AI marketing is the term for marketing work in which a statistical model, rather than a person, makes the decision. The decision varies. A model may select which audience sees an advertisement, calculate what to pay for one impression, write the headline, generate the image, or assign credit for the conversion that follows. What unites the cases is delegation: a marketer supplies an objective, a budget and a set of constraints, and software determines the specifics inside them.

The label is elastic, and the elasticity causes confusion. It covers three technical generations that arrived roughly a decade apart and now operate side by side. Predictive machine learning scores and prices. Generative models produce text, images and video. Agentic systems execute multi-step workflows across platforms with limited supervision. Published adoption figures rarely specify which of the three is being counted, which is why the same industry can be described as almost fully converted or barely started depending on the question asked.

The prediction layer

The oldest layer sits inside the auction. When an advertising request is created, a bidding model reads the signals attached to it and converts a predicted outcome into a price. Google groups four strategies under Smart Bidding, its term for auction-time bidding, and describes more than 20 signals read at the moment of the auction, among them device, operating system, physical location, time of day, browser and the query itself. Meta's equivalent is packaged as a suite of automation covering audience, placement, budget and creative, where an advertiser supplies an objective, a budget, a country and assets, and the system resolves everything else.

The same machinery runs outside media buying. Customer relationship management platforms score leads, rank product recommendations and choose email send times using the same class of model. Google has built a vocabulary around the inputs those systems need, reorganising its framework into named strength categories covering data connections, content and performance.

Manual alternatives are being withdrawn. Microsoft will retire maximum cost-per-click fields on new campaigns using automated strategies from 1 October 2026, and OpenAI made an automated strategy the default in new ChatGPT Ads ad groups in mid-August 2026.

The generation layer

Generative models entered advertising platforms as asset production. Google's text customisation feature, the renamed successor to automatically created assets, completed its generative rollout in February 2024 for English-language advertisers in the United States and United Kingdom. It combines extractive techniques, pulling snippets from landing page titles and meta tags, with generative models grounded in page content. Assets are refreshed at least every 48 hours, and machine-written versions serve only when predicted to outperform advertiser copy.

Image and video followed. Google launched Asset Studio on 10 September 2025 after a beta that began on 22 August, adding generation, bulk editing and style reference controls inside Google Ads. Meta introduced branded text and image generation at Cannes Lions on 17 June 2025, letting advertisers feed logos, colours, fonts and visual styles into the system. Adobe extended GenStudio with custom models and direct activation into advertising platforms in October 2025, removing the manual file transfers that had separated creative production from campaign launch.

The agent layer

The newest layer is software that acts. Standardisation began in October 2025, when a coalition including Scope3, Yahoo, PubMatic, Swivel, Triton and Optable launched the Ad Context Protocol, an open-source interface for agents discovering inventory and activating campaigns. IAB Tech Lab published a competing roadmap on 6 January 2026 and named its umbrella initiative AAMP, for Agentic Advertising Management Protocols, on 26 February 2026. Its Agent Registry reached ten entries on 11 March 2026, every one classified as a Model Context Protocol server and none as agent-to-agent.

The specifications have kept moving. AAMP 2.3, released on 30 July 2026, added a pricing provenance field intended to stop buying agents fabricating bid prices when real market data is unavailable, alongside a vendor approval gate and audience embeddings. On 20 August 2026 IAB Tech Lab chief operating officer Shailley Singh counted thirteen functions where the two frameworks overlap, covering almost the whole arc of a media buy from the brief to the impression.

Origin and evolution

Autonomy claims are older than the current wave. Adgorithms, founded in 2010 and later renamed Albert Technologies, marketed Albert as the first fully autonomous artificial intelligence marketing platform, listed on London's AIM market in 2015, and was acquired by Zoomd in March 2022. The vendor category arrived with Salesforce Einstein, announced on 19 September 2016 and embedded across Sales Cloud, Marketing Cloud, Commerce Cloud and others, according to the company's announcement.

Trade bodies formalised the subject later. IAB Europe's AI Working Group published a 15-page whitepaper on 7 July 2025, Artificial Intelligence and Europe's Digital Advertising Frontier, with contributors from Google, Meta, WPP Media, Criteo, Verve, Adform, Scope3 and BVDW setting out applications, safety principles and policy recommendations. The organisation's first pan-European adoption survey followed on 18 September 2025.

What the numbers say

That survey found 85% of companies using AI-based tools for marketing, most commonly for targeting at 64% and content generation at 61%, from roughly 95 responses collected in five languages. Advertising technology firms reported the strongest results, with 60% citing improvements in key performance indicators, against 48% for agencies and under a third of publishers reporting higher CPMs.

Budgets moved next. Mediaocean's survey of 320 professionals in November 2025 found 54% planning to increase AI media investment against 47% for search, the first time a nascent channel had overtaken search in that series. IAB's 2026 Outlook Study, drawn from more than 200 brands and agency buyers, forecast 9.5% United States advertising growthwith two-thirds of respondents prioritising agentic systems.

Forecasts diverge on where the money lands. EMARKETER, in a report published on 4 June 2026 by principal analyst Nate Elliott, projected US AI advertising at $32.03 billion in 2026 rising to $68.25 billion by 2030, with more than 80% of it appearing beside AI-generated content rather than inside chatbots.

