Google Ads for small business: what to set up first, in order

Google Ads has something like forty screens of settings, and almost none of them matter in week one. What matters is a short list of decisions that are expensive to reverse: which campaign type you build, how loosely your keywords are allowed to match, what gets excluded before you spend anything, and whether the account can tell a sale from a bounce. Everything below about Google's own rules was read on Google Ads Help on September 12, 2026, and dated in the text. Where Google does not publish a number, this article says so rather than borrowing one from a benchmark table.

One thing this article is not. If you are still deciding whether to buy demand that already exists or create demand that does not, that is a different question with a different answer, and it is written up separately at aevin.dev/blog/facebook-ads-vs-google-ads. Everything from here assumes that came out on the side of Google: somebody is typing the words you sell, and the job now is to build the thing without setting fire to the first month.

Search or Performance Max, and why the answer is usually Search

These are the two campaign types a small business is realistically choosing between, and they are not variations of each other. A Search campaign shows text ads against words you chose. Performance Max, in Google's own description read on September 12, 2026, is a goal-based campaign type that gives you all Google Ads inventory — YouTube, Display, Search, Discover, Gmail and Maps — from a single campaign, combining the text, images and video you supply into combinations it assembles itself.

Google is unusually clear about the relationship between the two. The same page positions Performance Max as something you use to complement your keyword-based Search campaigns and to increase reach beyond them. Its keyword matching page spells out the priority order: if a search is identical to an eligible exact match keyword in your Search campaign, that campaign is prioritized over any Performance Max campaign, while Performance Max search themes carry the same priority as phrase and broad keywords. One footnote in that same section is worth keeping: if your Search campaign is limited by budget, Performance Max may serve on those exact terms after all.

The practical case for starting on Search is not that Performance Max is bad. It is that Search is the one you can debug. A Search campaign hands you a search terms report — the actual words people typed — and that report is where every useful decision of the first two months comes from: what to exclude, what to promote, whether the demand you assumed exists actually does. Performance Max spends across six surfaces and reports back in aggregates. It also runs on Smart Bidding, which learns from conversion history a new account does not have yet. Starting there means paying for an education you cannot read.

Search first because it is legible; Performance Max later because it is powerful. Reversing that order is the most reliable way to spend a first Google budget and learn nothing from it.

The account structure that keeps the rest debuggable

Google's hierarchy is account, then campaign, then ad group, then keywords and ads. Only one rule about it matters at the start: an ad group is a theme, not a container. The keywords in it, the ad text they trigger and the page that ad points to should all be about the same thing, so that a person who types a phrase, reads a headline and lands on a page is having one continuous conversation. When those three drift apart you cannot tell whether a bad result came from the keyword, the ad or the page, so you change all three at once and learn nothing from that either.

• –One campaign to start, not five. A small budget split across several campaigns leaves every one of them below the volume where numbers mean anything, and you end up with five inconclusive tests instead of one answer.
• –Ad groups by theme, roughly five to twenty keywords each, with their own ad and their own landing page. One keyword per ad group is a dead practice: close variant matching made the precision it was built for impossible years ago.
• –Your own brand name in its own campaign, if you bid on it at all. Brand searches convert at a rate nothing else touches, and leaving them mixed in makes every other campaign look better than it is.
• –One language per campaign. Ad text is not translated for you, and Google's location page advises keeping language and location targeting consistent with each other.

Two settings deserve to be opened deliberately rather than accepted. A Search campaign can also serve on Search Partners and on the Display Network: separate inventory with a different audience, and not what you are trying to read in month one, so check where those two switches are sitting before you launch. And location targeting has more in it than a city name. Google's location page, read on September 12, 2026, calls the "Presence or Interest" option broad geo targeting, recommends it as a best practice on Search, and names the case against it in the same breath — choose "Presence" when you only want people in your locations rather than people elsewhere who are interested in them. For a business that can only serve someone physically nearby, that second case is yours, and the difference between the two settings is the difference between a plumber's budget and a plumber's budget spent on another continent.

Match types, and why broad match is where the money goes

A keyword is not an instruction to appear on that word. It is an instruction about how loosely Google may interpret it. Google's keyword matching page, read on September 12, 2026, puts it plainly: match types dictate how closely the keyword needs to match the user's search query for the ad to be considered for the auction. There are three of them, they are set with punctuation rather than a menu, and the choice between them is most of the difference between a controlled first month and an expensive one.

That last cell is the whole problem, and it is worth stating without melodrama. Broad match is not a trap Google conceals: it is the documented default, and the same page explains exactly why it works — Smart Bidding reads the signals present at auction time and bids only where a particular search is worth it. The catch is that Smart Bidding learns from conversions, and a new account has none. So the out-of-the-box pairing on day one is the widest possible matching with the least possible intelligence steering it, and the bill for that combination arrives a fortnight later as a search terms report full of words you would never have chosen.

