What each level controls
Google documents which settings live at which level, and the table below is the basis for every structural decision: if two things need different values of a campaign-level setting, they need different campaigns.
| Level | Controls | Cannot be split below this level |
|---|---|---|
| Account | Billing, currency, time zone, conversion actions and default goals, shared negative lists, shared budgets, portfolio bid strategies, audience and asset libraries, auto-tagging, account-level negatives | Currency and time zone; conversion definitions |
| Campaign | Campaign type and sub-type, budget, bidding strategy and target, networks, locations and location options, languages, ad schedule, start and end dates, campaign-level conversion goals, campaign negatives, campaign-level assets | Budget; bid strategy and target; location; goal |
| Ad group | Keywords or other targeting, ad-group default bids under manual bidding, ad-group negatives, audience segments with observation or targeting, ad-group assets | The pairing of targeting with ads |
| Ads and assets | Headlines, descriptions, final URLs, images and video, asset-level reporting | The message |
Two things on this table are frequently misunderstood. A shared budget lets several campaigns draw from one pool, which is useful when the split between them should float; a portfolio bid strategy lets several campaigns share one target and one learning pool, which is useful when campaigns are split for reasons other than bidding. Both exist precisely so that a campaign split for one reason does not force a split for another. The bidding strategies guide covers when a portfolio is the right tool. In a brand-new account the account row of the table is settled before any campaign exists, and the new Google Ads account setup tutorial works through it in order.
Principles: consolidate data, separate by control need
The first principle is consolidate. Smart Bidding predicts the likelihood of a conversion at each auction from the history it has, and Google's guidance consistently favours giving a strategy more conversions in one place rather than fewer in many. Every unnecessary split divides the data, lengthens learning after any change and produces ad groups whose statistics are too thin to act on. An account built on the older pattern of hundreds of tiny ad groups is not more controlled; it is less measurable.
The second principle is separate only by control need. There are exactly five reasons to put something in a different campaign: it needs its own budget; it needs its own bid target or strategy; it needs its own location or language settings; it optimises towards a different conversion goal; or it needs reporting that the campaign level alone can provide. Anything else can be handled inside a campaign with ad groups, audiences, labels and the reporting dimensions. Ad groups are split for one reason: the ads that would serve one set of targeting would be wrong for the other.
Structure is a means to control, not a tidiness exercise. A structure with fewer campaigns than feels comfortable but which answers every control need is better than one that mirrors the website's navigation and splits the data forty ways.
A useful test for any proposed split: name the campaign-level setting that will differ. If the answer is "none, but it is a different product", the split belongs at the ad group level. If the answer is "none, but we want to see it separately", labels and segment reporting answer that without a split. The CPA calculator helps compare whether two proposed segments actually need different targets or merely feel different.
Search structure: intent groups, brand and generic, geography, device
On Search, ad groups are organised around intent groups: sets of queries that want the same answer and can share a landing page and a set of headlines. "Emergency plumber", "plumber near me" and "plumber open now" are one intent; "boiler installation cost" is another. Match types, covered in the keyword match types guide, decide how widely each group reaches; the group itself is defined by what the person wants, not by the words they used. Quality Score (Quality ScoreGoogle's 1–10 keyword diagnostic.) is reported per keyword, and its relevance components improve when the ads in a group speak to every keyword in it, which is the strongest practical argument for intent grouping.
Brand and generic terms belong in separate campaigns in almost every account. Brand clicks are cheap (CPCAverage cost of each click.) and convert well; generic clicks are expensive and convert less. Sharing a budget means brand starves generic or generic exhausts the budget before the brand searches of the evening; sharing a target means the blended cost per acquisition (CPACost of each conversion.) hides the real figures for both. Separation also makes Impression ShareShare of eligible impressions you received. meaningful: lost share on brand is a different problem from lost share on generic. Whether the brand campaign earns its budget at all is a separate question, which the brand search holdout test answers with a regional pause.
Geography is a campaign split when locations need different budgets, targets, languages or ad schedules, or when the business has different offers by region. It is not a split when the only difference is a city name in the ad, which location insertion and ad customisers handle. Over-splitting by geography is the most common way accounts end up with dozens of under-funded campaigns.
Device has not been a campaign split for years in the standard interface; bid adjustments and the Smart Bidding signals handle it. A device-specific campaign is justified only when the offer or the landing experience genuinely differs by device, for instance an app-install flow on mobile. Before a new campaign is built, these splits can be written into the structure map of a search campaign build sheet, which records each ad group and why it is separate so a reviewer can challenge the split before it exists in the account.
Structure for Smart Bidding versus legacy granular structures
The granular structures of the manual-bidding era, often called single-keyword ad groups, had a logic: one keyword, one ad, one bid, perfect control. Three changes removed that logic. Close variants made the "exact" keyword match a range of queries anyway; responsive search ads made the one-ad-per-group idea moot; and Smart Bidding moved bidding from the keyword to the auction, where the keyword is one signal among many. What remains of the granular structure is fragmentation: ad groups with a handful of conversions a month, each one too thin to tell whether a change helped.
A Smart Bidding structure inverts the priority. Campaigns are as few as the control needs allow; ad groups are intent groups large enough for their statistics to mean something; match types are broader than the legacy approach would tolerate, with the search terms report and negativesKeywords that stop an ad from showing for searches containing them. doing the pruning. The strategy and target sit at the level with the most data, and portfolios combine campaigns that had to be split for other reasons. The Smart Bidding strategies guide explains what each strategy needs; structurally, the job is to make sure it gets it.
This does not mean one campaign for everything. Campaign-level controls still exist for good reasons, and an account whose single campaign cannot be given a different target for its most profitable product line has over-consolidated. The test is the same as before: does a control need exist, and at which level?