Limitations and disputes

Return remains unproven at the aggregate level. TransUnion research found 53% of marketers reporting meaningful return on investment from AI, with data and process readiness rated high by 36%. Typeface research in June 2026 found enterprise campaign timelines lengthening rather than shortening as adoption rose. StackAdapt research published on 19 August 2026 recorded 91% tool usage but only 6% of marketers acting on in-platform AI recommendations, a gap the research attributes to unresolved accountability rather than capability.

Accuracy is contested. WordStream research published on 10 July 2025 found 20% of AI responses to pay-per-click questions contained inaccurate information, with Google AI Overviews at 26% incorrect and Gemini at 6%, across 45 identical questions put to five systems.

Vendor projections conflict openly. PubMatic chief executive Rajeev Goel forecasts 25% of digital advertising executing autonomously by 2028 and 50% by 2030. Magnite chief executive Michael Barrett capped his 2027 expectation for protocol-based agentic spend at around $700 million, describing the market as still in discovery. Neither figure has independent corroboration.

Disclosure carries a measurable cost. IAB published version 1 of its AI Transparency and Disclosure Framework on 16 January 2026, citing New York University research showing AI labels reducing click-through rates by 31.5% alongside a 37-point perception gap between advertisers and consumers.

Regulation

Article 50 of the European Union AI Act, Regulation (EU) 2024/1689, became applicable on 2 August 2026, imposing transparency duties on providers and deployers of generative systems. According to the European Commission, systems placed on the market before that date have until 2 December 2026 to meet the machine-readable marking obligation under Article 50(2), and content generated before 2 August 2026 needs no retroactive labelling. The Commission published finalised guidelines and a Code of Practice on 20 July 2026. Google has placed responsibility for AI ad labelling entirely on advertisers, and Dutch trade body VIA Nederland has mapped four disclosure triggers for agencies, noting that the lighter creative-works regime does not cover advertising where the commercial message dominates.

What it is not

Agentic AI is one layer of AI marketing, not a synonym for it: it describes systems that plan and execute across steps, and most marketing AI still predicts or generates rather than acts. AI advertising in the EMARKETER sense means advertising inventory inside or beside AI products, a media category rather than a working method. Generative engine optimisation concerns brand visibility inside AI answers, closer to search practice than to campaign automation. Marketing automation predates all of it and follows rules a human wrote; the distinguishing feature of AI marketing is that the rule is inferred from data rather than specified.

Recent developments

Google began converting campaigns running automatically created assets or the campaign-level broad match setting into AI Max for Search on 1 September 2026, with creation of the legacy configurations switched off nine days before a 12 August developer post set out the mechanics. Agents gained live account access across the same period: Innovid added Meta campaign data to its NIVO layer through an ads MCP integration on 1 September 2026. Media agencies have started building audit logs and daily token caps as agents begin buying real campaigns, after one connected television test returned five times a $25,000 media spend.

Timeline

  • 2010: Adgorithms founded, later marketing Albert as an autonomous artificial intelligence marketing platform
  • 2015: Albert Technologies lists on London's AIM market
  • 19 September 2016: Salesforce announces Einstein across its cloud products
  • March 2022: Zoomd acquires Albert Technologies
  • February 2024: Google completes the generative rollout of automatically created assets in the United States and United Kingdom
  • 17 June 2025: Meta introduces branded generative text and image tools at Cannes Lions
  • 7 July 2025: IAB Europe publishes its AI whitepaper
  • 10 September 2025: Google launches Asset Studio
  • 18 September 2025: IAB Europe publishes its first pan-European AI adoption report
  • 15 October 2025: Ad Context Protocol launched by a supply-side coalition
  • 6 January 2026: IAB Tech Lab publishes its agentic roadmap
  • 16 January 2026: IAB publishes version 1 of its AI Transparency and Disclosure Framework
  • 28 January 2026: IAB forecasts 9.5% United States advertising growth for 2026
  • 26 February 2026: IAB Tech Lab names AAMP
  • 11 March 2026: The Agent Registry reaches ten entries
  • 4 June 2026: EMARKETER publishes its US AI advertising forecast
  • 20 July 2026: The European Commission publishes final Article 50 guidelines and Code of Practice
  • 30 July 2026: AAMP 2.3 adds a pricing provenance field
  • 2 August 2026: Article 50 of the EU AI Act becomes applicable
  • 1 September 2026: Google begins converting legacy Google Ads settings to AI Max for Search
  • 2 December 2026: Marking obligation deadline for generative systems already on the market

Summary

Who: Advertisers, agencies, advertising technology vendors and publishers, operating through platform automation from Google, Meta, Microsoft, Amazon and OpenAI, marketing clouds from Salesforce and Adobe, and standards work at IAB Tech Lab, IAB and IAB Europe.

What: Marketing in which a model rather than a person makes the decision, spanning three layers: prediction and pricing, generative asset production, and agentic execution across platforms.

When: Autonomy claims date to platforms founded around 2010 and vendor branding to Salesforce Einstein in September 2016. Generative production reached advertising platforms in February 2024. Agentic protocols emerged between October 2025 and August 2026, and the EU AI Act transparency regime became applicable on 2 August 2026.

Where: Across search, social, retail media, connected television, display, email and customer relationship management, in both open programmatic markets and closed platform environments.

Why: The volume of pricing, targeting and creative decisions in digital marketing exceeds what people can make, and platforms have steadily removed the manual alternatives. The trade-off is unsettled: adoption measured at 85% or higher sits alongside a 53% rate of meaningful reported return, documented accuracy failures, vendor forecasts that differ by an order of magnitude, and a European disclosure regime whose labels have been measured to cut click-through rates by 31.5%.