Quick answer: what is AI advertising in 2026?
AI advertising is the use of machine learning to decide who sees an ad, what to bid, which creative and copy to show, and how results are counted. In 2026, Meta and Google switch most of this on by default. Your job is to feed the AI profit-true signals and varied creative, then check its results honestly.
AI advertising in 2026 means the platform makes most decisions
AI advertising now means the ad platform, not your media buyer, makes most moment-to-moment decisions. It chooses who sees the ad, what to bid in each auction, which placement to use and which creative to show. On Meta and Google this is the standard way campaigns run, not an optional extra.
It helps to split AI advertising into four layers. You control each one differently, and most wasted budget comes from mixing them up.
| Layer | What the AI decides | What you still control |
|---|---|---|
| Delivery | Audience, placement, timing | Conversion goal, exclusions, countries you ship to |
| Bidding | The bid in every auction | Budget, ROAS or CPA targets, the value you send back |
| Creative | Which ad, headline and text variant each person sees | Concepts, offers, claims, brand rules, approvals |
| Measurement | Which conversions it credits to itself | Which source you trust, and holdout tests |
ROAS is return on ad spend, meaning revenue divided by ad cost. CPA is cost per acquisition, meaning ad cost divided by orders or leads.
Platform AI vs third-party tools
Most of the gain now comes from the platform layer, because only Meta and Google see every auction and every user signal. Third-party tools sit on top. They read data across Meta, Google and your store, flag problems and sometimes make changes. They cannot bid in auctions the platform does not show them.
That is why the order matters. Set up the platform defaults and the conversion signals first, then decide whether a tool adds anything. We cover that split in detail in what Advantage+ and Smart Bidding already do for free.
What is switched on by default now: a dated map
Most guides still describe AI as something you turn on step by step. In 2026 the reverse is true. Meta Sales campaigns start as Advantage+ (according to Meta's developer docs as quoted by PPC Land), and new Google Search campaigns start with AI Max selected. Here is what changed, and when.
| Date | Platform | Change | What to do |
|---|---|---|---|
| Jan 2025 | Performance Max (PMax) campaign-level negative keywords roll out, with a new-customer goal, demographic exclusions and device targeting | Add brand and irrelevant terms as PMax negatives | |
| 2025 | Meta | Meta's developer docs now refer to Advantage+ Shopping as Advantage+ Sales, and the legacy Shopping format can no longer be created through the API. | Update old naming, scripts and reports |
| 7 Aug 2025 | Shared negative keyword lists for PMax announced, after a channel performance report beta at Google Marketing Live | Use one shared list across PMax campaigns | |
| 8 Oct 2025 | Meta | Marketing API v24.0 stops new Advantage+ shopping campaigns through legacy APIs. v25.0 extends the block to every API version (from 19 May 2026). | Check any tool that builds campaigns by API |
| Feb 2026 (reported) | Meta | Manual and Advantage+ Shopping options merged into one campaign creation flow | Expect AI settings on by default in every new Sales campaign |
| 3 Mar 2026 (reported) | Meta | Trade press reports that click-through attribution counts link clicks only and that other interactions move to a 1-day engage-through bucket. | Re-baseline reported results from this date |
| Sep 2026 | Dynamic Search Ads (DSA), automatically created assets and broad match campaigns upgrade to AI Max | Review search terms and landing pages in upgraded campaigns |
The February 2026 merge comes from third-party reports. Check your own Ads Manager to see the current flow. Google's own help page confirms that on a new Search campaign AI Max is selected by default and text customisation is enabled. Google's announcement of the DSA upgrade to AI Max gives the September timing.
Defaults to review on day one
- Meta: Advantage+ audience, placements and campaign budget. Check all three, and decide on purpose.
- Meta: creative enhancements. Review what Meta may change in your ads before you publish.
- Meta: past-buyer exclusions. Practitioner guides report that the old existing-customer cap is gone, so add past buyers as an exclusion yourself.
- Google: AI Max on new Search campaigns, including text customisation and final URL expansion.
- Google: upgraded DSA campaigns. Check that they still point only at pages you want to advertise.
