.png)
Most Google Ads structure advice in 2026 is a 2017 account plan with new vocabulary bolted on. Single-keyword ad groups, alpha/beta campaigns, and one campaign per match type made sense when I set the bid and Google honored it.
Last spring, I rebuilt an account with 43 campaigns and 178 ad groups on roughly $34k a month. It looked tidy in the UI. Every campaign averaged fewer than a dozen conversions a month. tCPA targets swung from $60 to $210 week to week. The previous manager reported campaign-level CPA as though those numbers meant something.
They didn't.
Segmentation used to buy you control. Now it mostly buys you noise.
Smart Bidding builds its model per bid strategy. It needs enough conversion events inside that model to predict anything useful. Split spend across 43 campaigns and you create 43 underfed models instead of one well-fed one. Each model re-enters learning when you touch it, overreacts to a fluke conversion, and never gathers enough volume to separate a good auction from a lucky one.
My rough floor is 30 conversions per campaign per 30 days before I trust a tCPA target. I want roughly 50 before I trust tROAS, because value-based bidding has to model both conversion probability and conversion value. Below those numbers, you are not managing a bid strategy. You are watching a random number generator with a budget.
The structural question for 2026 is not, "How do I organize my keywords?" It is, "How few campaigns can I run, and what has to be true before I split one?" Every extra campaign taxes the algorithm's learning. It has to earn that tax back.
I use two tests. Both need a yes:
A product line converting at a $40 CPA and one converting at $300 belong apart. One target cannot serve both. Six locations in the same state selling the same service usually do not, no matter how much the client wants a campaign named after each city. You can report on geography without paying for it in structure.
For most accounts below $50k a month, the structure fits on one line:
That is usually enough. I have run $200k-a-month accounts with six campaigns and seen $8k-a-month accounts split into 19. The bigger account was easier to manage and performed better because each campaign had enough conversions to support its target.
Keep brand separate for two reasons:
Exclude brand terms from everything else, including PMax, and stop congratulating yourself on blended CPA. Blended brand and non-brand reporting is how a flat account gets presented as growth.
I used to build 300-keyword Search campaigns and mine the search terms report every week for terms to add. I was wrong. I stopped around 2023.
Adding a keyword no longer adds control the way it did when each keyword carried its own bid. With Smart Bidding, the exact, phrase, and broad versions of the same term feed one model and mostly compete for the same auctions. A huge keyword list adds fragmentation and a maintenance burden that scales with nothing useful.
My default is 10 to 25 keywords per Search campaign, weighted toward phrase and broad match and backed by a serious negative keyword list. Exact match earns its place in two cases:
"Emergency plumber" and "plumber jobs" are not near variants to you. They are to the matching engine on a bad day.
Use one to three ad groups per campaign. Build each around one intent that deserves distinct copy and a distinct landing page, not around a keyword. If two ad groups point to the same URL and say roughly the same thing in their headlines, they are one ad group wearing two hats. Splitting them only halves the data behind both.
Two responsive search ads per ad group are enough. Google allows three, but the third rarely gets enough impressions to tell you anything before the offer changes. Pin sparingly, and only for compliance or a required disclaimer. Every pin narrows the combinations the system can test. That testing is the part of RSAs that genuinely works.
Split Search only when economics, landing pages, or budget control demand it.
Performance Max is a distribution engine attached to a feed. If you sell products and your Merchant Center feed is clean, it belongs in the account and will usually deserve most of the non-brand ecommerce budget. For lead generation, use it only after you set brand exclusions and confirm your conversion actions reflect qualified leads rather than form fills. Otherwise, it will find the cheapest form fill on the internet and buy a thousand of them.
Use a simple cannibalization test. Track brand Search impression share and cost per conversion for two weeks before and after PMax launches. If brand CPCs rise while PMax reports a suspiciously good CPA, you are paying PMax to take credit for demand the brand campaign already had.
I cover the asset-group and signal work in more detail here. For structure, start with one PMax campaign. Add a second only when two product groups truly need different ROAS targets.
Demand Gen gets added for the wrong reason, then killed for the wrong reason. It is not a Search substitute, and it will not hit your Search CPA. Its job is to create demand that shows up later as branded search and direct traffic. Judging it on last-click conversions is like judging a billboard by how many people pull over.
If you run Demand Gen:
If you cannot afford to leave budget in a channel that pays back over months, skip it and put the money into Search. That is not a failure of the campaign type. It is a budget-stage decision.
Turning on AI Max in a Search campaign changes where your control lives. Keywords stop being the fence and become a hint, because the campaign can serve on queries your list never contained. Your real control surface becomes account-level and campaign-level negatives, brand inclusions and exclusions, and the landing pages Google can send traffic to.
That argues for fewer, cleaner campaigns. The reason to split by keyword theme mostly disappears once matching stops depending on your keyword list. It also means a sloppy account gets expensive faster. If your negative list has not been touched since the last agency handoff, fix that before you expand matching. Not after the invoice arrives.
Everything above assumes the conversions feeding your bid strategies are real and complete. In 2026, they usually are not. The gaps have just moved.
