Ads in Google's AI Overviews are not a separate product you book. Eligible Search, Shopping and Performance Max campaigns can show text or shopping ads above, below or within AI-written answers, labelled Sponsored. Before spending, tell clients that reporting is aggregate only, you cannot opt in or out, and the right first step is a capped test.
That last point is the one clients mishear. They read "ads in AI Overviews" in the news and ask, "Can we get in on that?" The honest reply is that you may already be in it, you cannot see how it performed on its own, and the lever you control is matching and budget, not placement. So the work sits in three places: what you say, what you check on the account, and how you set a test that ends in a decision.
How the placements work, in plain words
Everything in this section comes from Google's own help page, "About ads in AI-powered experiences in Google Search", as of November 2026. Google updates the page, so recheck it before a client meeting.
- Which campaigns qualify. Search, Shopping and Performance Max campaigns are described as automatically eligible to serve ads within AI Overviews and AI Mode. Text ads and Shopping ads can appear. Local inventory ads and vehicle ads are described as being tested.
- Where the ads sit. Directly within or alongside the AI-generated answer, including during follow-up questions in AI Mode. All of them are labelled Sponsored. Ordinary Search ads can still appear above or below an AI Overview.
- Markets and language. Availability is limited to English and select markets. Confirm that your client's target audience is covered before you promise anything.
- How a query is matched. Google says it matches an ad either directly to the query, when commercial intent is clear, or contextually to both the query and the AI response, when commercial intent emerges from the answer.
- Which match types get which. AI-powered targeting (AI Max, broad match, Shopping, Performance Max) can use both kinds of match. Exact and phrase match keywords qualify only when the query shows direct commercial intent, and do not qualify for contextual matching on broader questions.
- Controls. You cannot target AI Overviews or AI Mode separately, and you cannot opt out. Existing guardrails such as negative keywords and asset pinning still apply, and brand exclusions are another way to limit where ads show.
- Reporting. Performance is included in aggregate in your standard campaign reports (impressions, clicks, conversions, cost). The queries also appear in the search terms report, in aggregate, not labelled by placement.
- Cost. The ads go through the same auction, with the same cost-per-click or conversion-based bidding, as standard Search ads, and Google says there is no extra charge to take part. A placement that costs nothing extra to enter is still spend, which is why the test below caps it.
The pattern is that Google treats these placements as an extension of auto-matching. The more you let the system read intent, the more often you may appear. The tighter you keep to exact and phrase match, the less likely it is. That makes the question for a client a trade-off between control and reach, not a yes or no on a new product.
What a client hears versus what is true
| What clients assume | What the help page supports |
|---|---|
| "We can buy an AI Overviews slot." | No separate buying route. Eligible campaigns can serve there; you cannot target or opt out. |
| "We need to switch on AI Max." | One of several AI-powered routes. Broad match, Shopping and Performance Max are also listed. |
| "The report will show AI Overviews results." | Aggregate within standard reports. No separate line. |
| "Our exact-match keywords already cover it." | Only for queries with direct commercial intent. |
| "It will replace our normal search ads." | Standard ads can still show above or below an AI Overview. |
| "It works for any market." | English and select markets only, with some formats still in testing. |
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThe pre-spend checklist
Run through these with the client before a penny of extra budget is approved. Each item has a reason and a way to verify it.
1. Is the market and format covered?
Check the help page for the language and market list on the day of the meeting. If the client sells vehicles or local stock, note that those formats were described as tests, not general availability.
2. Is the account clean enough to let the system read intent?
Broad and AI-powered matching amplify whatever data you feed them. Check three things: conversion tracking fires on real leads or sales, only primary actions are set as the optimisation goal, and the last 90 days of search terms do not show large volumes of irrelevant queries already getting spend. If any of those fail, fix the account first. Loosening matching on a messy account buys messy traffic.
3. Are the guardrails in place?
Because you cannot opt out, negatives and brand exclusions are your steering. Build a negative list from your own search terms history: competitor names you do not want, job-seeker phrases, free, DIY and "what is" informational patterns where relevant. Google's negative keywords handle casing and misspellings automatically, but synonyms and singular or plural forms still need adding. Also check that asset pinning is set where wording must appear exactly.
