Sohail Kasmani. Brand and marketing strategist in London: pharma brand management at Cipla and Alembic, now marketing strategy and AI adoption at ACTAsia. MBA, Brunel University London.
AI in marketing and advertising: what works and what doesn’t
AI in marketing and advertising in 2026 mostly means machine learning deciding who sees an ad, how much to bid and which creative variant to serve, with generative AI now producing a growing share of the creative itself. The media side (bidding and targeting on Google and Meta) usually pays back if your conversion data is sound. The creative side is where the evidence says slow down.
The clearest recent evidence comes from Adam Peruta, an associate professor at Syracuse University’s Newhouse School, who co-led a study with the research firm Ipsos. They showed 3,000 US consumers 20 video ads for major brands, half made by people and half rebuilt entirely with AI from the same briefs. Viewers struggled to tell which was which, yet the AI ads were 14% weaker on short-term sales potential and 17% weaker on long-term brand equity, as he reported in Harvard Business Review on 3 September 2026.
How is AI actually used in advertising in 2026?
It helps to split the work into four jobs (targeting, bidding, creative and measurement), because AI is very good at some of them and mediocre at others, and the vendor decks tend to blur them into one glossy slide. The pillar piece on what AI in marketing is covers the wider basics.
IAB UK’s Digital Adspend study, run with Oliver Wyman, put the UK digital ad market at £40.5bn in 2025, with search at £17.9bn and social at £11.5bn (social grew 21% in the year). Both of the big platforms taking that money now switch their AI campaign types on by default (on Meta, Advantage+ is the default for new sales, app and leads campaigns), so “should we use AI in advertising” was settled a while ago. What’s still open is how much of the decision-making you hand over, and what you check afterwards.
Audience targeting, and what really changed with cookies
For five years we were told third-party cookies were about to die, and a lot of us spent budget preparing for a funeral that never happened. In April 2025 Google’s Privacy Sandbox lead, Anthony Chavez, announced that Chrome would keep its current approach to third-party cookies and would not roll out a standalone cookie prompt. In October 2025 Google went further and retired most of the Privacy Sandbox, including Topics, Protected Audience and the Attribution Reporting API, citing low adoption.
So the cookie is still alive in Chrome, but in the UK that matters less than people think, because the law never depended on the browser. The ICO’s guidance on storage and access technologies, finalised on 29 April 2026, adds the new exceptions from the Data (Use and Access) Act 2025 (statistical purposes, a site’s appearance) but still says you must get consent for tracking and profiling for online advertising. Google’s own EU user consent policy applies to UK users, but Google says it does not expect advertisers to send it a verified consent signal for UK traffic the way it does for the EEA. That is Google’s policy, not a legal exemption, so don’t let a developer read it as permission.
The practical result is that AI targeting now runs mainly on first-party data (your customer lists, your site conversions, your CRM stages) plus the platform’s own signals about its logged-in users. Meta’s developer documentation for Advantage+ audience is unusually candid about the split: location, minimum age, language and custom audience exclusions are what it calls “non-negotiable business constraints” that it never expands, and once the feature is on, delivery can reach beyond the rest of the targeting you entered. How the segments themselves get built is covered in AI customer segmentation.
Automated bidding and the campaign types doing it
Google’s Smart Bidding has four strategies (Target CPA, Target ROAS, Maximize conversions and Maximize conversion value) and sets a bid for each auction using signals such as device, location, time of day, remarketing list membership and the actual search query. Google recommends judging it over periods with at least 30 conversions, or 50 for Target ROAS. Below that it’s guessing with confidence, which to be fair is also how a lot of us do media planning.
On top of the bidding sit the campaign types:
- Performance Max runs one campaign across YouTube, Display, Search, Discover, Gmail and Maps. Search campaigns with an exact match keyword take priority over it on that exact query. After years of complaints about PMax as a black box, Google’s 2025 highlights list channel-level reporting, full search terms reporting and up to 10,000 negative keywords per campaign.
