AI in marketing
Last reviewed
May 30, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v2 · 3,574 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 30, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v2 · 3,574 words
Add missing citations, update stale details, or suggest a clearer explanation.
Artificial intelligence in marketing refers to the use of machine-learning systems, statistical models, and, more recently, generative AI to plan, produce, target, deliver, and measure marketing and advertising activity. The field spans a wide range of tasks, from automated bidding in online ad auctions and product recommendation engines to copywriting, image generation, audience segmentation, customer-service chatbots, and the statistical attribution of sales to marketing spend. AI in marketing is closely tied to the broader disciplines of machine learning, natural language processing, and computer vision, applied to commercial communication and customer data.
Marketing is among the business functions where AI has been adopted most widely. In McKinsey's global "State of AI" survey conducted in mid-2025, 88 percent of respondents reported that their organizations used AI in at least one business function, up from 78 percent a year earlier, with marketing and sales among the functions most commonly cited.1 The economic context is large: the Interactive Advertising Bureau (IAB) reported that United States internet advertising revenue reached 294.6 billion dollars for full-year 2025, a 13.9 percent year-over-year increase, of which programmatic advertising, a heavily automated channel, accounted for 162.4 billion dollars.2 Most of that programmatic spend is allocated by algorithms rather than negotiated by hand.
The commercial use of algorithms to influence what consumers see predates the modern generative-AI era by roughly two decades. One of the earliest large-scale deployments was Amazon's product recommendation system. In a frequently cited 2003 paper in IEEE Internet Computing, Amazon engineers Greg Linden, Brent Smith, and Jeremy York described "item-to-item collaborative filtering," an approach that recommends products by computing similarities between items rather than between users, allowing it to scale to tens of millions of customers and products.3 The technique became a template for recommendation systems across retail and media.
Interest in recommender algorithms was amplified by the Netflix Prize, an open competition announced on 2 October 2006 that offered one million dollars to any team that could improve the accuracy of Netflix's "Cinematch" rating-prediction system by at least 10 percent. The grand prize was awarded on 21 September 2009 to the team "BellKor's Pragmatic Chaos," whose ensemble method beat the benchmark by 10.06 percent.4 The competition popularized collaborative filtering and matrix-factorization methods and drew sustained academic and industry attention to personalization.
A second strand of the history is the automation of media buying. In the mid-2000s, advertising technology firms built "ad exchanges" that auctioned individual ad impressions. Brian O'Kelley, then at Right Media, is widely credited with building one of the first such exchanges, introducing the idea of matching publishers and buyers on an impression-by-impression basis; Yahoo acquired Right Media in 2007, the same year Google acquired DoubleClick.5 Real-time bidding (RTB), in which an auction for a single impression is conducted in the milliseconds while a web page loads, became widespread around 2009 with the launch of exchanges such as Google's DoubleClick Ad Exchange.5 RTB depends on predictive models that estimate the value of an impression to a given advertiser, and it established the pattern of millisecond-scale, model-driven decision making that characterizes much of digital advertising today.
From roughly 2014 onward, deep learning improved the accuracy of click-through prediction, image recognition, and recommendation, but the outputs of these systems were largely classifications, rankings, and scores. The release of capable text and image generators, beginning with large language models and text-to-image systems in the early 2020s, shifted attention toward content creation. Generative tools made it possible to draft ad copy, produce images, and generate variations at scale. Vendor adoption followed quickly: Salesforce announced Einstein GPT, which it described as generative AI for customer relationship management, on 7 March 2023,6 and other large marketing-software vendors added generative features over the following two years. McKinsey's surveys recorded a sharp rise in the share of organizations using generative AI, from roughly a third in 2023 to a large majority by 2025.1
AI is applied across most stages of the marketing process. The table below summarizes major application areas and the kinds of models typically involved; later sections describe several in more detail.