- Google: PMax negatives and brand terms, if you run a separate brand Search campaign.
Meta Advantage+ Sales: structure and creative supply at ₹10 lakh+ a month
According to Meta's developer docs (as quoted by PPC Land), a Sales campaign becomes Advantage+ when three automations are on: Advantage+ campaign budget, Advantage+ audience and Advantage+ placements. Turn any one off and it is still a Sales campaign, just not in the Advantage+ state.
Keep all three on unless you have a specific reason. Good reasons include a placement your creative formats cannot fill properly, or a hard audience rule such as a minimum age for your category.
Use ad sets for offers and product lines
Compared with the old Shopping campaigns, ad sets are back and you can exclude custom audiences. The existing-customer budget cap is gone, though, according to practitioner guides that cite Meta's API changes. If you used that cap before, the replacement is an ad-set-level exclusion of past buyers. You can also run a separate retention ad set with its own budget.
At ₹10 lakh to ₹1 crore a month, these splits usually make sense:
- One ad set per major product line or price band, so a ₹4,000 hero product does not starve a ₹900 accessory line.
- A separate ad set for a time-bound offer, such as a festive sale, so you can switch it off cleanly afterwards.
- Past purchasers excluded from prospecting ad sets, and targeted on purpose in a retention ad set.
Our Advantage+ Sales optimisation guide covers budgets and structure for this campaign type in more depth.
Creative is now your main targeting lever
With broad audiences, the creative decides who responds. A skincare ad showing acne scars finds a different buyer from one showing a wedding glow, even inside the same audience setting. Variety of concepts matters more than the number of near-identical versions.
Practitioner guides (such as Jon Loomer's) report that each Advantage+ Sales ad set is capped at 50 ads, while the old format allowed 150 across the whole campaign. We could not confirm this on a Meta help page. You will rarely need 50, but the cap means you should plan concepts per ad set, not dump everything into one.
Keep 8-15 live ads per ad set, spread across at least 4 distinct concepts (different hook, angle or format, not just colour swaps). At ₹30 lakh+ a month, add 3-5 new concepts a week and retire ads only after they have had a fair share of spend.
Where generative AI fits, and where humans stay
Generative tools are good at volume: resizing, background swaps, first-draft scripts and headline variants from a working ad. Keep a human on four things: product accuracy, price and offer terms, health or performance claims, and anything that looks like a customer testimonial.
A synthetic "customer" giving a review is a misleading endorsement in most markets, whether a person or a model produced it. Check your local advertising code before you run AI-generated people who appear to be real users.
Google AI Max, Performance Max and Search: when to use which
AI Max is not a new campaign type. It is an optimisation layer inside a standard Search campaign. It adds broader search-term matching, text customisation (Google writes headline and description variants) and final URL expansion (Google picks the landing page).
PMax is a separate campaign type that runs across Search, Shopping, YouTube, Display and other Google inventory from one budget.
| Setup | Best for | Watch out for |
|---|---|---|
| Search, AI Max off | Brand terms, regulated claims, tight keyword control | Missing new queries that convert |
| Search with AI Max | Generic and category searches, once tracking is clean | Off-brand text variants, wrong landing pages |
| Performance Max | Shopify catalogues with a good product feed | Brand and existing-customer traffic inflating ROAS |
Test AI Max with the 1-click experiment first
Do not switch AI Max on across every Search campaign at once. Google offers a 1-click experiment that splits traffic 50/50 between your base campaign and an AI Max version. Run it on one high-volume generic campaign and judge it on store orders, not only Google-reported conversions.
Do not run the test on a budget-limited campaign. Google itself says AI Max will not be effective if campaigns are limited by budget.
Google's internal 2025 data says advertisers who activate AI Max typically see 14% more conversions or conversion value at a similar CPA or ROAS. That figure is for non-retail advertisers. Treat it as Google's claim about Google's product, and let your own experiment decide.