- AI Max for Search campaigns is a set of switches on a normal Search campaign: search term matching (broad match plus keywordless targeting), text customisation that writes ad copy from your landing pages and assets, and final URL expansion that picks the landing page for you. Brand inclusions and exclusions and URL exclusions are the controls. Google’s own 2025 data says advertisers who switch it on typically see 14% more conversions or conversion value at a similar CPA or ROAS, for non-retail advertisers. That is Google marking its own homework, so treat it as a hypothesis to test.
- Meta’s Advantage+ is now the default setup for Sales, App and Leads campaigns. As Meta ads consultant Jon Loomer explains, it switches on three levers together (Advantage+ audience, Advantage+ placements and Advantage campaign budget). On Meta’s Q2 2026 earnings call on 29 July, CFO Susan Li said Advantage+ end-to-end solutions had passed $75bn in annual revenue run-rate.
By coincidence, Google’s AI Max claim and Peruta’s creative penalty are both 14%, pointing in opposite directions, and only one of them comes from a test against a human-made control rather than from the company selling the product.
AI creative production and testing
Google’s Asset Studio, launched in September 2025, turns product photos into lifestyle images using Imagen 4, edits up to 100 images in bulk and builds video from images and text. Mark Zuckerberg said on the same Meta earnings call that 9 million small businesses now use at least one of Meta’s AI ad creative tools. TikTok’s Smart Creative combines your videos, text and calls to action into variants, tests them for three to five days using up to 10% of the ad group budget, then swaps out tired ones.
Peruta’s study is the useful corrective. The AI versions did best on conventional, product-led work: the AI Febreze spot (people spraying fabric around the house) and the AI Herbal Essences ad performed at least as well as the human originals on short-term sales potential. They did worst where the idea needed a joke or emotional precision. The human Chewy ad (Chewy is a US pet-supplies retailer) ages a golden retriever from puppy to senior as a sweater is pulled over his head; the AI version showed a generic dog near a Christmas tree, and it wasn’t even clear the young and old dogs were the same pet. What worries me is how easily the weaker ad gets through internal review, because it looks finished, and we only find out a year later when the brand tracker dips and we blame the category, the pricing, the media mix and probably the weather.
On disclosure, the UK position from the ASA and CAP (May 2025) is that there’s no blanket rule, the CAP Code applies however the ad was made, and you should disclose AI use where the audience would otherwise be misled. An “AI-generated” label does not rescue an AI skincare result that the product can’t deliver. Peruta also points out in HBR that the EU AI Act requires transparency for certain AI-generated or manipulated content from August 2026, which catches any UK brand also running ads in the EU. If you’re in a regulated category like medicines, the UK pharma advertising rules apply on top of all this.
Measurement
Every platform now uses AI to tell you how well its AI did, which is a bit like asking the chef to write the restaurant review. Meta says its incremental attribution model, which tries to count only conversions the ad caused, drove 24% more incremental conversions than its standard attribution model in Q4 2025, according to its January 2026 update. That’s a step in the right direction, but it’s still Meta’s model of Meta’s ads.
For a view across channels, Google made Meridian, its open-source marketing mix model, available to everyone in January 2025. It can take your incrementality experiment results as priors, so you run a geo or holdout test and use it to calibrate the model. I cover how to set this up in measuring the ROI of AI in marketing.
Can AI improve ROI in digital advertising campaigns?
Yes, on the media side, when three conditions hold: you have enough conversions (Google’s 30 a month is a floor, not a target), the conversion you feed the machine is the one you actually care about, and you’ve set hard limits on where it can go. Automated bidding can weigh signals in every auction that no human buyer could, and Google’s own 14% figure for AI Max points the same way, self-interested as it is.
