Why Google PPC still works in 2026, and where its AI actually earns its keep

Every year someone announces that paid search is finished. The reasons change — rising costs, ad blockers, AI answers eating the results page — but the conclusion is always the same, and it is always aimed at businesses who are nervous about their spend. So rather than argue about it, let's look at what the numbers actually did this year.

The honest picture: clicks got more expensive, leads got cheaper

LocaliQ and WordStream publish the most useful open benchmark set in paid search, drawn from thousands of live Google Ads accounts across 23 industries. Their 2026 edition, covering campaigns from April 2025 through March 2026, reports these cross-industry averages:

• Click-through rate: 6.64%
• Cost per click: $5.42
• Conversion rate: 8.18%
• Cost per lead: $66.69

Two of those numbers matter far more than the other two. Cost per click is the one everybody quotes when they want to argue that Google Ads has become unaffordable, and on its own it looks damning: $5.42 today against roughly $2.32 ten years ago. Clicks have more than doubled in price.

But cost per lead over the same decade went from about $59.18 to $66.69. That is a rise of roughly 13% across ten years — slower than inflation in most of the economy, and slower than the cost of almost every other lead source you could name. And this year it did not rise at all. It fell, for the first time in five years.

None of that means every account is healthy. Averages hide enormous spread. Attorneys and legal services pay around $9.87 a click, home and home improvement $8.33, dentists $8.00. Restaurants pay $2.05 and arts and entertainment $1.63. If you are in a high-cost vertical, a sloppy account will destroy a budget faster than it ever could in 2016. The margin for error has narrowed even as the average outcome improved.

The reason PPC still works has not changed

Search advertising works for one structural reason: it reaches people at the moment they have decided they need something. Nobody types "emergency plumber near me" out of curiosity. That intent is why 88% of searches for a local business end in a call or a visit.

Social advertising interrupts people who were doing something else. SEO reaches the same intent but on a timeline you cannot control — months, sometimes quarters. Paid search is the only channel that lets you appear in front of ready-to-buy demand this week. That is worth paying a premium for, and the premium is what you see in the CPC.

AI answers in the results page have changed how some informational searches behave, and that genuinely matters for content strategy. It has not changed what happens when someone's furnace fails in February. Commercial intent still ends in a click and a phone call.

Where Google's AI genuinely earns its keep

Google now runs most of the machinery that used to be manual. It is worth being precise about which parts of that are working, because the honest answer is: some of them, very well.

Bidding

This is where automation is strongest, because it is a narrow, data-rich problem: given everything known about this auction, this user and this moment, what is this click worth? A human cannot evaluate that in the milliseconds available. A model can.

Google reports that campaigns using Smart Bidding Exploration see 27% more unique converting users, and that it has shipped more than twenty improvements to Search and Shopping bid strategies between January 2025 and March 2026. Earlier figures put the lift at 18% more unique converting query categories and 19% more conversions.

Budget pacing

The other quietly useful piece. Campaign total budgets let you set a figure for a whole flight rather than a daily cap, and Google reports a 66% average reduction in manual budget adjustments compared with running daily budgets. For a seasonal business — an HVAC company through a heatwave, a retailer through December — this removes a genuine operational headache.

Creative assembly

Responsive ads testing combinations at a scale no human would attempt is a real gain, provided you give the system good raw material. The quality ceiling is still set by your headlines and your offer. Automation tests what you give it; it does not invent a better proposition.

Where it does not, and the tests that show it

The weakest link in Google's automation is query expansion — the system deciding which searches your ad should appear for. This is where independent testing consistently disagrees with the marketing.

In March 2026, Search Engine Land published an analysis of 23 separate tests across 16 mature advertisers looking at AI Max, Google's expanded matching layer for Search campaigns. The findings are worth reading carefully:

• Only 46% of the queries AI Max captured were genuinely new to the account. The other 54% were searches the account was already winning through existing campaigns — traffic reshuffled, not gained.
• Net incremental improvement in conversion value at the account level came out at roughly 3%.
• Where advertisers did enable text customisation, weighted Quality Score improved from 6.8 to 7.3, with ad relevance showing the clearest gain.
• Accounts using all three AI Max features together saw a 40% higher uplift in test success rates than those enabling the baseline only — the feature rewards commitment, not dabbling.