4. Are the assets worth showing?
Contextual matches put your ad next to a written answer. Headlines that read like fine print, or promise things the landing page does not deliver, look worse in that setting. Review responsive search ad assets for clarity. The tutorial on AI ad copy for responsive search ads covers building and reviewing them at scale.
5. Does the landing page answer the question the ad raises?
A person who arrives from an AI-written answer is often one or two questions into their research. The page should confirm the answer in the first screen, not restart the sales pitch. Test the page on a phone and check that it loads quickly.
6. Is there a baseline?
Without reporting by placement, your only evidence is before and after at campaign level. Record four weeks of baseline data: impressions, clicks, conversion rate, cost per conversion and impression share. Note anything else that changes during the test (seasonal peaks, promotions, price changes), because aggregate reports cannot separate causes.
7. Is there a written stop rule?
Agree in advance what ends the test: a cost per conversion more than a set percentage above baseline for two consecutive weeks, or a drop in conversion rate beyond a set threshold. Put it in the email that approves the test.
8. Who owns the decision?
The client approves the budget and the stop rule. The agency owns the reporting. Naming those roles prevents a vague "let's see how it goes" turning into spend without a decision date.
Testing contextual reach for a physiotherapy clinic
Illustration with invented figures. A two-site physiotherapy clinic spends $4,000 a month on Search for new-patient bookings. The campaign is mostly exact and phrase match around "physio for sciatica" and similar terms, with a cost per booking of $38 over the last four weeks and 105 bookings a month. The client has read about AI Overviews and asks if they should be spending there.
The agency's plan:
- Do not create new money. Pick the single ad group with the strongest booking volume, say sciatica and back pain, which accounts for $1,500 of the $4,000. Because budgets sit at campaign level, move that ad group into its own campaign for the test, so its spend can be capped and read on its own.
- Change one thing. Add broad match on that ad group's core keywords with smart bidding on booking conversions, leaving the others unchanged. Add a negative list built from the search terms report (job and training terms, "exercises at home pdf", competitor clinic names).
- Cap the exposure. Set the test campaign's daily budget so the month cannot exceed $1,725: the existing $1,500 plus 15 per cent ($225) of headroom for the wider matching. Review weekly and do not raise the cap mid-test.
- Run it for five weeks. Long enough to cover a full booking cycle and past the learning period.
- Stop rule. Pause the broad match if cost per booking in that ad group exceeds $50 for two weeks running (about 30 per cent above baseline), or if the search terms show more than a fifth of spend on irrelevant queries.
What the agency tells the client afterwards depends on the numbers, but notice what it cannot say: "AI Overviews produced X bookings". What it can say is "widening matching in this group changed cost per booking from $38 to $41 and added 9 bookings a month, and the new search terms were these". That is a decision about matching, which is what the client controls.
A second look: when exact match on purpose is the right answer
Not every account should loosen anything. A bookkeeping practice with $900 a month and a cost per lead it can barely afford will do better to keep tight control, because there is no budget for experiments that could drift. Its answer to the client can be: "You are eligible where a searcher's intent is direct. We are not chasing the contextual placements because your budget cannot absorb exploration. We will revisit when the account can afford a test." That is a perfectly respectable recommendation, and often a more honest one than a rushed test.
A sample client email you can adapt
Subject: AI Overviews ads: what we can and cannot do
Hi [first name],
Short version. Google's ads in AI Overviews are not a separate
placement we can switch on. Your eligible Search campaigns can
already appear there, depending on how queries are matched.
Google does not report them separately, so we cannot show you
"AI Overviews results" as a line.
What we can do is a controlled test. We would widen matching in
one ad group ([ad group]) for five weeks, add negatives from your
search terms, and cap the extra spend at $[amount]. We stop if cost
per [lead/booking] stays above $[amount] for two weeks.
We will compare against four weeks of baseline. If it helps, we
roll it out; if not, we revert and you keep the learnings.
Are you happy for us to go ahead?
[your name]The message does three things: it corrects the assumption about buying a slot, it makes the limits of reporting clear before there is any disappointment, and it gives the client a decision with a number in it.
