed changes. If your tools need more, pull change_event every day and store it.

For performance data, Google’s BigQuery Data Transfer Service runs every 24 hours by default (Load Google Ads data into BigQuery). Each run re-reads up to the past 30 days, 7 by default. Automating your Google Ads exports helps you pick the route that fits your store, from scheduled reports to Sheets and BigQuery.

Product-level numbers for Shopping and Performance Max also come from the API, and a few traps can make them wrong. The queries and the traps are in Google Ads API product reporting. If you need a regular report rather than raw data, start with Google Ads reporting for online stores.

AI assistants: should they read before they write?

Yes. An MCP server is the connector between an assistant and a data source. The open-source Google Ads MCP server, published by the googleads organisation on GitHub, offers three tools: search, get_resource_metadata and list_accessible_customers. All three read data and leave the account as it is.

The server’s README also notes that it passes your data to the agent or model you connect it to. If your setup needs a developer token, the same README says it must have at least Explorer access to query live accounts.

Start with tools that only read data, and use them for questions in plain language and drafts of queries. With these tools, the risk sits in the answer. A model can fill a gap with a plausible number, so every figure in an answer should trace back to a query result.

AI analysis of Google Ads data shows how to prompt for that. For setup and what to lock down first, see connecting an assistant via MCP.

AI agents with write access: what has to happen before Apply?

An agent with write access sits at the top of the ladder. Google’s Ads Advisor applies changes after you review and approve them. A third-party agent makes changes through the Google Ads API, with the same single permission scope as any other tool. Four rules fit any agent, and the fourth comes from the portal’s methodology:

Read-only first. The agent starts with tools that only read data, as the Google Ads MCP server does.
Every change shows the current value and waits. The agent states the current value and the new one, then acts only after an explicit yes.
Structure changes arrive as a file. The agent saves a new campaign structure as a CSV for bulk edits in Google Ads Editor. A person reviews it there before import.
Changes go in small steps and get judged. The methodology moves a target ROAS by up to 15% at a time, at most once every one to two weeks. A budget moves by up to 20% at a time. The methodology judges each change over 14 days.

AI agents that change your account covers the other risks and guardrails.

The portal works as a reading layer. It reads Google Ads and Merchant Center and shows what to change in a plan of recommendations. You make the change in your own account.

Which decisions should stay with a person?

Bids in each auction, checks, reports and alerts can run on automation. The inputs that set what the account learns from and buys stay with a person:

Decision	Hand to automation	Keep with a person	Why
The bid in each auction	Smart Bidding	—	Google’s models weigh more signals per auction than a person can
What counts as a conversion, and its value	—	Yes	Smart Bidding learns from exactly that
Target ROAS or target CPA	Alerts from rules or scripts	Yes, in steps of up to 15%, at most once every 1–2 weeks	A target sets what the whole account buys
Budget	Dated caps and alerts	Yes, in steps of up to 20%	A typical month’s ROAS sits 16% from the store’s own median
Switching products off	A script that flags candidates	Yes, after a long window	35.4% of “loser” products in our study converted later
Campaign structure	A draft as an Editor CSV	Yes: review, then import	One import touches many products at once
Auto-applied recommendation types	Types you would approve by hand anyway	Types that move targets or add broad match keywords	They change what the account buys
Reports, exports and alerts	All of it	—	They only read data

Fill in this table for your own account, and you have most of an automation plan. For each decision you keep with a person, name that person and a step size.

What to do this week

Read the change log by source. In Google Ads, open Change history from the Campaigns menu and look at the last 90 days. The User column shows an email for people, and names such as “Google Ads API” or “Google Ads system” for tools and Google’s systems. If tools and Google’s systems already make many of the changes, decide who reviews them before you add another layer of automation.
Check what Google applies for you. In Recommendations, open the auto-apply settings from the top bar. Keep only the types you would approve by hand.
List every rule and script. For each one, note the owner, what it changes and which data window it reads. Remove what nobody owns.
Keep the history you’ll need. If a tool reads changes through the API, store them daily, because change_event reaches back only 30 days.
Give AI tools read access first. Use tools that only read data, or a Google account with the Read only role.
Write down who presses the button. For targets, budgets, conversion goals, campaign structure and switching products off, name the person and the step size.

What to do next — and why exactly that. The portal doesn’t just show figures: for every finding it says what to do about it and explains what the advice rests on. The portal changes nothing without your consent.

See your plan →

Frequently asked questions

Is Smart Bidding the same as Google Ads automation?

No, it is one layer of it: Smart Bidding automates bids. Rules, scripts, the API and AI tools automate other jobs, and each needs its own owner.

Do automated rules work in Performance Max?

Yes, at the asset group level. Google Ads Help describes creating them from the Asset groups page or from Rules in the Tools menu.

How far back can I see who changed what?

Two years in the Google Ads interface. Through the API, change_event covers 30 days and change_status 90 days, so a tool that audits changes needs its own copy to look further back.

Sources

About Smart Bidding — definition of Smart Bidding, auction-time bidding, the four strategies, signals beyond what a person can compute. Checked 2 October 2026.
Create a Performance Max campaign — the only bidding strategies Performance Max supports. Checked 2 October 2026.
About applying recommendations automatically — account-level auto-apply, the list of types, no budget raising, History tab and Change history. Checked 2 October 2026.
Google’s AI advisors: agentic tools to drive impact and insights — Ads Advisor: announcement date, rollout, capabilities, changes applied with review and approval. Checked 2 October 2026.
Set up automated rules — what rules change; asset group rules for Performance Max. Checked 2 October 2026.
Common ways to use automated rules — 2-hour turnaround; conversions taking up to 30 days; long data ranges for conversion rules. Checked 2 October 2026.
Using scripts to make automated changes — what scripts do, outside data, manager accounts, preview and authorisation. Checked 2 October 2026.
Shopping campaigns (Google Ads scripts) — what scripts can and can’t do in Shopping campaigns. Checked 2 October 2026.
Limits (Google Ads scripts) — run time, timeout behaviour, iterator and selector limits, logging, manager scripts, 250 authorised scripts. Checked 2 October 2026.
Access levels and permissible use — Explorer, Basic and Standard operation limits. Checked 2 October 2026.
OAuth 2.0 internals for Google Ads API — the single OAuth scope of the Google Ads API. Checked 2 October 2026.
About change history — 2 years of changes in the interface; how users, tools and Google’s systems appear. Checked 2 October 2026.
Change Event — 30-day window, 10,000-row limit, old and new values. Checked 2 October 2026.
Change Status — 90-day window, 10,000-row limit, only which resources changed. Checked 2 October 2026.
ChangeClientType — sources of a change; Google Ads Editor changes are not returned. Checked 2 October 2026.
Load Google Ads data into BigQuery — default 24-hour schedule, refresh window of up to 30 days, 7 by default. Checked 2 October 2026.
googleads/google-ads-mcp — the three tools of the Google Ads MCP server, the note on data passed to the model, the developer token requirement. Checked 2 October 2026.
CSV file columns (Google Ads Editor Help) — Google Ads Editor makes changes from CSV files. Checked 2 October 2026.
GetProfit data: 146 stores, June 2025 – June 2026 — share of budget in Performance Max.
GetProfit data: 110 stores with at least eight months of history, June 2025 – June 2026 — monthly ROAS swings.
GetProfit study: 1,404,808 products, 130+ stores, 13 months — accuracy of a fixed switch-off threshold; later conversions of “loser” products.
GetProfit portal methodology — trend rule, step sizes for targets and budgets, 14-day judgement of a change, exclusion rule for products.