What happens to your DSA campaigns
From September 2026, DSA, automatically created assets and broad match campaigns upgrade to AI Max. Open each upgraded campaign and check three things: the search terms report, which landing pages are being used, and the text variants Google is writing. Add URL exclusions for pages such as careers, blog posts or out-of-stock collections.
Keep brand and generic traffic apart in PMax
PMax now has campaign-level negative keywords, and they apply to Search and Shopping inventory. If you run a separate brand Search campaign, add your brand terms as PMax negatives. Otherwise PMax can claim cheap brand searches from people who already intended to buy, and its ROAS will look better than its real contribution.
Use the new-customer acquisition goal if most of your growth must come from first-time buyers, and read the channel performance report to see where the spend actually went.
Feed the AI profit, not platform revenue
The AI optimises towards whatever value you send it. If you send gross order value, it will chase gross order value, including tax you pass to the government and COD orders that never get delivered. Clean, net signals are the biggest lever you control.
The baseline is the Meta Conversions API (CAPI), which sends orders server-side, and Google enhanced conversions. Both help match orders the browser pixel misses. If either is not deduplicated, fix that first with our Pixel and Conversions API setup guide.
What to subtract before the value reaches the platform
- Tax: GST in India and Australia, VAT in the UK, EU and Gulf, sales tax in some US states. In India, the Gulf and the UK, shop prices usually include tax, so the order value sent to Meta usually includes it too.
- Returns: deduct your typical return rate by category, or send a refund adjustment when returns are processed.
- COD and RTO: cash on delivery (COD) is common in India, the Gulf and parts of South-East Asia. RTO (return to origin) means the parcel comes back undelivered. Either value COD orders by expected delivery rate at the time of order, or send the purchase only when the order is delivered.
- New vs existing customers: if first-time buyers are worth more to you, send a higher value or use the platform's new-customer settings.
Sending purchases only on delivery is the cleanest option, but it adds a delay of several days, which slows learning. Most brands at ₹10 lakh+ a month get a better balance by discounting COD order values by region-level delivery rates at the time of purchase.
Checking reported ROAS against real profit
Before you trust any AI's results, convert platform ROAS into profit ROAS. Our break-even ROAS guide explains the margin side in detail. Here is the full chain on one account.
Worked example: a ₹30 lakh-a-month Meta account in India
Step 1: Remove GST from reported revenue
Example: Meta reports ₹1,20,00,000 in purchase value on ₹30,00,000 of spend, a ROAS of 4.0. The products carry 18% GST in this example, and prices are GST-inclusive.
Net revenue after GST
₹1,20,00,000 ÷ 1.18 = about ₹1,01,70,000Step 2: Remove RTO and returns
Example: 40% of orders are COD, and 25% of those come back undelivered. That loses 10% of revenue. Prepaid and COD returns take another 5%. So 85% of net revenue is kept: ₹1,01,70,000 × 0.85 = about ₹86,44,500.
Step 3: Calculate true ROAS
True ROAS = ₹86,44,500 ÷ ₹30,00,000 = 2.88. If the brand cannot claim input credit on the 18% GST on its ad invoices, the real ad cost is ₹35,40,000 and true ROAS drops to 2.44.
Step 4: Compare with break-even
Example: product cost is 30% of net delivered revenue. Shipping, packaging, payment fees and RTO return shipping add another 15%. That leaves a 55% contribution margin, so break-even ROAS is 1 ÷ 0.55 = 1.82. The account is profitable, but the headroom is far smaller than a 4.0 platform ROAS suggests. Contribution after ads is about ₹47,54,000 minus ₹30,00,000, or roughly ₹17,54,000.
Tax on ad invoices, market by market
- India: advertising carries 18% GST. Google says advertisers billed by Google Asia Pacific are charged 18% GST unless they provide a valid GSTIN, and SEZ advertisers pay 0%. For invoices from Google Ireland or Meta Ireland, tax advisers recommend self-assessing 18% IGST under reverse charge each month. A ₹1,00,000 spend becomes a ₹1,18,000 invoice, so a 4.0 ROAS before tax is 3.39 against the invoice.