Sensible test budgets by account size
There is no universal figure, so use a ratio. The aim is to buy enough data for a decision, not a perfect experiment.
| Account | Monthly spend | Approach |
|---|---|---|
| Small | Under $1,500 | No test. Keep exact and phrase match and refine negatives. Revisit when there is slack. |
| Mid | $1,500 to $10,000 | One ad group or campaign, extra spend capped at 10 to 15 per cent of that campaign's monthly budget, four to six weeks. |
| Larger | Over $10,000 | One campaign split test where the platform allows an experiment, with a pre-agreed holdout, and a weekly review. |
These bands are my rule of thumb, not Google guidance. Their purpose is to give clients a defensible number to approve and to keep the test small enough that a bad result is an annoyance, not a crisis. If you handle several accounts, AI media plans and weekly budget checks shows how a test fits into the weekly routine.
Reading the results without placement data
Since you will not see an AI Overviews row, use what you do have.
- Search terms report. Read it weekly for the test ad group. Look for long, conversational queries that were not in your keyword list. They are a sign that contextual matching is reaching new phrasing, and each can be turned into a keyword or a negative.
- Impression share and top-of-page rates. Movement in these on the test group tells you if you are appearing in more auctions. It does not tell you where.
- Conversion rate and cost per conversion. The decision metrics. Compare with the baseline and, where possible, with an untouched sibling group over the same weeks.
- Branded versus non-branded split. Make sure the apparent lift is not just more branded searches arriving for unrelated reasons.
- Organic side. If the client also wants to know about appearing organically in AI features, that is a different report with different limits. See what agencies can report from Search Console on AI Overviews.
Reading a search terms sample, line by line
Here is an invented extract from week two of the physiotherapy test, with the agency's reading beside each line. It shows what "contextual" reach looks like in practice.
Query Clicks Conv Action
physio for sciatica 14 3 Keep (exact)
why does my leg go numb when i sit 9 2 New keyword, phrase
sciatica exercises pdf free 11 0 Add negative: pdf, free
how long until sciatica goes away 8 1 Watch; informational
sciatica physio jobs 3 0 Add negative: jobs
best chair for sciatica 6 0 Add negative: chair
physiotherapy for nerve pain in buttock 7 2 New keyword, phraseThree of those queries are conversational and were not in the keyword list. Two of them converted, which is the kind of signal that justifies the test. Three are noise that a negative removes. One sits in between. The week's work is exactly this: promote what converts, exclude what does not, and keep the ratio of useful to useless spend moving the right way. If the useless share stays above a fifth after two rounds of negatives, that is the stop rule talking.
A shopping example: a small online homeware store
Another case, with invented numbers. An online homeware shop spends $6,000 a month, mostly on Shopping and Performance Max. Shopping ads are described by Google as eligible, so the shop may already be appearing. The agency's job is not to "enter" but to make sure the product feed gives the system something to match: accurate titles, descriptions that answer typical questions ("dishwasher safe", "fits a 60 cm oven"), correct availability and prices. Google's page says to keep Merchant Center feeds complete and current for Shopping ads, and that is where an agency has most influence. The test here is a feed improvement on one product category, measured by conversion rate and cost per sale over the same four-to-six-week window, not a budget increase. Many clients will find that cheaper than they feared, and it is a better use of the first hour than arguing about placements.
When the client asks if you are wasting money by waiting
This question arrives in every review. A calm, honest answer has three parts. First, eligible campaigns may already be serving, so "not doing it" is partly a false picture. Second, there is no evidence in the account that a particular spend is wasted because the placement cannot be isolated. Third, you have a plan to learn at a cost the client chose. That answer protects trust better than any promised uplift, and it keeps the conversation on the one thing you control.
How AI Overviews tests go wrong for agencies
- Loosening matching on a leaky account. A plumbing-style "emergency" campaign with poor negatives starts spending on do-it-yourself queries in AI answers. The search terms report shows it within a week. The fix is negatives first, matching second.
- Claiming credit. An agency tells a client that a rise in conversions "came from AI Overviews". With no placement report, that is a guess. Say what changed and what the numbers did.