On the creative side the answer is “sometimes”. AI creative saves money you can see (production) and costs money you can’t (effectiveness). A rough illustration, with invented numbers and the big assumption that Peruta’s 14% sales-potential gap translates one-for-one into revenue:
| Line | Human-made creative | AI-made creative |
|---|---|---|
| Media spend | £100,000 | £100,000 |
| Production cost | £30,000 | £10,000 |
| Revenue attributed | £400,000 | £344,000 (14% lower) |
| Revenue minus media and production | £270,000 | £234,000 |
The £20,000 saving is swamped by a £56,000 revenue gap. AI creative only wins on ROI when the production saving is bigger than the revenue lost to weaker creative. On a big-media campaign that’s a high bar, while for a small always-on product ad (the Febreze type) the gap may be zero, which is exactly where AI should go first.
A guide to implementing AI for dynamic ad targeting on the major platforms
This is the order I’d do it in. The usual mistake is to skip steps one and two and go straight to pressing the Advantage+ button with great enthusiasm.
- Fix the conversion data first. Check consent is captured and passed properly, send server-side conversions (Meta’s Conversions API, Google’s offline conversion imports from your CRM) and remove duplicates. The machine optimises towards whatever you tell it is a success, including your broken form-submit tag.
- Choose the right goal. If you bid on leads, you’ll get cheap leads, including the student downloading your whitepaper for an essay. Feed back qualified leads or revenue values where you can, and use value-based bidding (Target ROAS or Maximize conversion value) once you have 50 conversions in the period.
- Set the hard limits before you switch anything on. On Google: brand exclusions, account-level negative keywords, URL exclusions for pages that should never be landing pages (careers, T&Cs, the old promotion that ended in 2023). On Meta: location, minimum age, language and custom audience exclusions, which Advantage+ audience won’t override.
- Give the system signals, not fences. Upload customer lists, add PMax audience signals and your starting targeting on Meta, and let broad targeting do its work inside the limits from step three.
- Supply creative variety, with a human choosing the ideas. Following Peruta’s split, use AI for versioning, resizing and product demos, and people for the emotional idea.
- Run it as an experiment. Use Google Ads experiments, Meta’s A/B testing or a geo holdout against your current setup, long enough to reach the conversion thresholds, and decide the success metric before you look.
- Review the machine weekly. The search terms report, the PMax channel report and Meta’s placement breakdown are where most first-month waste shows up.
| Platform | Main AI targeting product | What you control | What to watch |
|---|---|---|---|
| Google Search | AI Max for Search campaigns | Brand inclusions/exclusions, URL exclusions, turning off text customisation or search term matching | Irrelevant queries, generated copy making claims you can’t support, odd landing pages |
| Google (all inventory) | Performance Max | Audience signals, search themes (up to 50 per asset group), negative keywords, brand exclusions | Spend drifting to brand search and remarketing that you’d have won anyway |
| Meta | Advantage+ sales, app and leads campaigns | Location, minimum age, language, custom audience exclusions; everything else is a starting point | Lead quality on leads campaigns, existing customers soaking up “new customer” budget |
What are the best AI-driven tools for optimising pay-per-click advertising?