There is a structural point underneath it that saves work. Google notes that broader match types capture everything the narrower ones do, plus more: a phrase keyword matches every search the same keyword would match on exact. You therefore never need the same keyword listed three times in three match types. Start narrow, read what comes in, and widen deliberately once there is conversion data to steer with — as a separate decision taken weeks later, not as a setting you never touched.

Negative keywords: the list that exists before launch, not after

Negative keywords exclude searches you do not want to pay for, and on a small budget they are not housekeeping — they are the main cost control you have. The starter list is predictable enough to write before a single impression is served: jobs, hiring, careers, salary, free, cheap, DIY, "how to", tutorial, plus competitor brand names and the towns you do not serve. Every one of those describes somebody who will click and was never going to buy.

Two mechanics decide whether that list actually does its job, and both surprise people. First, Google's negative keywords page, read on September 12, 2026, states that negative keywords will not match to close variants or other expansions, and gives its own example: excluding the broad negative "flowers" stops your ad on "red flowers" but not on "red flower". Plurals, word forms and synonyms are your job, while casing and misspellings Google handles for you. Second, negative match types do not behave like their positive counterparts. Negative broad, the default here too, blocks only searches containing all your terms in any order. Negative phrase blocks the terms in that order. Negative exact blocks only the literal query with nothing added, which is narrower than most people assume when they add one straight from the report.

• –Load the starter list before the campaign is switched on, not after the first invoice explains why it was needed.
• –Read the search terms report twice a week while the campaign is new, then weekly. Junk goes to negatives; a query that actually converted gets promoted to its own exact keyword.
• –Add the singular and the plural yourself, and reach for a negative phrase when you want to block a whole class of queries rather than one literal search.
• –Expect the negative list to end up longer than the keyword list. On a tight budget that is a healthy account, not an over-managed one.

Conversion tracking, and why it comes before the budget

An account that cannot tell a sale from a bounce is an account where every other number in this article is decoration. Measurement is also what unlocks the bidding that makes Google worth running at all: Google's conversion measurement page, read on September 12, 2026, lists Maximize Conversions, Target CPA and Target ROAS as strategies available once conversions are being counted, and names where conversions come from — the website, a mobile app, phone calls from the ad or from the site, and offline imports for sales that close later. Count the real thing, a purchase or a qualified enquiry, and keep micro-events in a separate column rather than mixing them in. A bidding strategy optimizing toward "scrolled the page" will get you a great deal of scrolling.

One published number is worth planning the first fortnight around. The same page states that Smart Bidding strategies require a standard 7 to 14 day learning phase to gather sufficient data and stabilize ad delivery, and warns against frequent manual changes to budgets, targets or conversion goals during it, because those adjustments reset the learning window and delay optimization. That is the mechanical reason behind the advice to leave a new campaign alone, and it is not patience for its own sake: a well-meant tweak on day three restarts the clock, and an account that gets tweaked daily never finishes learning at all.

Quality Score: what it is, and what it is not

Quality Score is a number from 1 to 10, reported per keyword, built from three components: expected clickthrough rate, ad relevance and landing page experience. Each component is graded Above average, Average or Below average, and Google's page — read on September 12, 2026 — says the comparison is against other advertisers whose ads showed for the exact same search over the last 90 days. A dash instead of a number means there were not enough exactly matching searches to calculate one, which is common on a young account and is not a fault.

Now the part that contradicts a decade of articles about it. Google's own page states that Quality Score is not an input in the ad auction, and that it is not a key performance indicator and should not be optimized or aggregated with the rest of your data. It describes it as a diagnostic tool, full stop. The page adds that changing keyword match types will not affect Quality Score, because it is calculated from exact-search history regardless of how the keyword is set. So the widely repeated rule that a Quality Score of 8 or more cuts your cost per click by some specific percentage is describing a mechanism its own source does not claim exists. We went looking for where that percentage comes from and found only articles citing each other, so this one does not print a number for it.

Read Quality Score one keyword at a time, like a smoke alarm rather than a scoreboard: Below average on landing page experience is a page to fix, Below average on ad relevance is ad text that does not answer the query. Read it as a score to raise and you will spend a month optimizing a diagnostic.

How much money it takes to get a signal

Start with what Google actually publishes, which is less than people assume. There is no minimum daily budget and no advertised floor to start. What is documented is how a budget behaves: Google's overdelivery page, read on September 12, 2026, states that on a given day a campaign might spend up to twice your average daily budget to take advantage of traffic fluctuations, and that at the end of the month you will have spent no more than 30.4 times it. So a $20 daily budget carries a monthly ceiling of about $608, and a single $40 day inside that month is normal pacing rather than a billing error to panic about.