Three pieces of plumbing carry most of the load:
Set up the first two even if tracking looks fine. The failure mode is silent. Nobody sends an alert saying 12% of your conversions stopped being attributed in March.
Offline imports are the integration that changes account structure. When a bid model optimizes toward form fills, it buys form fills. Every lead-generation account I have inherited that complained about lead quality was really complaining about what it asked for.
Feed CRM stages back into Google Ads and the account reorients. Cost per lead often rises. Cost per real customer falls. Campaigns you were about to pause turn out to be the ones producing revenue. That is also the point where tROAS becomes usable in lead generation, because you finally have values to optimize toward instead of a flat conversion count.
If your CRM cannot export a lead ID and a stage, that is the highest-value engineering ticket in your marketing backlog this quarter.
One boring habit is worth naming: review conversion actions every quarter and count how many are marked primary. Sites get redesigned. Someone adds a PDF-download tag. A developer ships a new thank-you page. Six weeks later, your target CPA is being calculated against events that mean nothing. I have found chat-widget opens marked as primary conversions in accounts where three separate people were paid to be watching.
Better structure cannot rescue a bid strategy trained on the wrong outcome.
When the system decides who sees the ad, what the ad says becomes your main steering wheel. Fill the asset slots. An RSA supports up to 15 headlines and four descriptions. I want at least 10 headlines that make genuinely different points rather than 10 rewrites of the same benefit:
Then let the combinations sort themselves out.
Do the same with the extensions people skip: four or more sitelinks pointing at pages that actually exist, callouts, structured snippets, a logo, images, and the business name. Ad Strength is a mediocre metric, and I do not chase "Excellent." But the question underneath it is real: how many distinct combinations can the system test? Ad groups with three headlines do not lose because Google punishes them. They lose because there is nothing to optimize.
The bigger gap sits downstream. I have never audited an account where landing-page coverage matched ad-group coverage. Twenty ad groups, four pages, and one generic services page catching traffic from six different intents is normal.
The cost hides inside the average. Someone searching "emergency AC repair tonight" lands on a page about heating, cooling, and indoor air quality, spends nine seconds there, and leaves. You paid $22 for that click, and your conversion rate absorbed it quietly. Message match is the cheapest conversion-rate improvement available to most accounts. It is also the one nobody does because it means asking a developer for pages, and developers have a queue.
This is the part of the job I stopped doing by hand. groas takes an existing landing page and deploys dynamic versions that adapt to the intent behind each search. A query about arm hair trimmers lands on a page about arm hair trimmers instead of a category page listing every product. It is not magic. It is the same message-match discipline a good conversion copywriter applies, run per search intent instead of per campaign and deployed without a dev ticket.
If you build pages by hand, start with your five highest-spend ad groups. Those five almost always account for more waste than the other 40 combined.
The hardest discipline in a 2026 account is not doing the work. It is avoiding work that feels productive but keeps the model relearning. Every target change, budget shift, and new keyword nudges the model back toward learning. A manager who logs in daily and makes three small improvements runs an account that never leaves its learning period.
I keep a fixed cadence instead. It makes the work manageable and makes it obvious when nothing needs doing.
Anything outside those three lists is optional. In a Smart Bidding account, optional usually means harmful.
Read that cadence again and mark the tasks that require judgment. Adding negatives for junk queries does not. Neither does spend pacing, asset rotation, catching a broken tag, or noticing that an ad group's landing page no longer matches its intent. Those are pattern-recognition tasks against data that refreshes hourly, done by a human on a weekly schedule.
The average error sits in the account for three and a half days before anyone sees it. That gap is where a lot of wasted spend lives. It is not a competence problem. It is a sampling-rate problem.
At groas, models act continuously on the same decisions a manager makes on Monday morning. That is why the customer results cluster around fast, structural changes: CPA cut 36% in the first 30 days for a supplements account spending $200k to $300k a month, and conversions up 170% with CPA down 72% in seven days for an electronics account spending $40k to $50k. Those are not clever strategies. They are the same obvious fixes, applied the moment they become obvious.
What does not get automated is the layer above the account, and I would be suspicious of anyone selling otherwise. Nobody but you decides what counts as a qualified lead, what margin you will accept, whether you can afford a channel that pays back in month three, or whether the offer on the page is actually competitive. Structure decisions with business consequences stay human too. Splitting a campaign because two product lines have different economics is a finance call wearing a Google Ads costume.
Groas draws the line by spend. Below $25k a month, it runs fully managed, with a dedicated account manager and no dashboard for you to babysit. Above that, you can take software access, keep your team on day-to-day management, and put a strategist behind them. If you spend $6k a month, another login is not what is holding you back.
That 43-campaign account ended up with five campaigns. Same budget, same offer, same landing pages for the first month. CPA fell 31% in six weeks because the bid strategies could finally see enough conversions to make a decision. The client's first reaction was to ask what else we had deleted.
Run the same check this week. Open the campaigns view, sort by conversions over the last 30 days, and count the rows sitting below 30. If most of them are, you do not have an account structure. You have a filing system, and you are paying to maintain it.