- UK and EU: tax guides say that with a VAT number, Meta shows no VAT on the invoice and you account for it by reverse charge, and that without one 20% UK VAT is added. Confirm on Meta's tax page for your billing country.
- Australia: 10% GST. US: sales tax applies in some states.
- Gulf, Canada and South-East Asia: rates and invoicing rules vary, so check each platform's tax page for your billing country.
Agree internally whether your team reports ROAS before or after ad tax, and stick to it. Mixing the two makes month-on-month comparisons meaningless.
Measuring AI campaigns honestly after March 2026
Meta changed how it counts conversions on 3 March 2026 (as reported by Search Engine Land and other trade press), so reported results moved without any change in real sales. Click-through attribution now counts genuine link clicks only. Likes, shares, saves and comments moved into a new engage-through bucket.
Industry coverage (Search Engine Land, CMO Tech) reports that engage-through counts conversions within 1 day after a non-link interaction, or after someone watches at least 5 seconds of a video ad, and that billing did not change. We could not confirm these details on a Meta help page, so check the attribution settings in your own Ads Manager.
Re-baseline before you compare periods
If you compare April 2026 click-through conversions with January 2026, you are comparing two different definitions. Pick a fresh baseline from March onwards. Look at click-through and engage-through as separate columns, and give engage-through less weight when you judge prospecting ads.
When you compare Meta with GA4 (Google Analytics 4) or Shopify, the link-click-only definition is now closer to what GA4 sees. Gaps will remain, because GA4 cannot see view-through or engage-through journeys. Store orders remain your source of truth for revenue.
Check the leaks the AI cannot fix
Compare landing-page views with link clicks to spot slow or broken pages. Then compare checkouts started with purchases. In India, UPI (Unified Payments Interface) checkouts can lose buyers when they switch to their payment app and do not come back. No bidding algorithm fixes a checkout leak, so do not blame the campaign for it.
Test incrementality, not just attribution
Attribution tells you who the platform credits. Incrementality tells you what would not have happened without the ads. The simplest test at this spend level is a regional holdout. Pause or sharply cut spend in a few comparable regions for 2-3 weeks, outside festive peaks, and compare store orders against regions that kept running.
Treat synthetic-audience pretesting, where AI predicts how people will react to an ad, as a rough filter for ideas. It is not a replacement for live spend.
Our rule of thumb before scaling any AI campaign
- Signal check: platform-reported purchases over 7 days should be within about 20% of store orders from that channel. If not, fix tracking first.
- Profit check: convert reported ROAS to true ROAS after tax, returns and RTO, as in the worked example.
- Buffer check: true ROAS should be at least 1.2 times break-even ROAS for 7 days in a row.
- Customer check: confirm past buyers and brand searches are excluded, or measured separately.
- Scale step: raise budget by 15-20% at a time, at least 3-4 days apart, and repeat the checks after each step.
This is the order ScaleWin's assistant follows before it suggests a change. It pulls the last 7-10 days and breaks results down by age, gender, placement, device, region, hour, ad set and ad, then checks learning-phase status and page drop-off. You can run the same checks by hand from Ads Manager exports.
What most guides get wrong about AI advertising
Most ranking guides were written before the 2026 changes, or they repeat each other's numbers. These are the errors that cost real money at ₹10 lakh+ a month.
- "Start with automated bidding and add automation slowly." On Meta and Google the AI is already on when you create the campaign. The work is deciding what to switch off, and why.
- "Run Advantage+ Shopping with an existing-customer cap." Meta's developer docs now call the format Advantage+ Sales, practitioner guides report that the cap is gone, and the workaround is to exclude past buyers at ad-set level.
- "One flat campaign with up to 150 ads." Ad sets are back, with up to 50 ads each according to practitioner reports.
- "PMax has no negative keywords." Campaign-level negatives reached all advertisers in 2025, and shared negative lists followed.
- "Meta and GA4 click-through numbers are like for like." Not since March 2026, according to trade press reports that click-through now means link clicks only.
- "AI delivers 28% higher ROI" or "67% higher ROAS." These figures circulate between pages without a primary source. Ignore any statistic you cannot trace to a named study or an official platform page.