Shopping, Performance Max and video structures
Shopping campaigns are structured by product groups rather than keywords. The campaign's product filter decides which items are eligible; ad groups and product groups subdivide the inventory by attribute such as brand, category, product type or custom label. The usual structural split is by margin or role: products that can bear a different return target, or that need their own budget during a season, sit in their own campaign with a product filter, and custom labels in the feed are set up precisely to make that split possible without touching the campaign each time the range changes. Priority settings between Shopping campaigns decide which campaign serves when a product qualifies for several.
Performance Max replaces most of the structure with asset groups inside a single campaign. Asset groups pair a set of creative with an audience signal and, for retail, a listing group of products. Campaign splits still follow the campaign-level rule: a different budget, target or goal justifies a second campaign; different creative for different product themes is an asset-group matter. Brand exclusions and campaign-level negatives, where available, decide how Performance Max interacts with the Search campaigns alongside it, and most accounts keep a brand Search campaign running next to it so that brand traffic is reported and controlled separately.
Video campaigns are structured by objective and format first, since the campaign sub-type fixes which formats and bidding strategiesThe rule an ad platform follows to set the bid in each auction. are available, then by audience in ad groups. A campaign that needs different creative for different audiences uses ad groups; one that needs a different budget or goal uses a separate campaign. The same five-reason test applies across every type, and it is the reason the account's structure can be explained on one page even when it mixes all four.
A kitchenware retailer consolidates
A kitchenware retailer inherits an account with 84 Search campaigns, one per product category and country, most with fewer than 10 conversions a month, all on manual bidding with a blended target. The restructure reduces Search to seven campaigns: brand, and generic campaigns for the three product families that have different margins, each split for the two country groups with different budgets. Ad groups are rebuilt as intent groups of roughly 15 to 40 keywords, and match types are widened with a shared negative list of around 600 terms carried over from the old account. Shopping is reorganised into two campaigns on custom labels for high-margin and clearance stock with different return targets. The migration runs as an experiment on the largest generic campaign for four weeks before the rest follows. Six months later the account converts about a third more per pound spent, with the gain concentrated in generic campaigns that previously had too little data in any one place for bidding to learn.
Experiments and drafts as structural tools
A structural change is a hypothesis, and Google Ads experiments are how it is tested. A campaign is copied, the copy is altered, and traffic is split between the two for a set period; the report shows the difference with a confidence indication. Older versions of the interface exposed a separate drafts step in which the copy was edited before the experiment was launched; current versions fold this into creating the experiment. The principle is unchanged: change one thing, run for long enough to pass the learning period and cover a full weekly cycle or more, decide on a rule set in advance, then either apply the change to the original or discard it.
Structural experiments that are worth running include a broader match-type set against the existing one, a consolidated campaign against a split pair, a new bid strategy against the current one, and a different landing page for the same ad group. Experiments that are not worth running are those whose outcome the control-need test already decides; if two campaigns must have different budgets, no experiment will change that. Naming the experiment and its campaigns consistently matters more than it sounds, and the naming conventions guide covers the pattern.
When to restructure and how to migrate
Restructure when the current structure prevents a decision, fragments data below the level bidding can use, or no longer matches the business: new product lines, new markets, a change in what the primary conversion is, or a move from manual to automated bidding. Do not restructure because a new account manager prefers a different pattern, because the account is untidy, or because performance dipped last month. An account audit is the usual way of deciding whether the structure is the problem or merely the most visible thing.
Migration is where restructures fail. Google does not document a way to move performance history from one campaign to another, so a new campaign begins with a learning period, and a wholesale cut-over resets the whole account at once. The safer sequence is to build the new campaigns alongside the old, migrate one segment at a time starting with the largest, run an experiment where the change is contestable, carry over negatives, audiences and the best-performing ads, keep the old campaigns paused rather than deleted so their history remains visible, and hold targets steady through the learning period before tightening them. Plan the migration against the media plan so that it does not land in the peak season, and record each step in change history with a note. The campaign management expertise page describes how such migrations are scheduled and reviewed on client accounts.
When an advertiser running several engines moves its search into Search Ads 360, each Google Ads account keeps the structure described here beneath an SA360 client account, and the tutorial on moving paid search into SA360 sequences that move.
Before restructuring an inherited account, work through the Google Ads account audit so the structural problems are recorded alongside tracking, bidding and query issues rather than fixed in isolation.
Where an account is weighing a Performance Max campaign against its existing Search campaigns, the Performance Max versus Search campaigns comparison explains how the two divide queries under Google's priority rules.
How broad and exact keywords can share an account, with exact match holding proven queries and broad match exploring around them, is covered in the comparison of broad match and exact match.
Once brand and generic terms sit in separate campaigns, the comparison of branded and non-branded search campaigns covers how to set their targets apart and test whether brand bidding adds sales.
Common mistakes
- Mirroring the website. Site navigation is for visitors; campaigns are for control needs.
- One keyword per ad group on automated bidding. Fragmented data, slow learning, no better relevance.
- Brand and generic in one campaign. Blended figures that hide both and a budget the brand terms win.
- Splitting by geography without a reason. Dozens of starved campaigns where ad customisers would do.
- Splitting for reporting. Labels and segments report; campaigns control.
- Restructuring everything at once. The whole account relearns at the same time, usually in a busy month.
- Deleting old campaigns. Pausing keeps the history visible for the next audit.
- Changing structure and bidding together. Two variables, one result, no conclusion.
- Ignoring shared budgets and portfolios. They exist so that a split made for one reason does not force a split for another; an account that never uses them usually has too many campaigns.