The most powerful PPC AI is the stuff built into the platforms, and third-party tools earn their place by giving you back control and visibility across accounts. This comparison is based on each tool’s own documentation:
| Tool | Good at | Less good at |
|---|---|---|
| Google Smart Bidding and AI Max | Auction-time bids using signals you can’t see; finding converting queries you didn’t think of | Explaining itself; judging whether a conversion was incremental |
| Google’s Ask Advisor (announced at Google Marketing Live 2026) | An AI agent across Google Ads, Analytics and Merchant Center for insights and optimisation | It’s Google’s adviser on Google’s products, so don’t expect it to suggest spending less with Google |
| Copilot in Microsoft Advertising | Chat-based diagnostics, conversion tracking checks, root-cause analysis of performance shifts, asset suggestions built from your URL | Only sees Microsoft’s own ad network, so it can’t tell you anything about your Google or Meta spend |
| Optmyzr | Rule Engine for custom automation, audits, budget monitoring, campaigns built from product feeds; covers Google, Microsoft, Amazon, Meta and LinkedIn | Rules are only as good as the logic you write |
| Adalysis | Automated ad testing, RSA asset analysis, 100+ daily audit checks | Google Ads and Microsoft Ads only, so no help on social |
| Smartly | Creative automation, dynamic templates and dynamic product ads across Meta, TikTok, Snapchat and Pinterest | Enterprise-priced and scoped; overkill unless you run lots of creative across social |
| General LLMs (Claude, ChatGPT, Gemini) | Reading an exported search terms report and clustering waste; drafting RSA headline options; writing Google Ads scripts | No live account access by default; will confidently invent a benchmark if you ask for one |
The cheapest, most useful thing on that list is the last one. A prompt I’d use on a search terms export:
Here is a CSV of 90 days of search terms from a Google Ads account for a UK [category] business. Columns: search term, match type, clicks, cost, conversions, conversion value. Our target CPA is £[X]. 1. Group the terms into themes by intent. 2. For each theme, give total cost, conversions and CPA. 3. List candidate negative keywords (themes with cost above £[Y] and no conversions), and flag any that could block good traffic. 4. Flag queries that look like competitor brands or job seekers. Do not invent any numbers that are not in the file.
Then check its maths on two or three themes by hand, because it will be right most of the time and wrong in a very convincing way occasionally.
How do AI solutions for marketing campaign optimisation compare?
In practice you’re choosing between three categories, and the choice depends more on your team than on the tech. Platform-native AI (Smart Bidding, PMax, AI Max, Advantage+) is free with your spend and has the best data, but it optimises inside one walled garden and reports on itself. Third-party tools (Optmyzr, Adalysis, Smartly) cost a subscription and add guardrails and auditing across accounts, which matters most to agencies and teams running many campaigns. Independent measurement (a marketing mix model like Meridian, or geo and holdout experiments) is the only one of the three that can tell you whether the other two are working.
If a small team can only afford one of those investments, I’d put money into the measurement before the tooling. Once a tool makes changes on your behalf rather than recommending them, it needs a different level of oversight, which generative vs agentic AI in marketing covers. I lead AI adoption in marketing at ACTAsia, a four-person team with no agency budget, and the ACTAsia case study shows what a team that size automated first.
What about AI in affiliate marketing?
Affiliate marketing gets squeezed from two sides. On the traffic side, a Pew Research Center study tracking the browsing of 900 US adults in March 2025 found people clicked a traditional search result on 8% of visits when Google showed an AI summary, against 15% when it didn’t. Review and comparison sites live on those clicks.
On the content side, the obvious response (generate a few thousand AI product round-ups) runs straight into Google’s spam policies, which name both scaled content abuse (many pages generated mainly to manipulate rankings) and thin affiliation (merchant descriptions copied with no original value). Google says affiliate pages with original reviews, rigorous testing or product comparisons are fine. So AI is useful to affiliates for the operational work (matching partners to products, spotting fraudulent traffic patterns, drafting copy that a person then tests and rewrites) and risky as a replacement for the testing that gives an affiliate page any reason to exist.
If I had to pick one thing to do this quarter, it would be to take your biggest always-on campaign, brief a human-made and an AI-made version of the creative against the same brief, and run them as a proper split test for long enough to hit the conversion thresholds above. What I’d want to know from it is whether your category behaves like Febreze or like Chewy.
FAQs
How is AI used in marketing ads day to day?
Mostly in three places: automated bidding that sets a price for each auction, audience expansion that finds people similar to your converters, and generative tools that write headlines or produce image and video variants. On Google that means Smart Bidding, Performance Max and AI Max; on Meta, Advantage+. Humans still set goals, budgets, exclusions and the core creative idea.
Are AI-generated ads as effective as human-made ones?
Not on average, according to a 2026 study by Adam Peruta of Syracuse University with Ipsos. Across 20 ads tested on 3,000 US consumers, fully AI-made versions scored 14% lower on short-term sales potential and 17% lower on long-term brand equity, even though viewers struggled to tell them apart. Simple product demonstrations were the exception.