| Application area | Typical task | Common techniques |
|---|---|---|
| Content generation and copywriting | Drafting ad copy, emails, product descriptions, images | Large language models, text-to-image models |
| Personalization and segmentation | Tailoring content and offers to individuals or groups | Collaborative filtering, clustering, supervised learning |
| Predictive analytics and lead scoring | Ranking prospects by likelihood to convert or churn | Classification and regression models |
| Programmatic ad targeting and bidding | Valuing and bidding on individual impressions | Click-through-rate prediction, reinforcement and optimization methods |
| Search and SEO | Improving visibility in search and AI answer features | Ranking models, language models |
| Email and lifecycle automation | Timing, subject lines, and content of messages | Predictive models, generative models |
| Chatbots and customer service | Answering queries and routing requests | Natural language processing, language models |
| Social media | Scheduling, listening, and creative testing | Sentiment analysis, generative models |
| Dynamic and personalized pricing | Adjusting prices to demand or to the individual | Demand forecasting, optimization |
| Marketing-mix modeling and attribution | Estimating the sales impact of marketing spend | Regression, Bayesian and causal inference |
Generative models are used to draft and vary marketing text and imagery, including advertisements, social posts, product descriptions, email bodies, and landing-page copy. Large language models can produce many variations of a message for testing, adapt tone for different audiences, and assist with translation and localization. Text-to-image models generate or edit visual assets. These tools are typically positioned as drafting and ideation aids that require human review rather than as autonomous publishers, in part because of the accuracy and rights issues discussed below.
Personalization uses behavioral and profile data to tailor content, product recommendations, and offers. The underlying methods include collaborative filtering and other recommender-system techniques, clustering for audience segmentation, and supervised models that predict which message or product a given user is most likely to respond to. Personalization is one of the oldest commercial applications of AI in marketing and remains central to e-commerce, streaming media, and content platforms.
Predictive analytics applies classification and regression models to estimate future customer behavior, such as the probability that a lead will become a customer, that a customer will churn, or that a user will respond to a campaign. Lead scoring, common in business-to-business marketing, ranks prospects so that sales and marketing effort can be concentrated where it is most likely to pay off. These models draw on customer-relationship-management data, web activity, and third-party signals.
Programmatic advertising automates the purchase of ad inventory through auctions, with most digital display advertising bought this way. In 2025, the IAB reported that programmatic accounted for 162.4 billion dollars of United States internet ad revenue, growing faster than the market as a whole.2 AI supports several steps: predicting the probability that a user will click or convert, valuing each impression, setting bids, allocating budget across inventory, and filtering invalid traffic. Two prominent examples of AI-driven, automated campaign products are Google's Performance Max and Meta's Advantage+ shopping campaigns. In both, the advertiser supplies a goal, a budget, and creative assets, and the platform's models decide which audiences, placements, and creative combinations to use across the company's surfaces.7 Industry coverage notes that these products can reduce manual workload but that their results depend heavily on the quality of the data signals provided, and that their decision making is relatively opaque to advertisers, a point of recurring criticism.7
The integration of generative AI into search engines has changed how marketers approach discoverability. Search engines now display AI-generated summaries above traditional links for many queries, which can answer a user's question without a click to the underlying website. This has given rise to practices variously called answer engine optimization (AEO) and generative engine optimization (GEO), which aim to make content more likely to be cited or summarized by AI search features. Industry analyses have reported declines in organic click-through rates on queries where AI summaries appear, prompting publishers and brands to adapt their content strategies; the precise magnitude of the effect is contested and varies by query type and source.8 Google's own documentation advises that content created for AI features should follow largely the same principles as conventional search optimization.9
In email and lifecycle marketing, AI is used to choose send times for individual recipients, to generate and test subject lines and body content, to segment audiences dynamically, and to score recipients by engagement or purchase propensity. Most major email and marketing-automation platforms now include generative content features and predictive send-time optimization as standard capabilities. These features are typically layered on top of behavioral data held in the platform's customer database.
