There are eight practical points where AI agents are changing connected TV advertising, running from forecasting reach to reconciling what was delivered against what was agreed. At each point the software takes on the volume work, and a person keeps one decision the software should not make alone.

TL;DR. In planning, AI forecasts reach and gathers publisher proposals. In buying, agents negotiate deals and move budget in flight, while in creative, generative tools build and reformat TV-ready ads. In delivery and measurement, agents watch suitability and frequency and reconcile results against the deal.

Where do AI agents sit in a CTV campaign?

The IAB 2026 Digital Video Ad Spend and Strategy report puts US digital video ad spend above $80 billion in 2026. Fox Advertising told Adweek its Fox AdStudio agentic system works across audience planning, media transactions and activation.

Alkimi is one worked example of how the stages connect. Buyers and sellers build their own agents in its Gym, set the rules those agents work to and spar against counterparty agents before real money moves, then trade on a neutral marketplace where people keep approval rights at deal level. Each use case below covers what AI does, what a person decides, an example and a risk.

How is AI used in CTV planning?

1. Forecasting reach and building the plan

What AI does: Planning agents model reach and frequency across streaming apps, group markets by business signals and draft a budget split by market, platform and format.

What a person still decides: The objective, the budget, which markets matter and how far to trust a forecast built on third-party audience data.

Example: On 1 October 2026 Locala launched CTV Planner, which groups markets by signals such as brand performance, competitive pressure and audience affinity. Locala says its AI agents turn those recommendations into parameters that can be activated directly within buying environments.

Risk: Forecast reach estimates the audience the inventory should deliver; it is not measured delivery. Treating it as a promise sets up an argument later.

2. Briefing publishers and gathering proposals

What AI does: A buyer agent turns the brief into a request for proposals, sends it to sellers and compares the replies.

What a person still decides: Which sellers are on the list, what counts as an acceptable substitute and which proposals are worth pursuing.

Example: IAB Tech Lab's AAMP 3.0 extends its Agentic Advertising Management Protocols to the brief stage with OpenProposal, a specification for automating campaign discovery and planning. IAB Tech Lab says sellers may find opportunities they were not previously considered for, and buyers may find plans from publishers they had not considered.

Risk: A publisher with a thinly described catalogue can drop off an agent's shortlist for reasons unrelated to its inventory.

How are AI agents used to buy and negotiate CTV?

3. Negotiating and setting up deals agent to agent

What AI does: A buyer agent and a seller agent exchange terms, agree a deal within limits each principal has set and set up the campaign. The deal usually still runs as a familiar deal type, as the four ways to buy CTV explains.

What a person still decides: Approval of the deal, plus the price ceilings, viewability floors and make-good terms written into the agent's mandate beforehand.

Example: AdExchanger reported that agency H/L's buyer agent worked with Magnite's seller agent through Model Context Protocol (MCP) to set up a campaign on LG smart TV home screens for the Northern California Toyota Dealers Association. Magnite's Mike Laband said the agent-led plan was approved by LG and H/L before it went live. The mechanics are covered in how AI agents negotiate CTV inventory.

Risk: A completed negotiation is not the same as a good deal. An agent judged on closing can accept weak protections, so the signed terms need checking as closely as the price.

4. Moving budget between platforms in flight

What AI does: An orchestration agent reads reporting across platforms, spots which inventory sources are growing or lagging and proposes changes, which it can make directly if permitted.

What a person still decides: Whether a shift goes ahead, and which kinds of change the agent may make without asking.

Example: AdExchanger described how InterMedia, an independent agency focused on CTV, connects Vibe and Olyzon so campaign managers can query performance and order changes from one dashboard. Sydney Brower, its director of digital media and strategy, prefers to double check inside each DSP before giving the go-ahead: "You can trust AI to fly a plane, but I still want a pilot in the cockpit to watch it."

Risk: Brower told AdExchanger that integrating several sources still takes troubleshooting and manual standardisation, because DSPs often use different terminology for the same concepts. An agent that misreads a field can move money on a false signal.

How is AI changing CTV creative?

5. Producing TV-ready ads from brand assets

What AI does: Generative tools turn logos, product images, brand guidelines and earlier ads into video spots, then produce variations for different audiences and placements.

What a person still decides: Approval of every claim, face and voice that reaches the screen, and whether a variant is on brand.

Example: Disney has opened a closed beta of Disney Ad Creative Studio inside Disney Campaign Manager, combining several AI models in one workflow that turns brand materials into CTV video ads and variations. Disney says "Human creativity, strategy, and oversight remain essential to the process."

