LinkedIn
Your 2026 best-practice list is a 2019 list with AI Max pasted on
Most Google Ads best practices for 2026 will waste your time. That list you bookmarked is probably a 2019 lever-pulling routine with AI Max pasted on top. I know because I used to preach it.
Read one closely and you will see the same moves: use responsive search ads with diverse headlines, apply Smart Bidding once you have data, layer audiences, and run weekly search-term reviews. That sounds like account management. But Google already handles much of the auction-level work, and some of the controls you keep adjusting no longer work the way you think. What still moves CPA lives on the input side: what you count, what you exclude, which page gets the click, and what you are actually selling.
The auction does not take instructions from your adjustment table
Smart Bidding does not treat every bid adjustment as an instruction. On Maximize Conversions, device adjustments are ignored except for a -100% exclusion that removes the traffic entirely. On Target CPA, a device adjustment changes the target, not the bid. Set a $10 target with +40% on mobile and you have told Google your mobile target is $14, not that you want to pay 40% more per click. Location and ad-schedule percentages do not steer those auctions either: Smart Bidding works at auction time, not from your adjustment table.
AI Max pushes the same shift further. It is not a new campaign type. It is a Search setting that expands matching beyond your keywords, can generate assets, and can expand the final URL. Your landing page becomes another source of signals. An informational query such as why is my pool green can bring up a product ad because Google matches inferred intent, not just the keyword string you typed. You still set boundaries, but you are not manually choosing the bid and query for each auction. You are feeding a system that makes those choices.
Five routines I used to preach and stopped
1. Manual bid adjustments by device and hour
I used to tell clients I would fix mobile with a -30% adjustment on Tuesdays. I was wrong. Under the Smart Bidding strategies above, that percentage either gets ignored or changes the target rather than the auction bid. The device move that reliably removes traffic is -100%, which blocks it. I spent too much time nudging percentages instead of asking why mobile traffic converted poorly in the first place. Delete the adjustment table before you mistake it for a strategy; fix the signal it was compensating for.
2. Single-keyword ad groups
SKAGs were my religion in 2017. One keyword, one ad group, total control. Then Google expanded close variants toward intent, and exact match stopped being the query filter the structure depended on. Keep PPC advertising and PPC management in separate SKAGs today and they can still overlap on a query such as PPC ad campaigns. You end up maintaining negative lists to defend a separation matching no longer reliably preserves. Consolidate related intent into fewer groups. Spend the saved hour checking which queries you do not want.
3. Hand-rotating ad copy
I used to run three expanded text ads per group and pause the loser every Monday like clockwork. Responsive search ads ended that job. You can supply up to 15 headlines and 4 descriptions per ad, and Google selects combinations for auctions. Pin everything to position one because you like your phrasing and you shrink the combinations it can test. With AI Max on, Google can also generate assets from your page. Your old rotation schedule never had that inventory. Feed the system varied, honest copy; stop pretending Monday is when ad selection happens.
4. Weekly bid-target nudges
My worst habit was the 10% Tuesday tweak. CPA ran high, so I raised the target a little. CPA ran low, so I lowered it a little. It felt like steering. Mostly, it made the account harder to read: when the target keeps changing, you cannot cleanly tell whether performance changed because of your input, the auction, or the new goal. The agency report calls the activity “responsiveness.” What the deck calls optimization, I call moving the finish line during the race. Set a target you can live with long enough to judge it, then change it for a reason you can name.
5. Treating every conversion as equal
This is the habit that still costs people real money. Count every form fill the same and Smart Bidding has a reason to buy more form fills, not better customers. Turn on broader matching with that setup and the weak signal travels further. One four-month practitioner test measured $100.37 CPA for AI Max against $43.97 for phrase and $52.69 for exact; Smarter Ecommerce reported 35% lower ROAS on AI Max than standard structure across 600 accounts (the tests behind those numbers). Those comparisons are not a forecast for your account. They are a warning against handing expansion a bad definition of success. Fix what counts before you ask the system to find more of it.
What still moves CPA: four inputs, not more levers
Google does the bidding math. It cannot look at a form fill and know whether that person became a customer unless you give it that information. Give it a clean signal and it has a basis for better decisions. Give it a bad one and it can buy the wrong thing efficiently. These are the four places I would spend the time I used to spend on adjustment tables.
Conversion values that reflect the business
If every conversion has a value of 1, the bidding system cannot distinguish a $400 tire-kicker from a $40,000 buyer. A flat value signal rewards volume, even when pipeline quality falls. The fix is unglamorous tracking work: import offline CRM stages so a qualified opportunity counts differently from a raw form fill, and pass revenue values for ecommerce instead of relying only on transaction counts (why broader matching needs a useful signal).