Chatbots and conversational agents handle customer queries, qualify leads, and route requests, and they increasingly draw on large language models to produce free-form responses rather than scripted replies. According to a Gartner survey of customer-service leaders published in December 2024, 85 percent said they expected to explore or pilot a customer-facing conversational generative-AI solution during 2025.10 Conversational systems in marketing and service contexts raise distinct risks around accuracy, because a model that produces confident but incorrect answers can create legal and reputational exposure for the business deploying it.
On social platforms, AI assists with content creation, scheduling, and the analysis of audience reaction. Sentiment analysis and social listening apply natural language processing to large volumes of public posts to track brand perception and emerging topics. Generative tools are used to produce and vary creative, and platform-side recommendation algorithms determine which organic and paid content reaches which users.
Retailers and travel companies have long used algorithmic pricing that adjusts to supply, demand, and competitor behavior. A more contested development is personalized or "surveillance" pricing, in which prices are tailored to inferred attributes of an individual shopper, such as device, location, or browsing history. In January 2025, the United States Federal Trade Commission published initial findings from a study, based on orders issued to eight companies, examining how intermediaries use detailed consumer data to set individualized prices; the agency described the practice as "surveillance pricing" and flagged consumer-protection concerns.11 The boundary between conventional demand-based dynamic pricing and individualized pricing is central to the policy debate.
Attribution is the problem of estimating how much each marketing activity contributed to sales or other outcomes. Two broad approaches coexist. Multi-touch attribution tracks individual user journeys across touchpoints, while marketing-mix modeling (MMM) uses aggregate, statistical regression of outcomes against spend by channel without relying on individual-level tracking. The deprecation of third-party tracking cookies and tighter privacy rules have renewed interest in MMM and in incrementality testing, because these methods do not depend on following individuals across the web. Modern MMM practice increasingly combines econometric regression with Bayesian and machine-learning methods.12
The techniques used in AI marketing are drawn from the wider machine-learning toolkit and applied to commercial data:
A practical constraint across these techniques is data: model quality depends on the availability and accuracy of first-party customer data, and privacy regulation increasingly shapes what data can be collected and how it can be used.
The marketing-technology market includes both dedicated AI tools and AI features added to established platforms. The following table lists several widely used examples and the vendors' own descriptions of them; capability and availability claims originate with the vendors and should be read as such.
| Tool or platform | Vendor | Description (as positioned by the vendor) |
|---|---|---|
| Jasper | Jasper | Generative writing tool aimed at marketing teams for producing campaign copy and content. |
| HubSpot AI (Breeze) | HubSpot | AI layer for HubSpot's customer platform, including a "Copilot" assistant and task-oriented "Agents," introduced at INBOUND 2024.13 |
| Einstein / Agentforce | Salesforce | Generative AI for customer relationship management, launched as Einstein GPT in March 2023 and later extended with agent products.6 |
| Adobe Firefly | Adobe | Family of generative models that Adobe positions as "commercially safe," trained on Adobe Stock, openly licensed, and public-domain content, with Content Credentials provenance metadata.14 |
| Performance Max | Automated, goal-based campaign type that places ads across Google's surfaces using the advertiser's assets and signals.7 | |
| Advantage+ | Meta | Automated shopping-campaign product that tests audience, placement, and creative combinations across Meta's apps.7 |
Many other platforms exist, including email and marketing-automation tools that have added generative content and predictive features, and a large number of specialized startups. The "Marketing ChatGPT plugins" ecosystem is one example of tooling built on top of general-purpose language models for marketing tasks.
Proponents and vendors describe several potential benefits of AI in marketing, which independent evidence supports to varying degrees:
Realized value, however, is uneven. McKinsey's 2025 survey found that although adoption was widespread, only a minority of organizations reported material financial impact from AI, and a small group of "high performers" accounted for a disproportionate share of the value, suggesting that benefits depend heavily on execution and data quality rather than on adoption alone.1
The application of AI to marketing has attracted scrutiny on several fronts.
Accuracy and hallucination. Generative models can produce fluent but false statements, sometimes called hallucinations. In marketing and customer-service contexts, an inaccurate claim about a product, price, or policy can mislead consumers and expose the business to liability, which is why such systems are commonly paired with human review and guardrails.
