A scenario analysis of how conversational AI could reshape advertising interfaces, products and companies—grounded in live AI hiring signals from AdTech job postings.
Scenario analysis
This is a thought experiment. It does not claim that OpenAI is entering advertising, and it does not describe an announced advertising product.
Search reorganized advertising around keywords and explicit queries. Social platforms reorganized it around feeds, identity and audience graphs. Each shift created a new interface between consumer intent and commercial demand—and a new stack of buying, selling, measurement and creative companies around it.
Conversational AI could become another interface. Assistants can understand a multi-step goal, retain context, compare options and help complete a task. If platforms such as ChatGPT or similar assistants ever supported advertising, the unit of value might move from an impression or click toward a relevant recommendation or completed action.
The diagrams below explore that possibility, then compare it with current hiring signals in the AdTechTalent jobs database. Conceptual sections are clearly separated from measured job-market data.
1 · Current model
The simplified chain separates demand, transaction infrastructure, supply and the user-facing experience. Real implementations add identity, data, verification, creative and measurement layers around it.
The advertiser buys access to an audience or context. The DSP evaluates opportunities, an exchange or SSP clears the transaction, and a publisher renders the experience. Optimization is usually anchored to delivery, clicks, conversions and attributed revenue.
2 · AI-first scenario
In this hypothetical model, the assistant is not merely another publisher page. It interprets the task, decides whether commercial input is useful and helps the user move toward an outcome.
A viable system would need clear sponsorship labels, relevance thresholds, user control and independent measurement. Commercial eligibility would not justify a recommendation by itself; the recommendation would still need to serve the user's stated goal.
3 · Operating model
These are directional shifts, not predictions that every existing advertising primitive disappears.
| Today | AI-first possibility |
|---|---|
| Keywords | Intent |
| Audience segments | Conversation context |
| Clicks | Completed tasks |
| Landing pages | AI actions |
| Display ads | Conversational recommendations |
| Campaign optimization | Autonomous agents |
| Cookies | Long-term user memory |
4 · Product layer
Product concepts are included to make the scenario concrete. None is presented as an existing or announced OpenAI product.
Clearly labeled recommendations selected because they fit the task a user is trying to complete.
A market where eligible commercial intents can be matched with relevant offers under explicit policy controls.
A user agent requests availability, pricing or fulfillment from merchant agents and returns comparable options.
Sponsored products appear inside a useful answer without disguising the commercial relationship.
A guided buying workflow that compares constraints, explains tradeoffs and can hand off to checkout.
Structured, attributable commercial results that third-party agents can request and render with disclosure.
5 · Industry structure
The likely change is not a clean replacement of the supply chain. Existing capabilities could be repackaged around intent, actions, consent and machine-readable commerce.
| Company type | Primary role today | Possible AI-first role |
|---|---|---|
| DSP | Bids on impressions | Competes for eligible intents or agent tasks |
| SSP | Packages publisher inventory | Packages trusted actions, answers or commerce access |
| Ad exchange | Clears impression auctions | Clears intent, recommendation or action auctions |
| Publisher | Creates pages and audiences | Creates trusted knowledge, utility and proprietary context |
| Measurement | Attributes clicks and conversions | Verifies influence, task completion and incrementality |
| Creative platforms | Produce ad formats and variants | Produce adaptive conversational assets and product data |
| Retail media | Monetizes shopper traffic and SKU data | Supplies transaction-ready offers to assistants |
| CDP | Unifies customer profiles | Governs consented memory and context across agent interactions |
| Data providers | Sell segments and enrichment | Provide verifiable signals, product knowledge and permissions |
6 · Measured hiring data
This section uses published, non-expired jobs in the AdTechTalent database. It measures explicit AI language in active postings—not AI adoption, investment or revenue.
Active jobs analyzed
4,506
Published and not expired
AI-related jobs
1,635
Each posting counted once
Share of active jobs
36.3%
Point-in-time dataset share
Companies hiring
80
At least one matching role
Ranked by matching jobs. Engineering and total job counts use all active postings from each company.
| Company | Open AI jobs | Engineering jobs | Total jobs |
|---|---|---|---|
| 1.Merkle | 128 | 148 | 410 |
| 2.Microsoft | 102 | 42 | 152 |
| 3.Epsilon | 101 | 86 | 223 |
| 4.StackAdapt | 81 | 43 | 215 |
| 5.Moloco | 77 | 25 | 99 |
| 6.PubMatic | 72 | 29 | 109 |
| 7.RTB House | 60 | 20 | 106 |
| 8.The Trade Desk | 53 | 36 | 197 |
| 9.Tatari | 52 | 21 | 93 |
| 10.Zeta Global | 46 | 46 | 143 |
Explicit mentions across titles, descriptions, requirements, responsibilities, tags and structured tech stacks. A job can mention more than one skill.
