Partly. Google's own AI already sets bids, finds searches, writes ad variations and, through Ads Advisor, suggests changes you approve. What it can't do is choose your goals, fix broken conversion tracking, judge which enquiries are worth having, or notice it's buying the wrong clicks. With about an hour a week of checks, many small advertisers can manage without an agency.
The condition that changes the answer is your conversion data. Google's bidding learns from whatever you tell it counts as a result. If your tracking counts a visit to the contact page as a "lead", the AI will get very good at buying contact-page visits. Fix tracking first; with accurate conversions, Google's automation does most of what a small-account agency used to do by hand.
What Google's AI already does inside your account
- Smart Bidding sets a bid for each auction based on how likely that search is to convert. After a change it needs time to calibrate: Google's help pages say a few conversion cycles (the time from click to conversion), and its updated guidance talks of up to about 50 conversion events or three conversion cycles.
- Performance Max runs one campaign across Google's channels, including Search, YouTube and Display, choosing where to show ads and which combinations of your headlines, images and videos to use. It now supports campaign-level negative keywords and a search terms report, which matter a lot for small budgets.
- AI Max for Search campaigns is an optional layer on ordinary Search campaigns that widens search term matching, customises ad text and can send people to the page on your site it judges most relevant. From September 2026, Google began automatically upgrading Dynamic Search Ads, automatically created assets and campaign-level broad match settings to AI Max, so check your campaign settings rather than assuming nothing has changed.
- Ads Advisor, announced in November 2025 for English-language accounts, is a Gemini-based assistant inside Google Ads that analyses performance, suggests keywords and assets, helps with policy issues, and applies changes only after you review and approve them.
- Ask Advisor is a newer conversational agent, still labelled beta, that can report on campaigns, troubleshoot and generate assets. Google's own help page says to review everything it suggests and to check billing and policy outcomes in your account settings rather than relying on it.
- Auto-apply recommendations can make certain changes without asking. It's a setting on the Recommendations page, and the History tab shows what has been applied.
For the plain-English version of how these systems decide what to show, how Meta and Google use AI to run your ads covers the mechanics.
The jobs that stay with you
| Job | Why Google's AI can't do it for you | Time |
|---|---|---|
| Decide what counts as a conversion, and what each is worth | Only you know a tasting booking is worth ten brochure downloads | 2 hours once, then quarterly |
| Check conversion tracking works | Google can't tell a real enquiry from a thank-you page reload | 30 minutes monthly |
| Review search terms and add negatives | Google's incentive is reach; yours is relevance | 15 to 20 minutes weekly |
| Set the budget ceiling and seasonal changes | Recommendations tend to suggest spending more | 10 minutes weekly |
| Feed back lead quality | Google sees the form, not whether the customer booked | 30 minutes monthly |
| Approve AI-written headlines and images | Generated text can make claims you can't back up | 10 minutes weekly |
| Improve the landing page and offer | Bidding can't rescue a page that doesn't persuade | A few hours a quarter |
Get the conversion setup right before anything else; setting up conversion tracking before you let AI spend your ad budget walks through it.
A catering company moves its ads in-house
Take an illustrative catering company that does weddings, private parties and office lunches. It spends $1,500 a month on Google Ads and pays an agency $500 a month to manage them, a fee equal to a third of its media budget. The account produces about 30 enquiries a month at $50 each, but the owner reckons one in five is junk: job seekers, students, people wanting a $5-a-head buffet for 200.
Month one was preparation. The owner asked the agency to confirm the business owned the ad account (it did, which made leaving simple), then gave notice. She checked the conversion setup and found the thank-you page was counted every time it loaded, so some enquiries were counted twice. She fixed that, gave wedding and corporate enquiries a higher conversion value than party enquiries, and switched off auto-apply for budgets, bidding and keywords.
Months two and three were routine: 45 minutes every Monday, using the checklist further down. The illustrative outcome after three months looked like this:
| With agency | DIY with Google's AI | |
|---|---|---|
| Media spend per month | $1,500 | $1,500 |
| Management cost per month | $500 fee | About 3.5 hours of the owner's time |
| Enquiries per month | 30 | 27 |
| Junk enquiries | About 6 | About 3 |
| Useful enquiries | About 24 | About 24 |
Fewer enquiries, the same number worth having, and $1,500 a quarter saved against roughly 45 hours of the owner's time over a year. Whether that's a good trade depends on what her hours are worth and whether she enjoys it. It was also possible only because the account was small and simple: two Search campaigns, one region, no product feed.