Do third-party cookies still matter for AI ad targeting in the UK?
Less than they used to. Chrome kept third-party cookies after Google dropped its phase-out in 2025, but UK law still requires consent for advertising cookies under PECR. Platform AI now leans mainly on first-party data such as customer lists and server-side conversion data, plus signals from logged-in users on Google and Meta.
How much data does Google’s AI bidding need to work?
Google recommends judging Smart Bidding over a period with at least 30 conversions, such as a month, and at least 50 for Target ROAS. Below those levels the algorithm has little to learn from, so results swing more. Low-volume accounts can bid on a higher-funnel conversion or pool campaigns to get enough data.
Should UK advertisers disclose that an ad was made with AI?
There is no blanket UK rule. The ASA and CAP say the CAP Code applies however an ad was made, and advertisers should disclose AI use where the audience would otherwise be misled, for example with realistic synthetic people or product results. A disclosure label does not make an otherwise misleading claim acceptable.
Sources
- Harvard Business Review, Adam Peruta, “Research: AI-Generated Ads Perform Worse Than Human-Made Ones, Even When Customers Can’t Tell Them Apart”, 3 September 2026. hbr.org
- IAB UK, “Digital Adspend 2025: UK’s digital ad market reaches £40.5bn”, 3 March 2026. iabuk.com
- Google Privacy Sandbox, Anthony Chavez, “Next steps for Privacy Sandbox and tracking protections in Chrome”, 22 April 2025. privacysandbox.google.com
- Google Privacy Sandbox, Anthony Chavez, “Update on plans for Privacy Sandbox technologies”, 17 October 2025. privacysandbox.google.com
- Information Commissioner’s Office, “Guidance on the use of storage and access technologies”, finalised 29 April 2026. ico.org.uk
- Google, “Help with the EU user consent policy”, accessed September 2026. google.com
- Meta for Developers, “Advantage+ audience”, Marketing API documentation, accessed September 2026. developers.facebook.com
- Google Ads Help, “About Smart Bidding”, accessed September 2026. support.google.com
- Google Ads Help, “About Performance Max campaigns”, accessed September 2026. support.google.com
- Google Ads Help, “Use search themes with your Performance Max campaign”, accessed September 2026. support.google.com
- Google Ads Help, “Google Ads Highlights of 2025”, accessed September 2026. support.google.com
- Google Ads Help, “How AI Max for Search campaigns works”, accessed September 2026. support.google.com
- Google Ads Help, “About AI Max for Search campaigns”, accessed September 2026. support.google.com
- Jon Loomer Digital, “Advantage+ Sales, App, and Leads Campaigns are Coming”, 10 February 2025. jonloomer.com
- Meta Platforms, “Q2 2026 Earnings Call Transcript”, 29 July 2026. investor.atmeta.com (PDF)
- Meta Newsroom, “2026: AI Drives Performance”, January 2026. about.fb.com
- Google, “Asset Studio: generate creative assets with Google AI”, 10 September 2025. blog.google
- Google, “Google Marketing Live 2026: News and announcements”, 20 May 2026. blog.google
- TikTok Ads Help, “About Smart Creative”, accessed September 2026. ads.tiktok.com
- ASA and CAP, “Disclosure of AI in advertising: striking the balance between creativity and responsibility”, 29 May 2025. asa.org.uk
- Google, “Meridian is now available to everyone”, 29 January 2025. blog.google
- Microsoft Advertising, “Copilot in Microsoft Advertising Platform”, accessed September 2026. about.ads.microsoft.com
- Optmyzr, “Official Optmyzr information”, accessed September 2026. optmyzr.com
- Adalysis, “Features”, accessed September 2026. adalysis.com
- Smartly, “All features”, accessed September 2026. smartly.io
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, 22 July 2025. pewresearch.org
- Google Search Central, “Spam policies for Google web search”, accessed September 2026. developers.google.com