What AI actually changed in Google Ads, and what it did not

Contents (12 sections)

• What Smart Bidding actually does
• What broad match changed
• What responsive search ads changed
• What Performance Max changed
• How ads appear inside Google AI Overviews
• What AI did not change
• Machine decisions and owner decisions
• The local-service example
• The control checklist
• When automation is useful
• The honest answer
• What to do next

Google Ads is more automated than it was a few years ago. Smart Bidding uses Google AI to adjust bids at auction time for conversion or conversion-value goals. Broad match can reach related searches. Responsive search ads assemble combinations from the headlines and descriptions you provide. Performance Max can use multiple Google inventory surfaces from one campaign.

That changed how quickly the platform can make decisions.

It did not change what a booked job is, what your margin is, whether your phone is answered, or whether the conversion data is trustworthy.

That distinction matters to a local-service owner. The machine can decide which auction to enter. It cannot decide whether the client is inside your service area, whether the work is profitable, whether the crew has capacity, or whether a form became a paying job unless you define and feed those outcomes properly.

I count conservatively: each call and form once, and nothing I cannot show.

• AI changed four things.Bidding became auction-by-auction, search matching became broader, ads are assembled from parts, and Performance Max reaches most Google inventory from one campaign.
• It still needs an accurate goal.Smart Bidding optimises toward the conversions you give it. Feed it form fills from price shoppers and job seekers, and it buys more of them.
• The owner's questions did not change.AI made Google Ads faster and broader, not accountable for your booked jobs. Guardrails, a fixed test period, and reconciliation with real work are still yours.

The four changes, stated plainly:

1. Bidding became more auction-specific. Smart Bidding uses contextual signals and conversion data to set bids in individual auctions rather than applying one simple manual rule.
2. Search expansion became more machine-assisted. Broad match can identify related queries beyond the exact words in your keyword list, especially when paired with Smart Bidding.
3. Ad assembly became more flexible. Responsive search ads test combinations of the headlines and descriptions you provide.
4. Campaign reach became more consolidated. Performance Max can access Search, YouTube, Display, Discover, Gmail, and Maps from one campaign, depending on the setup and objective.

What Smart Bidding actually does

Google defines Smart Bidding as strategies that use Google AI to improve for conversions or conversion value in every auction, which Google calls auction-time bidding. The listed strategies include Target CPA, Target ROAS, Maximize conversions, and Maximize conversion value.

The practical meaning is simple: Google estimates the likelihood and value of a selected conversion and adjusts the bid for that auction.

Google says Smart Bidding relies on conversion tracking data and contextual signals such as device and location. That does not mean the system knows your profit. It means the system can improve toward the signals and goals you configure, which is why the conversion tracking audit comes before the bid strategy, not after it.

What broad match changed

Broad match can reach extra relevant search terms beyond phrase and exact match. Google suggests using it with Smart Bidding, because the wider search set gives the bidding system more opportunities to evaluate context and intent.

That can be useful. It can also expose the difference between platform relevance and business relevance.

A roofing company may want searches for roof replacement, but not roofing jobs, roof-repair instructions, roofing material wholesalers, or employment. A plumber may want emergency drain service, but not a drain snake tutorial or a plumbing apprenticeship. A cleaning company may want recurring house cleaning, but not cleaning products or a job application.

Broad match did not make negative keywords unnecessary. It made search-term review more important.

1. Start with what you can actually sell

   The service and the location, written down before the keyword list exists.

2. Review the search terms that received traffic or conversions

   Not the keywords you added. The queries the platform matched them to.

3. Classify each term

   Relevant, irrelevant, ambiguous, or useful but misrouted. Ambiguous is a real category and it needs a decision, not a guess.

4. Add a negative only when the exclusion is supported

   By the service model and the observed data, and record why you added it. The negative keyword guide covers the layering.

5. Check what the exclusion removed

   A negative that also blocks a valuable service, location, or client type is a cost, not a saving.

The right negative list is not a universal list downloaded from someone else. It is a record of what your business does not want to pay for.

What responsive search ads changed

Responsive search ads let you provide multiple headlines and descriptions. Google Ads tests combinations and learns which are more relevant for different searches.

That is useful assembly. It is not independent strategy.

Google says responsive search ads may use up to 15 headlines and 4 descriptions, and the assets can appear in different combinations and orders. That creates a practical writing requirement: each asset should make sense on its own and when combined with the others.