Put a value on each kind of enquiry
Telling Google that every enquiry is worth the same is the most common reason small accounts buy the wrong leads. The fix is a quick sum per enquiry type: the share that turn into customers × the margin a customer brings in over a year. The caterer's illustrative figures:
| Enquiry type | Share that book | Margin per customer in a year | Value to give Google |
|---|---|---|---|
| Wedding | 1 in 6 | $2,400 (one event) | $400 |
| Corporate lunches | 1 in 3 | $1,500 (about five orders) | $500 |
| Private party | 1 in 4 | $450 (one event) | $110 |
The surprise was corporate lunches. The owner had thought of weddings as the prize, but repeat office orders made a corporate enquiry the most valuable lead in the account. Once values were set and the bid strategy optimised for conversion value, spend drifted towards corporate and wedding searches without her adjusting a single keyword bid. That is the point of Smart Bidding: it will chase whatever you tell it matters, so tell it accurately. Revisit the values every quarter using real booking rates from your own records.
Search terms, before and after
The single most valuable weekly job is reading the search terms report, which shows the actual searches your ads appeared for. Here's an illustrative extract from the catering account's first review:
| Search term | Clicks | Cost | Decision |
|---|---|---|---|
| wedding catering prices | 14 | $58 | Keep |
| corporate lunch catering for 30 | 6 | $27 | Keep, and write an ad for it |
| catering jobs weekend | 9 | $31 | Add "jobs" as a negative |
| how to start a catering business | 7 | $22 | Add "how to start" as a negative |
| catering course | 4 | $15 | Add "course" as a negative |
| cheap buffet catering | 11 | $36 | Add "cheap" as a negative |
| vegan wedding catering | 3 | $14 | Keep: a real, profitable niche |
$104 of that week's spend went on searches that could never become customers. A starter negative list for most service businesses includes jobs, careers, salary, course, training, how to, free, cheap, DIY and template. Add your own as they appear.
Use ChatGPT or Claude as your analyst, not your operator
General AI assistants are good at reading exported reports and spotting patterns; they shouldn't be clicking around your live account. Export the search terms report as a spreadsheet and paste it in with this prompt:
I run Google Ads for a [business type] that sells [services], to
[type of customer], with typical orders of $[value]. We do NOT offer:
[things you don't do].
Below is last week's search terms report (search term, clicks, cost,
conversions). Please:
1. Group the terms into: clearly relevant / clearly irrelevant / unsure.
2. For the irrelevant ones, suggest negative keywords (single words or
short phrases) and say which terms each would block.
3. Flag any negative that might also block good searches.
4. List relevant terms with conversions that deserve their own ad.
Do not suggest raising the budget.
[paste report]An illustrative reply: "Irrelevant: 'catering jobs weekend', 'catering course', 'how to start a catering business'. Suggested negatives: jobs, course, how to start. Caution: 'cheap' would also block 'cheap wedding catering ideas', which may be a price-conscious couple rather than a bad lead; consider 'cheap buffet' instead. Also consider excluding 'vegan', which appears in 3 terms with no conversions." The first suggestions were sound and the caution about "cheap" was a good catch. The owner ignored the last one: vegan weddings are a niche she wants, and three clicks is far too few to judge. That's the pattern to expect: useful sorting, with judgement calls left to you.
A 45-minute weekly routine
- Spend check (5 minutes). Compare last week's spend with your plan. Google can spend up to twice your average daily budget on a given day, but not more than 30.4 times the daily budget over a month, so look at the trend rather than one day.
- Conversions check (5 minutes). Do the enquiries Google reports roughly match the enquiries in your inbox? A big gap means tracking trouble.
- Search terms (15 minutes). Add negatives, note good terms.
- Recommendations and Advisor (10 minutes). Read each one. Accept what fits your goals; dismiss the rest. Treat "raise your budget" as a question, not advice.
- Assets (5 minutes). Read any new AI-generated headlines or descriptions. Remove anything that claims what you can't prove.
- Notes (5 minutes). Write down what you changed and why, so you don't judge results from the middle of a learning period you caused.
Once a month, also compare enquiries against actual bookings and adjust conversion values if one type of enquiry is turning into far more work than another.
Triaging a week of Advisor suggestions
Recommendations and Advisor suggestions are useful, but they aren't neutral: many of them increase reach or spend. Here's an illustrative Monday list from the catering account and what the owner did with each.
- "Raise your budget by 30%: your campaign is limited by budget." Declined. Being budget-limited is a normal state for a small account with a fixed marketing budget. She'd consider it only if cost per useful enquiry was well below target for a month.
- "Add broad match versions of your keywords." Declined for now. Broad match can work with good conversion values and plenty of data; with 27 enquiries a month it mostly meant more searches like "catering jobs".
- "Add these five headlines." Accepted two, edited one, rejected two. One rejected headline said "Same-Day Catering Available"; the business needs 48 hours' notice.
- "Fix a sitelink whose destination isn't working." Accepted immediately. It pointed to an old menu page the website had deleted.
- "Add a callout for free delivery." Declined: delivery isn't free.