Say you spend $20k a month on B2B search. Weight demo requests that became sales-qualified leads at $500 and raw ebook downloads at $20. You have now told the bidding system which kind of query deserves more of that budget. The numbers are an illustration, not a value table to copy. Use values your business can defend. Otherwise every lever tweak is just rearranging spend around the same bad definition of a win.
Exclusions and brand controls
As matching expands beyond the keywords you entered, negatives become more important, not less. An exact-match conversion appearing under AI Max in reporting does not, by itself, prove AI Max created an additional conversion. Practitioner reviews have also found competitor queries slipping through and relevance near 50% in some campaigns, which is why brand-plus-modifier queries deserve their own structure rather than blind faith in a toggle.
I use account-level negative lists for junk intent, mine search terms twice a month, and keep brand activity separate so expansion does not get to define the whole account. An exclusion is an input the system cannot infer from your sales pitch. If you turn on expansion without checking what it pulls in, you are volunteering to pay for the lesson.
Landing pages that match the query
Google reads your landing page for signals about queries you did not type in. AI Max can match beyond your keywords using page content and can send a click to another URL when final URL expansion is enabled. A vague “services” page that crams five offers into one paragraph gives it little help distinguishing the buyer from the browser. You can end up paying for legal-advice traffic when you sell solicitor representation.
I have seen a home-services account leave bids alone and split a generic page into three query-matched pages, with the price range and service area in the first screen. Same budget, fewer junk clicks, lower CPA in three weeks. The point is not that every account needs three pages. It is that the page tells both Google and the visitor what the click is for. Tight page, tighter traffic. Loose page, expensive ambiguity.
The offer itself
No bid strategy fixes an offer nobody wants. If your page promises a free consultation while three competitors show fixed pricing and next-day booking, Google can send the visitor, charge you for the click, and watch them leave. The offer affects whether the click converts; bidding reacts to that conversion history. I have seen accounts cut CPA 20% by adding a price range and a booking slot to the page without changing Google Ads. The bid system can only price the demand your offer creates. Before you rewrite another target, look at what a buyer can actually say yes to.
The same test exposes weak YouTube advice
This is not just a Search problem. Apply the lever-versus-input test to YouTube and the same kind of checklist starts to wobble. Demand Gen brings placements across YouTube, Shorts, Discover, Gmail, and Display into one creative setup. When distribution is handled across those surfaces, another placement-tweaking ritual is not where I would start. I would look at creative variety and the value signal.
The creative numbers are a useful reminder of what the machine cannot shoot for you. Google’s help cites a Nielsen finding that 49% of ROI comes from creative. For Demand Gen in-stream, it reports lifts associated with a human voice (+12% conversions), on-screen supers (+3%), and showing the brand in the first five seconds (+4%) (the creative inputs with numbers attached). Those figures are not a promise that adding a super fixes a weak offer. They tell you where to put the effort: make something a person can understand before they move on.
Who should ignore this argument
A tiny account does not have to pretend it has a rich bidding signal. If you run one Search campaign at $1,500 a month with 12 conversions, keep the setup simple: tight exact match around five buyer terms, one clear page, and manual CPC with a cap you check twice a month. The point is to control waste while you learn, not to build a grand automation strategy on a thin sample.
There are also manual-CPC holdouts with a reason, not a habit: grant accounts with strict CPC caps, brand defense with fixed position rules, or a test cell quarantined on purpose. But if your main account has volume and you still hand-bid because you do not trust the conversion count, manual bids are hiding a tracking problem. They are not solving it.
If you sit above roughly 30 conversions a month per bidding pool and still spend Fridays on device tables and ad rotation, you are paying for theater. groas takes the other side of that trade: continuous work on conversion values, exclusions, page relevance, and logged changes rather than a bill for routine lever-pulling. Its fee is a flat $1,499 per service with no cut of spend (see what that flat fee covers). That model has room to tell you to stop tweaking. An agency paid a percentage of spend has a harder time making the same argument.
The one-hour input audit
If you do nothing else this quarter, spend one hour here, in order. No new campaign type and no Tuesday target tweak. Check whether the signal is honest before asking Smart Bidding to do more with it.
- Confirm each primary conversion fires once and matches your CRM count.
- Set conversion values that reflect the business, not a flat 1. Equal values tell bidding to chase the easiest lead.
- Add account-level negatives for junk intent and check search terms from the last 30 days.
- Keep brand-plus-modifier queries in their own structure and check AI Max brand controls. Do not assume a setting contains every query.
- Match each active ad group to a page that answers its query in the first screen.
- Import offline qualified stages and revenue. Skip this most often, and you pay Smart Bidding to find more of the leads your sales team never wanted.
Keep pulling levers and you will get exactly what bad inputs make possible: a busy change history, a clean-looking report, and spend flowing toward the wrong conversions. Google can do the lever work. It cannot tell itself which customer paid you. Feed it that truth or keep paying it to scale your mistakes.