Python
556 · 34%
Machine Learning
517 · 31.6%
LLMs
298 · 18.2%
AI Agents
184 · 11.3%
Deep Learning
141 · 8.6%
RAG
112 · 6.9%
PyTorch
99 · 6.1%
TensorFlow
88 · 5.4%
Prompt Engineering
83 · 5.1%
MLOps
77 · 4.7%
Embeddings
66 · 4%
AI Evaluation
62 · 3.8%
Vector Databases
53 · 3.2%
LangChain
52 · 3.2%
Fine-tuning
42 · 2.6%
NLP
27 · 1.7%
Reinforcement Learning
24 · 1.5%
Computer Vision
19 · 1.2%
Transformers
14 · 0.9%
Model Inference
10 · 0.6%
Titles are grouped only when they explicitly match a defined family such as Machine Learning Engineer, AI Engineer or Data Scientist. Unmatched titles are not forced into a category.
Data Scientist
78 · 4.8%
Machine Learning Engineer
48 · 2.9%
AI Engineer
24 · 1.5%
Applied Scientist
23 · 1.4%
AI Product Manager
17 · 1%
AI Solutions Architect
2 · 0.1%
ML Platform Engineer
2 · 0.1%
Research Engineer
1 · 0.1%
Ranked by AI-related share, not total volume. Companies need at least 5 active openings to qualify.
Ranked by active AI-related jobs with a structured city. Multi-location postings count once in each listed city.
Active AI-related jobs
Remote, hybrid and on-site shares among matching active roles.
Remote
383 jobs · 23.4%
Hybrid
620 jobs · 37.9%
On-site
632 jobs · 38.7%
7 · Scenario framework
This quadrant is an illustrative strategy framework, not a measured growth forecast. “Demand” reflects broad current relevance; “growing relevance” reflects the AI-interface scenario explored in this report.
8 · Value creation
The strongest positions may belong to companies that can make recommendations more useful, measurable, trustworthy or executable.
Inference, vector retrieval, evaluation and observability become critical operating layers.
They could supply the reasoning and language systems behind intent-aware commercial experiences.
AI-mediated advertising would expand demand for compute, data pipelines and governed model deployment.
New recommendation surfaces intensify the need for independent outcomes, incrementality and disclosure controls.
Retailers already connect product availability, transaction data and high commercial intent.
Conversational surfaces require modular assets that can adapt to context without losing brand control.
They can coordinate discovery, comparison, negotiation and action across merchant systems.
Original expertise and first-party relationships could become premium inputs to answer systems.
Demand-side intelligence could evolve from impression bidding into intent eligibility and agent decisioning.
Supply platforms could package trusted content, actions and commerce access for assistant interfaces.
9 · Constraints
Conversational advertising raises familiar AdTech problems in a more consequential interface: the system may influence not only what a user sees, but what it concludes and does.
Recommendation systems can reproduce skewed data or systematically advantage some sellers and users.
Advertisers and users need to understand why an offer was eligible, selected, ranked and priced.
Conversation context can be more sensitive than a browsing event and requires strict purpose limits.
Automated recommendations, sensitive inference and platform self-preferencing may attract new rules.
Users may stop relying on assistants if commercial influence is hidden or routinely outranks relevance.
Answer interfaces may reduce referral traffic without creating a durable way to reward original content.
Control of interface, intent and transaction could place excessive leverage in a small number of platforms.
10 · Conclusion
If conversational assistants become a major way people discover products, compare options and complete tasks, advertising could move closer to the moment where intent becomes action. That would create a powerful commercial surface—but only if relevance, disclosure and user control are treated as product requirements rather than compliance details.
The current hiring data does not prove this ecosystem will emerge. It does show where AdTech employers are already building AI capability: the companies, technical skills, role families, locations and workplace models visible in active postings. Those signals offer a grounded view of the talent base available for whatever model develops next.
Search transformed advertising. Social transformed advertising. AI assistants could become the next major advertising interface. Whether OpenAI ever becomes an advertising company or not, the talent market suggests that AdTech employers are preparing for more AI-native products and workflows.
The measured analysis covers published, non-expired AdTechTalent jobs. A posting qualifies when its title, description, summary, requirements, responsibilities, tags or structured tech stack contains an explicit AI term. Matching favors precision and avoids ambiguous lowercase letter sequences.
Skill, title and location charts are shown only when matching records exist. Mentions indicate expected exposure or capability, not that a role is dedicated exclusively to AI. Historical AI hiring is not charted because the current snapshot schema does not preserve comparable AI-signal history.
Explore AI-related AdTech work
Browse active opportunities that mention AI, machine learning, LLMs, agents and related capabilities.