Google's 14% AI Max figure has a source and a caveat: Google's internal data, for non-retail advertisers. Even sourced platform numbers describe averages across many accounts. Your own split test is the only number that applies to your account.
Where to start: a 90-day plan by spend tier
The first 30 days are an audit, days 31-60 are controlled experiments, and days 61-90 are for scaling what passed. What changes by spend tier is how many tests you can run at once. The full planning method is in our profit-first AI advertising strategy for D2C brands.
₹10-30 lakh a month
Fix CAPI, enhanced conversions and net-value signals first. Then run one experiment at a time: for example, the AI Max 50/50 test on your largest generic Search campaign, or a past-buyer exclusion on Meta prospecting. Keep Meta to a few ad sets so each gets enough conversions to learn.
₹30 lakh to ₹1 crore a month
Run one Meta test and one Google test in parallel. Add a regional holdout once a quarter. Build a weekly creative pipeline, because creative supply usually becomes the limit before budget does.
₹1 crore+ a month
Run holdouts by market, report ROAS before and after tax per country, and send margin-weighted conversion values where your catalogue margins vary widely. At this level, a small error in signal value is worth more than most bid changes.
Avoid starting big tests during Diwali, Ramadan or BFCM (Black Friday and Cyber Monday). CPMs (cost per thousand impressions) rise and buyer behaviour shifts, so results will not hold afterwards. ChatGPT ads are a new surface. An agency source reports a self-serve OpenAI Ads Manager with CPC and CPM bidding and no minimum spend as of May 2026. Confirm the details with OpenAI, and treat it as a small test budget until you can measure store orders from it.
Read next in this series
- Best AI Advertising Platforms in 2026: 10 Tools Compared for Meta and Google Ads
- AI Advertising for Service Businesses: Optimise for Bookings, Not Just Leads
- AI Advertising Mistakes That Waste Budget (and How to Avoid Them)
- AI Advertising Costs: What You Really Pay for Tools, Platforms and People
- AI Advertising Strategy: How to Build a Plan for a D2C Brand
- ROI Calculator - Free ROI formula calculator for ads or any investment
- CTR Calculator - Use the CTR formula (clicks ÷ impressions × 100) with a free calculator
- Inventory Turnover Calculator - Free inventory turnover ratio calculator
Key takeaways
- AI is now default-on. Meta's developer docs (as quoted by PPC Land) say Sales campaigns start as Advantage+, and new Google Search campaigns have AI Max pre-selected. The real decision is which defaults to switch off.
- Trade press reports that since 3 March 2026, Meta counts click-through conversions from link clicks only. Check your own Ads Manager and re-baseline ROAS before comparing periods.
- The AI optimises to the value you send it. Send revenue net of GST or VAT, returns and RTO, and value COD orders by expected delivery.
- Exclude past buyers and brand traffic, or reported ROAS will include people who would have bought anyway.
- Scale only when tax- and RTO-adjusted ROAS clears break-even with a buffer for 7 straight days.
Ai advertising: frequently asked questions
What is AI advertising?
Is Advantage+ Shopping still available in 2026?
Why did my Meta conversions drop after March 2026?
Is AI Max a new campaign type, and is it on by default?
What happens to my Dynamic Search Ads campaigns?
Can I add negative keywords to Performance Max?
How do I stop Advantage+ spending on existing customers?
How should COD orders be counted as conversions for AI bidding?
Do I need a third-party AI ad tool, or are platform tools enough?
Sources
- Google Ads Help: AI Max for Search campaigns setup
- Google Ads Help: AI Max 1-click experiments
- Google: Dynamic Search Ads upgrade to AI Max
- Google Ads Help: GST for advertisers in India
- Google Ads API: AI Max for Search campaigns
- Google Ads Help: AI Max and budget limits
- Google Ads announcement: PMax negative keyword lists and reporting
- Digital Applied: Meta click-through attribution change, March 2026
Let ScaleWin read your ad data every day
ScaleWin checks your Meta and Google Ads daily, judges every ad on profit and tells you exactly what to change.
Start free, no card