Brand safety and authenticity. AI-generated content raises questions about disclosure and trust, including whether audiences are told that content or endorsements were machine-generated. Synthetic media, including deepfakes, can be used for unauthorized endorsements or impersonation; conversely, brands face the reputational risk of their own ads appearing alongside harmful content placed through automated systems.
Privacy and data use. Targeting and personalization depend on consumer data, which is subject to privacy law and to growing public concern. Personalized pricing based on individual data has drawn particular criticism, as discussed above.11
Bias and discrimination. Models trained on historical data can reproduce or amplify bias, for example by delivering opportunities such as job or housing advertisements unevenly across demographic groups, which has been a recurring concern in ad-delivery research and litigation.
Ad fraud and invalid traffic. Automated buying is vulnerable to invalid traffic and fraudulent impressions; AI is used both to perpetrate and to detect such fraud.
Deceptive practices and fake reviews. Generative tools make it cheaper to produce fake reviews, testimonials, and endorsements at scale, a practice that regulators have moved to restrict (see below).
Opacity and dependence. Highly automated ad products have been criticized as "black boxes" that give advertisers limited visibility into where and how their budgets are spent, and reliance on a small number of platforms concentrates control over targeting and measurement.7
Several legal regimes bear on AI in marketing, even though few target it by name.
Consumer protection and advertising law. In the United States, the Federal Trade Commission finalized a rule on the use of consumer reviews and testimonials, announced on 14 August 2024 and effective 21 October 2024, that prohibits, among other things, creating or selling fake or AI-generated reviews and testimonials and fake indicators of social-media influence.15 The FTC has also examined "surveillance pricing" and has signaled scrutiny of deceptive or unsubstantiated AI-related marketing claims.11
Transparency and disclosure of AI content. The European Union's Artificial Intelligence Act imposes transparency obligations under its Article 50. Providers of generative systems must mark AI-generated audio, image, video, and text in a machine-readable, detectable format, and deployers must disclose deepfakes and, in certain public-interest contexts, AI-generated text, with the information given clearly at the time of first exposure. These transparency obligations are scheduled to apply from 2 August 2026.16 Such requirements directly affect marketing uses of synthetic media and AI-generated creative.
Data protection. Privacy laws, including the EU General Data Protection Regulation and various United States state privacy laws, govern the collection and use of the personal data on which targeting and personalization rely, including provisions on profiling and automated decision making. Compliance obligations under data-protection and AI-specific law can apply cumulatively to the same system.
Industry commentary in 2025 and 2026 has focused on the shift from generative content tools toward "agentic" systems that can carry out multi-step marketing tasks with limited human intervention, such as assembling campaigns, managing pipelines, and conducting outreach. McKinsey reported that a majority of organizations were at least experimenting with AI agents, though far fewer had scaled them.1 The IAB has likewise framed automated, AI-driven media buying, including emerging agent-to-agent protocols, as the next phase of programmatic advertising.2 At the same time, the measurement environment is being reshaped by the decline of third-party tracking and by privacy regulation, pushing the field toward aggregate and experimental methods of attribution. Whether the productivity and personalization gains attributed to AI translate into durable, measurable business value, and how regulation around disclosure, privacy, and pricing develops, remain open questions as of 2026.