What Google does not publish is the number everybody actually wants: how much a campaign has to spend before its results mean anything. Nobody publishes it, because it is not a budget figure at all. It is a click figure, and it depends entirely on what a click costs in your category. The usable form of the question is whether your daily budget buys enough clicks a day to produce a readable number this month. Look your terms up in Keyword Planner, take its cost estimate, and divide. Ten clicks a day is the working floor we use — it is the threshold our own pre-launch check applies before it warns that a plan is too thin — and it is a practitioners' heuristic rather than anything Google states. At a $4 click that is $40 a day; at a $12 click it is $120 a day, and if that is out of reach the honest conclusion is about the niche rather than about the settings.

On the conversion side Google does publish a floor, and only for one strategy. Target ROAS on Search and Shopping campaigns requires at least 15 conversions in the past 30 days at the conversion tracking level, per its help page read on September 12, 2026. Target CPA has no published equivalent: the same set of pages says outright that advertisers can start using Target CPA with no conversion history. Both facts point the same way for a small account. Fifteen conversions in thirty days is a sane line below which any target you set is learning from noise, whichever strategy is willing to accept it, and it is also why budget and measurement are one conversation rather than two. Too few clicks means too few conversions, which means nothing for a target to optimize toward.

When Google Ads is honestly the wrong channel

There are four situations where the right advice is not to run Search ads at all, and nobody selling you a campaign has much incentive to say so.

• –Nobody is searching. Keyword Planner shows the monthly volume for your terms in your area, and it costs nothing to check before you spend. If it is empty, no budget conjures demand into existence, and the honest next move is a channel where demand gets created rather than bought.
• –The arithmetic does not close on paper. Profit on one sale, multiplied by the share of enquiries that become sales, is the most you can pay for a lead. If that ceiling sits below what a handful of clicks costs in your category, no campaign settings rescue it — the offer or the margin has to change. That calculation, and the two numbers people routinely confuse while doing it, is at aevin.dev/blog/cost-per-lead-vs-cost-per-acquisition.
• –The budget cannot buy enough clicks. At a $15 click and $20 a day, a campaign spends months crawling toward the volume where a decision becomes possible. Narrow the geography or narrow the service until the budget clears the floor, or accept that the channel is premature and say so out loud.
• –There is nowhere for the click to land and nothing measuring it. A search visitor arrives mid-decision with a specific question, and a homepage with a menu is not an answer to it. The craft of building the page is written up at aevin.dev/blog/landing-page-for-facebook-ads — that one was written about Meta traffic, and nearly all of it transfers, with one difference in your favour: a search visitor has already told you what they wanted, so the page can say it back to them.

Where Aevin fits, and what it will not do

Aevin is our own product, which is why this sits second from the end and why it names its limits. It is $97, $197 or $497 a month, and it connects Meta, Google Ads and TikTok in one panel, alongside a landing page builder, image generation for creative, conversions sent back to Meta, a CRM for the leads and the MAD training programme. On Google specifically, and as precisely as we can put it:

1. 1.It builds Search campaigns, and only Search campaigns. The campaign type is fixed in our code rather than chosen at runtime, with the Display Network, Search Partners and the partner search network switched off explicitly rather than left wherever they default.
2. 2.It builds them for two objectives, leads or website traffic, because those are the two a search campaign has an honest answer for. Performance Max, video and app campaigns are named in the interface and greyed out with the reason on them rather than hidden, so you can see what the product does not do yet.
3. 3.The plan it assembles covers keywords with their match types, the negative list, geo and language, the daily budget, the bid strategy and the responsive search ad. It checks that plan against Google's own limits first, so a problem comes back as a named field rather than as one flat rejection of the whole request.
4. 4.New campaigns are created paused. Switching one on is a separate decision you make on a morning you chose, which is also the only sane way to launch anything.
5. 5.On a campaign that already exists it changes exactly four things: pause, resume, the daily budget and the bid strategy. The last of those is the one thing it does on Google and does not do on Meta or TikTok.
6. 6.Nothing reaches Google without a person approving it. Every change arrives as a before and after with a button on it. And the image generation mentioned above is a Meta-side feature: a search ad is text, so on Google the creative work is headlines and descriptions rather than pictures.

The order to do it in

If you keep one thing from this article, keep the sequence: measurement, then structure, then keywords, then negatives, then budget, then bids — and then leave it alone for a fortnight. Expensive first months tend to go wrong the same way, by running that list backwards and starting with a budget. Pick the four or five phrases a customer actually types, run them on exact and phrase, load the negative list before you switch anything on, fire a test conversion and confirm it lands, and then let the search terms report tell you what the market is instead of guessing at it for a quarter.