Earlier independent work was harsher still. Smarter Ecommerce, analysing more than 250 retail campaigns across 600+ accounts, found AI Max delivering conversions at roughly 35% lower return on ad spend than traditional match types. In a separate four-month test, Xavier Mantica measured AI Max at a $100.37 cost per conversion against $43.97 for phrase match — and reported it losing to every match type he compared it against.

How to make the AI produce qualified leads instead of just conversions

Here is the part most accounts get wrong, and it is the single biggest lever on lead quality.

Smart Bidding optimises toward whatever you tell it a conversion is. If your only conversion action is a form submission, the system will learn to find people who submit forms. Some of those people are tyre-kickers, students, competitors and bots. The model is not failing — it is succeeding perfectly at a goal you defined badly.

1. Define a conversion as a qualified lead, not a form fill. If your CRM knows which enquiries became real opportunities, that is the signal worth optimising against.
2. Send that signal back to Google. Offline conversion import pushes your CRM outcomes — qualified, quoted, closed — back into the account so bidding learns from revenue instead of form submissions. This is the highest-leverage change most accounts can make and the one most often skipped.
3. Value your conversions differently. A booked consultation and a newsletter signup should not carry the same weight. Conversion values let the model spend more to win the one that matters.
4. Keep the guardrails on. Review search terms weekly, build negatives aggressively, exclude your own brand where it inflates apparent performance. Expanded matching without negative discipline is how budgets leak.
5. Give it enough data to learn from. Smart Bidding needs a meaningful volume of conversions before its decisions beat sensible manual bidding. Below that, automation is guessing with confidence.
6. Match the landing page to the query. The best-targeted click in the world converts badly against a generic homepage. This is usually the cheapest fix available and the one with the fastest payback.

Do those six things and the automation works with you. Skip the second one in particular, and you will spend the year optimising enthusiastically toward the wrong outcome.

When Google Ads is the wrong answer

We would rather say this plainly than sell you a campaign that will not work.

• Nobody is searching for what you do. If you have invented a category, there is no existing demand to capture and you need to create awareness first.
• Your economics do not survive the click price. In a vertical at $9 a click, converting at 8%, a lead costs well over $100 before you have sold anything. If your margin cannot carry that, fix pricing or the offer first.
• You cannot follow up quickly. Paid search generates enquiries at the moment of intent. If those sit unanswered for two days, you are buying leads for whoever calls them back first.
• You have no budget to learn with. Automated bidding needs conversion volume before it performs. Starving a campaign guarantees the disappointing result that gets used as proof PPC does not work.

What we would actually check in your account

If you are running Google Ads now and it is not producing what you expected, the diagnosis is usually one of five things, in roughly this order of frequency:

1. Conversion tracking is measuring the wrong event, or double-counting, so every decision downstream is built on a bad number.
2. The account is optimising toward form fills with no feedback about which leads had value.
3. Search terms have drifted through expanded matching and nobody has added a negative in months.
4. Traffic lands on a page that answers a different question than the one that was searched.
5. Budget is spread thinly across too many campaigns for any of them to exit the learning phase.

None of these are exotic. All of them are common, and all of them are fixable within a reporting cycle. That is the real reason paid search still works in 2026: the channel is not broken, and the failures are nearly always specific, findable and mechanical.

The short version

Google Ads in 2026 costs more per click and less per lead than it did a year ago. The AI running underneath it is genuinely good at deciding how much to bid and how to pace a budget, and genuinely unreliable at deciding which searches you should show up for. The accounts that win are the ones that let the machine do the first job, keep a human on the second, and take the trouble to tell it which leads were actually worth having.