- Changing five things at once. New bidding, new matching, new landing page and new creative in the same fortnight leaves no way to learn anything.
- Assets pinned to hide weak copy. Pinning controls wording, but if a pinned headline reads badly beside an AI answer, nothing rescues it.
- Forgetting platform drift. Google changes eligibility, formats and defaults often. Text customisation and final URL expansion, for example, default to opted out for AI Max in Search campaigns in Google's API documentation, even though Google's AI Overviews help page recommends turning them on, so read the current settings rather than assuming. Record the date of every check in your notes.
Talking about organic and paid together
Clients tend to conflate "being in AI Overviews" with "being cited in AI Overviews". Paid placement is the first, buyable only through eligibility and matching. Being cited as a source is organic and follows from content and normal search practice; Google's own guidance says AI features need no special optimisation beyond normal SEO. Keep the two conversations apart in the meeting, or the budget discussion will drift into content promises. For the local-business view of the organic side, how AI Overviews change local search is a useful companion.
A one-page summary to leave with the client
AI Overviews ads: client summary
What it is: Eligible Search, Shopping and Performance Max ads can show
in or beside AI-written answers, labelled Sponsored.
What we control: matching, bidding, budgets, negatives, assets.
What we do not control: opting in or out, or targeting the placement.
What we can report: campaign totals and the search terms report.
What we cannot report: results by placement.
Test: [ad group], [weeks] weeks, cap $[amount].
Baseline: four weeks before start.
Stop rule: [metric] above [threshold] for two weeks.
Decision date: [date], owner: [client name].If you want help drafting a test plan for a specific account, or working out where AI sits in the rest of your agency's workflow, that is what my AI implementation consultation covers; start at the contact page. The AI guide lists related tutorials.
What clients ask before approving the spend
Can we buy ads in AI Overviews on their own?
Not as a separate placement. Google's help page says you cannot target AI Overviews or AI Mode separately or opt out of them. Eligible Search, Shopping and Performance Max campaigns can serve in them, and your keywords, bidding and assets decide whether they do.
Will reporting show what AI Overviews ads earned?
Not separately. Google's help page says performance from AI Overviews and AI Mode is included in aggregate in standard campaign reports. Plan on comparing campaign totals before and after changes, not on reading a placement report.
Do we need AI Max switched on?
Not strictly. Google lists several routes, including broad match, Shopping and Performance Max, and says AI-powered targeting can match both direct and contextual queries. Exact and phrase match qualify only when the query shows direct commercial intent.
How much should a first test cost?
Enough to collect decisions, not statistics. Many accounts can test inside their existing budget by widening matching on one campaign for four to six weeks, with a hard cap such as 10 to 15 per cent of that campaign's monthly spend and a stop rule written down first.
Further reads
- AI for Meta Ads: What Agencies Hand to Advantage+ and What They Keep — The same control-versus-automation trade-off on Meta.
- Best AI Overviews Tracking Software: What a Local Business Needs — Tools that track organic AI Overviews appearances.
- AI SEO for Plumbers: Get Named by ChatGPT and Google AI Overviews — A trade-business view of AI search visibility.
- Semrush AI Visibility vs Ahrefs Brand Radar: Which Suits an Agency? — Visibility tools an agency may add for the organic side.
- Best AI Tools for Advertising Agencies: Copy Variants, Mock-Ups, Reporting — Tools for copy variants, mock-ups and reporting in an agency.
- How to Advertise on ChatGPT: Approval, Budget and Reporting for Agencies — Run ChatGPT ads for a client with sign-off, a capped test budget, tracking and a weekly report, based on OpenAI's own Ads Manager documentation.
- Using AI in Advertising: What a Small Agency Can Hand to a Tool — A task-by-task map of what a small ad agency can hand to AI, where each approval gate sits, and the jobs that stay human.
- 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 pages About ads in AI-powered experiences in Google Search and About negative keywords, the Google Ads API documentation for AI Max for Search campaigns, and Google Search Central's guidance on AI features, all checked November 2026. Google changes these placements often, so check the help page the day you plan spend.