Do not give the platform a pile of interchangeable claims and assume it will invent a defensible offer. Write inputs that state:

• The service offered.
• The location actually served.
• The client problem solved.
• The next step the business can handle.
• Any material qualification or limitation.
• The evidence-backed reason to choose the business.

“Fast service” is not a measurement. “Same-day service” is not safe unless the business can deliver it under defined conditions. “Licensed and insured” should be used only when the statement is accurate for the business and market.

AI can rearrange the words. You still own the words.

What Performance Max changed

Performance Max is a goal-based campaign type that can access Google inventory from one campaign, including Search, YouTube, Display, Discover, Gmail, and Maps. Google says it uses AI for bidding, budget improvement, audiences, creative, attribution, and more, based on the goals and inputs the advertiser provides.

That can expand reach beyond a keyword list. It also changes the control question.

A Search campaign often starts with a visible query and a defined service theme. Performance Max can distribute spend across more surfaces and use a wider set of signals. That can be appropriate when the business has a clear goal, reliable conversion data, usable creative, enough capacity, and a reason to seek incremental reach.

It is not automatically appropriate for every local-service account.

The old question was, “Can I see every keyword?” The better question is, “Can I control and verify the client outcome this campaign is buying?” That is the same question that decides whether Smart Campaigns are a reasonable place to start.

How ads appear inside Google AI Overviews

Google now serves Search, Performance Max, and Local Services Ads directly above, below, and inside AI Overviews across the US, UK, Canada, and other English markets.

The platform matches ads to both the user’s search query and the AI Overview summary. A homeowner asking a long question about a leaking water heater or a garage floor coating can see a local ad right inside the AI answer.

To qualify for AI Overview placements without wasting spend, a local service account needs three things:

• Verified primary conversions: Clean call and form tracking in GA4 and GTM so Smart Bidding bids on real leads, not page views.
• Specific service and location assets: Clear headlines, landing pages, and location signals that match the exact trade problem.
• Active negative keyword lists: Blocking DIY, parts, and job-seeker queries so informational AI searches do not drain the daily budget.

What AI did not change

This half is the argument. The four changes above are product facts. The six below are the reason the owner still has a job.

It did not change conversion definitions

Smart Bidding improve toward the conversion actions and values available to it. If the main conversion is a form submission and the form receives price shoppers, job seekers, spam, and out-of-area requests, the platform receives a noisy instruction.

A local-service measurement chain should keep these separate:

A machine can improve for the first event you configure. It does not automatically know that the seventh event is the one that pays the bills.

It did not change your service area

Google can use location signals. That does not give the business permission to serve every location shown in a report. Travel time, dispatch cost, licensing, territory agreements, and crew availability remain business constraints.

It did not change your capacity

If the schedule is full, increasing automated reach may buy more unbookable leads. If the team needs a particular service line, a campaign should not improve toward a different one merely because that signal is cheaper.

It did not change your margin

Google can improve for conversion value if values are provided. It does not know whether the value is revenue, gross profit, contribution, or an unverified estimate. The value definition remains yours.

It did not change the need for negative keywords

Automation can identify patterns. It can also spend on patterns that look related but cannot become a job. Search-term review remains the owner’s protection against paying for DIY, employment, education, parts, rentals, products, and out-of-area demand where those are not services offered.

It did not create a decision date

Google provides bid-strategy reports, status information, experiments, simulators, and alerts. These help evaluate platform performance. They do not decide whether the campaign paid back for the owner.

The business still needs a written spend cap, a fixed test period that reflects conversion lag, and a decision date.

Machine decisions and owner decisions

The operator’s role did not disappear. It moved upward, from changing every bid manually to designing the rules, signals, exclusions, tests, and business definitions around the machine.

The local-service example

A pest-control campaign records these events in one month: 100 ad clicks, 20 forms and calls recorded as conversions, 8 leads inside the service area, 5 qualified leads, 3 booked treatments, 2 finished treatments.

The platform may report 20 conversions. The owner has 3 booked jobs and 2 finished jobs.

Neither number is automatically wrong. They answer different questions.

The platform number describes configured activity. The business number describes verified progression. If the campaign is improved to the 20-event total, the machine may prefer the cheapest way to produce more of those events. If the campaign can receive later booked-job outcomes with reliable identifiers and appropriate consent, the owner may be able to evaluate and improve closer to the money.

The first job is not to blame the algorithm. The first job is to define what the algorithm is being paid to find.