Five suggestions, ten minutes, two accepted as they stood. That ratio is typical, and it's why auto-apply for anything beyond housekeeping is risky for a small advertiser.
Where DIY with Google's AI tends to go wrong
- Budgets creeping up. An illustrative butcher with online meat boxes found his daily budget had been raised twice by auto-applied recommendations. Spend rose 40% with no rise in orders.
- Paying for people already looking for you. Performance Max and broad matching can pick up searches for your own business name, which you'd often get for free. Results look great; little of it is new. Check the search terms and consider brand exclusions.
- Generated claims. An illustrative farm shop found an automatically generated headline calling it "award-winning". It isn't. Review assets weekly and remove anything untrue; keeping your brand safe when AI places your ads covers the controls.
- Constant tinkering. Every significant change can send a bid strategy back into learning. Make changes in a weekly batch, then leave it alone.
- Judging too early. Two weeks of data on a small account is mostly noise. How much to spend on AI-run ads before judging results gives the maths.
Most of the classic Performance Max errors for small accounts are collected in Performance Max mistakes small businesses make.
Leaving an agency without losing your history
If the ad account is already in your business's name, leaving is mostly a checklist. Ask the agency for these before the last day of the contract:
- Admin access confirmed on the Google Ads account, and on any linked analytics property and tag manager container.
- A written list of conversion actions: what each one counts, where its tag sits on your site, and which ones are used for bidding.
- Call tracking details. Agencies often use their own call-tracking numbers in ads and on your site.
- Any shared negative keyword lists or audiences held at the agency's manager level rather than in your account.
- The original creative files: images, videos, logos in usable formats.
- A note of anything automated: scripts, rules, auto-apply settings.
The call-tracking item is where a real problem tends to show up. An illustrative farm shop left its agency on a Friday; the agency's call-tracking numbers, still showing in the ads and on the shop's contact page, were switched off with the contract. For ten days calls from ads went to a dead line, and the shop only noticed when Christmas order enquiries dropped. Replacing those numbers with the shop's own number, or with call tracking in an account the shop controls, should be done before the handover, not after.
When an agency or specialist still earns its fee
- You can't give it an hour a week. Google's AI without anyone checking it drifts towards spending more on broader traffic.
- The account is complex: a shopping feed with hundreds of products, several regions, several languages, or ads for products with extra policy rules, such as alcohol.
- Spend is large relative to margin. When a 10% efficiency gain is worth more each month than the fee, a skilled specialist pays for themselves.
- Tracking is broken and you can't fix it. Paying someone for a one-off tracking and account audit, then running it yourself, is often the best middle path.
If none of those apply, the realistic answer is that Google's AI plus an owner who checks it every Monday can manage a small account perfectly well.
More questions about running Google Ads yourself
Should I switch off auto-apply recommendations?
For most small advertisers, yes, at least for anything that changes budgets, bidding, keywords or match types. Leave recommendations visible and review them weekly instead. Some low-risk ones, such as fixing ad assets flagged for policy issues, are reasonable to accept, but you should always know what changed. The History tab in the auto-apply settings shows what has been applied.
Is Performance Max a good idea for a small business?
It can be, if your conversion tracking is accurate and you have enough conversions for it to learn from. Without those, it tends to spend on whatever is easiest to count. Many small service businesses do better starting with a tightly themed Search campaign and adding Performance Max later, once they know which enquiries turn into sales.
Can ChatGPT or Claude log in and run my Google Ads for me?
Browser agents can click through interfaces, but handing one your ad account is a poor idea: a misread screen can change budgets or targeting. Use general AI assistants as analysts on exported reports, and make the changes yourself or through Google's own Advisor, which asks for your approval before applying anything.
What if my conversions are phone calls rather than forms?
Google Ads can count calls from ads and from your website, usually with a minimum call length you set. Pick a length that filters out wrong numbers and sales calls, and check a sample of real calls each month. If call quality varies a lot, record which calls turned into bookings and feed that back as conversion values.
Further reads
- How to Test Ad Variations With AI on a Small Budget — Test headlines and offers without wasting a small budget.
- How to Build a Landing Page With AI That Converts — The page your ads send people to matters as much as the ads.
- How to Do Keyword Research With AI for a Small Business Website — Find the searches worth paying for before you bid on them.
- AI Consultant vs Agency vs Freelancer: Which Should You Hire? — Compare the kinds of outside help if you still want some.
- Should a Small Business Use Meta Advantage+ AI Campaigns? — The same question for Meta's AI campaigns.
- Red Flags When Buying AI Marketing Software or Services — Warning signs in ad management tools and services.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Google Ads Help (learning period, auto-apply recommendations, AI Max for Search campaigns, Ask Advisor, negative keywords in Performance Max); Google announcements on Ads Advisor (Nov 2025) and the Dynamic Search Ads upgrade to AI Max; Google Ads Help on budget overdelivery.