McKinsey & Company. "The State of AI: How organizations are rewiring to capture value." (Global survey, fielded mid-2025; 1,993 respondents across 105 countries.) https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai Accessed 2026-05-31. ↩ ↩2 ↩3 ↩4
Interactive Advertising Bureau. "Digital Ad Revenue Climbs to Nearly $300B as IAB Celebrates 30 Year Anniversary" (IAB/PwC Internet Advertising Revenue Report, full-year 2025: $294.6 billion total, +13.9% year over year; programmatic $162.4 billion). https://www.iab.com/news/digital-ad-revenue-climbs-to-nearly-300b-as-iab-celebrates-30-year-anniversary/ Accessed 2026-05-31. ↩ ↩2 ↩3
Linden, G., Smith, B., and York, J. "Amazon.com Recommendations: Item-to-Item Collaborative Filtering." IEEE Internet Computing, vol. 7, no. 1, 2003, pp. 76 to 80. https://dl.acm.org/doi/10.1109/MIC.2003.1167344 Accessed 2026-05-31. ↩
"Netflix Prize." Wikipedia (competition announced 2 October 2006; $1,000,000 grand prize awarded 21 September 2009 to BellKor's Pragmatic Chaos for a 10.06% improvement). https://en.wikipedia.org/wiki/Netflix_Prize Accessed 2026-05-31. ↩
VideoWeek. "Who Invented What in Ad Tech? Part One" (history of ad exchanges and real-time bidding, including Right Media and the 2007 Right Media and DoubleClick acquisitions). https://videoweek.com/2018/06/07/who-invented-what-in-ad-tech-part-one/ Accessed 2026-05-31. ↩ ↩2
Salesforce. "Salesforce Announces Einstein GPT, the World's First Generative AI for CRM" (press release, 7 March 2023). https://www.salesforce.com/news/press-releases/2023/03/07/einstein-generative-ai/ Accessed 2026-05-31. ↩ ↩2
Comparative industry analyses of Google Performance Max and Meta Advantage+, describing how each automates targeting, placement, and creative selection and noting their opacity and dependence on input data quality. Example: "Performance Max vs Meta Advantage+." https://www.buildmvpfast.com/blog/ai-ad-targeting-meta-advantage-plus-google-performance-max-2026 Accessed 2026-05-31. ↩ ↩2 ↩3 ↩4 ↩5
Industry analysis of Google AI Overviews and the emergence of answer engine optimization and generative engine optimization, including reported declines in organic click-through rates on queries where AI summaries appear. SevenAtoms. https://www.sevenatoms.com/blog/google-ai-overviews Accessed 2026-05-31. ↩
Google. "Guidance about AI-generated content and Google Search / optimizing for AI features." Google Search Central documentation. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide Accessed 2026-05-31. ↩
Gartner. "Gartner Survey Reveals 85% of Customer Service Leaders Will Explore or Pilot Customer-Facing Conversational GenAI in 2025" (press release, 9 December 2024). https://www.gartner.com/en/newsroom/press-releases/2024-12-09-gartner-survey-reveals-85-percent-of-customer-service-leaders-will-explore-or-pilot-customer-facing-conversational-genai-in-2025 Accessed 2026-05-31. ↩
U.S. Federal Trade Commission. "Issue Spotlight: The Rise of Surveillance Pricing" (initial 6(b) study findings, published January 2025, based on orders to eight firms). https://www.ftc.gov/system/files/ftc_gov/pdf/sp6b-issue-spotlight.pdf Accessed 2026-05-31. ↩ ↩2 ↩3
Industry overview of marketing-mix modeling, attribution, and incrementality testing in a privacy-constrained, cookieless environment, including the combination of regression, Bayesian, and machine-learning methods. Haus. https://www.haus.io/blog/how-traditional-marketing-mix-modeling-mmm-works-in-2025 Accessed 2026-05-31. ↩
HubSpot. "HubSpot Launches New AI, Breeze, Plus Hundreds of Product Updates at INBOUND 2024" (investor-relations news release, September 2024). https://ir.hubspot.com/news-releases/news-release-details/hubspot-launches-new-ai-breeze-plus-hundreds-product-updates Accessed 2026-05-31. ↩
Adobe. "Adobe Firefly: Comprehensive and Commercially Safe AI Content Creation for Businesses" (vendor description of training data and Content Credentials). https://business.adobe.com/products/firefly-business/firefly-ai-approach.html Accessed 2026-05-31. ↩
U.S. Federal Trade Commission. "Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials" (announced 14 August 2024; effective 21 October 2024). https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials Accessed 2026-05-31. ↩
"Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems." EU Artificial Intelligence Act (transparency obligations applicable from 2 August 2026). https://artificialintelligenceact.eu/article/50/ Accessed 2026-05-31. ↩