The control checklist

When automation is useful

Automation is useful when the goal is clear, the conversion signal is trustworthy enough for the decision being made, the campaign has enough relevant activity to learn from, and the business can handle the resulting demand.

It may be reasonable to test broader matching or Performance Max when the account can answer these:

1. What is the campaign meant to produce?
2. Which conversion action is used for bidding?
3. What evidence supports that conversion action?
4. What searches or clients are outside the offer?
5. What amount can be spent before a verdict?
6. How will a booked or finished job be tied back to the ad?

If the answers are unclear, adding more automation usually increases the speed of uncertainty.

The honest answer

AI made Google Ads faster at auction decisions, broader in matching, more flexible in ad assembly, and more consolidated across inventory.

It did not make the platform accountable for your booked jobs.

The useful division of labour is straightforward. Let the machine process auction-level decisions. Give it accurate conversion signals. Keep the offer, service area, exclusions, claims, capacity, spend cap, and decision date under human control. Reconcile platform activity with the deepest business outcome you can verify.

The system is not the strategy. The structure and measurement around it are the strategy.

Your one-page monthly report

Platform activity, measurement chain, business outcome

Money out Advertising spend for the automated campaigns, with the network scope stated Impressions, clicks, and query expansions are activity. They belong on this side of the page, not in the verdict.

Measured outcomes Calls, forms, messages, and configured conversions, then qualified leads, booked jobs, finished work, and confirmed revenue The measurement chain in between is event definition, tag firing, consent, identifier continuity, attribution, deduplication, and offline import. Contribution is estimated at the client's stated margin and labelled that way.

Decision Continue, adjust, stop, or limit the conclusion Limit it when the outcome record is incomplete, and name the stage the judgement actually rests on.

An automated campaign can be efficient at producing the wrong event. I count conservatively: each call and form once, and nothing I cannot show.

What to do next

Whatever the machine now decides, the owner’s questions did not change. The fixed test period guide is the structure that keeps an AI-managed account answerable to a date and a number.

Questions owners ask

Does AI replace a Google Ads manager?

It can replace some repetitive manual work. It does not replace the need to define the offer, choose the economic objective, verify the conversion chain, review search intent, write accurate inputs, protect the service area, manage capacity, and decide whether the campaign deserves more money. A manager who only changes bids may be easier to replace. An operator who owns the measurement and the business decision is doing a different job.

Does Smart Bidding know whether a job is profitable?

No. Google says Smart Bidding uses conversion tracking data and contextual signals such as device and location to optimise for conversions or conversion value at auction time. It optimises toward the conversion actions and values you configure. Whether that value reflects contribution, revenue, or a convenient number is your definition, not the platform's.

Did broad match remove the need for negative keywords?

No. Broad match can reach related searches beyond the words in your keyword list, which makes search-term review more important rather than less. Automation can find patterns that look related but cannot become a job, such as DIY, employment, education, parts, rentals, or out-of-area demand. The exclusion list is a record of what your business does not want to pay for.

Should a local-service business run Performance Max?

Sometimes. It can be appropriate when the business has a clear goal, reliable conversion data, usable creative, spare capacity, and a reason to seek incremental reach. It is not automatically appropriate for every local-service account. Answer what conversion the campaign optimises toward, whether the service areas and exclusions are accurate, and whether the result can be compared with Search on a separate budget before you fund it.

How many assets do responsive search ads use?

Google says responsive search ads may use up to 15 headlines and 4 descriptions, and that the assets can appear in different combinations and orders. That creates a writing requirement: each asset has to make sense on its own and alongside the others, because you do not control which combination a searcher sees.

Why does the platform report more conversions than my calendar shows bookings?

Because they measure different things. A platform conversion is a configured event. A booked job is a commitment the business recorded. Neither number is automatically wrong, but if the campaign is optimised to the event total, the machine may prefer the cheapest way to produce more of those events rather than more booked work.

Does automation shorten how long a test needs to run?

No. Google provides bid-strategy reports, status information, experiments, simulators, and alerts, which help evaluate platform performance. None of them decide whether the campaign paid back for the owner. You still need a written spend cap, a fixed test period that reflects conversion lag, and a decision date.

What should stay under human control in an automated account?

Keep the money decisions under human control. That covers the offer, the conversion definition, and the service area and exclusions. It also covers the accuracy of the claims in the assets, the campaign separation by urgency and economics, and the spend cap. Keep the decision date and the reconciliation between platform activity and the deepest business outcome you can verify.