Risk: Rights travel with the asset. Disney's description stresses safeguards for intellectual property and for image, voice and likeness, a sound checklist for any generated spot.

6. Reformatting creative for new CTV ad units

What AI does: Home screen tiles, pause ads and split-screen units need assets in different shapes and lengths, so an agent resizes and reformats existing creative to each specification by rules set in advance.

What a person still decides: The rules: what can be cropped, where the logo and legal copy must sit and which formats the brand will accept.

Example: Kargo chief executive Harry Kargman told Ad Age, in coverage republished by Kargo, that its agent is "reformatting your existing assets, with rules, very clear rules, into all of the different correctly sized and formatted assets that run in all those unique places."

Risk: Automated crops fail quietly. A tile that drops a disclaimer still serves, so sample checks on each new unit type should continue after launch.

How does AI manage CTV delivery and measurement?

7. Steering delivery for suitability and frequency

What AI does: Delivery agents check content before bidding, report against the advertiser's objectives afterwards and pass findings to the system that optimises the campaign.

What a person still decides: What counts as unsuitable for this brand and how often is too often, written down rather than inferred.

Example: At CES 2026 DoubleVerify announced a pre-bid agent that lets advertisers describe preferences and goals conversationally, and a post-bid agent that reports measurement against those objectives, according to the same Ad Age coverage.

Risk: Frequency caps set separately on each platform do not add up to a cap on the viewer. In a MediaPost column, Dave Morgan argued that when many siloed companies each promise a daily cap of three, one viewer might see the same ad a dozen times a day.

8. Reconciling delivery against the deal

What AI does: A measurement agent compares delivery logs with the agreed terms, flags gaps in impressions, completion or viewability and drafts any make-good claim.

What a person still decides: Whether a discrepancy is real, whether to accept a make-good and when a dispute needs a conversation with the seller.

Example: On 1 October 2026 IAB Europe published its final CTV Measurement Framework and Transparency Principles, which treat a clean, valid ad impression as the foundation for audience, exposure and outcome measurement. Providers reporting person-level reach are asked to say whether it is observed or modelled and to disclose assumptions such as co-viewing.

Risk: An agent can only reconcile against a record both sides accept. If each side keeps its own version of the deal, the agent checks delivery against a document the other party disputes.

How much should an agent decide on its own at each stage?

Earned autonomy offers a five-stage scale: Observe, Recommend, Draft, Human-approved action and Bounded automatic action. Planning and proposals suit Recommend or Draft, and negotiation belongs at Human-approved action, as in the LG and H/L campaign. Budget moves and reformatting can reach Bounded automatic action once limits are written down, while generated creative and disputes stay with a person.

What holds the eight use cases together?

Every stage hands a record to the next, and the deal becomes the yardstick for measurement. If each party keeps its own copy, small differences compound by the time delivery is checked.

A simulation published by Alkimi (Agree. Transact. Verify.) found mismatched records in 95.3% of simulated deals when each side kept its own, against 0.19% with a shared record. Its DealSheet is written at agreement, timestamped and held by both parties, so either side can retrieve the agreed terms independently.

Frequently asked questions

What are the main uses of AI in CTV advertising?

They are reach forecasting, proposal gathering, agent-to-agent negotiation, in-flight budget moves, generative production, reformatting, suitability and frequency control, and reconciliation. They span planning, buying, creative, and delivery and measurement.

Can an AI agent buy CTV without a person approving it?

An agent can act within limits set in a mandate, but approving the deal is the step most worth keeping with a person. In the Magnite and LG Ad Solutions campaign reported by AdExchanger, LG and the agency approved the agent-led plan before it went live.

Which standards cover AI agents trading CTV?

IAB Tech Lab's AAMP covers agent workflows from the brief, through OpenProposal, to transactions including programmatic guaranteed, PMPs and linear GRPs. AdCP, the Ad Context Protocol published by AgenticAdvertising.org, is an open protocol for agents to discover, plan, buy, sell and measure media.

Does AI make CTV measurement more reliable?

AI makes reconciliation faster, but reliability still depends on clean impressions and clear methodology. IAB Europe's CTV Measurement Framework sets out how delivery, audience and outcome metrics build on one another and what providers should disclose.

Is AI-generated creative ready for TV screens?

Broadcasters and platforms are building generative tools for CTV-ready ads, such as Disney Ad Creative Studio, now in closed beta. The open questions are rights, likeness and who signs off each variant.

AI in CTV Advertising: 8 Use Cases From Planning to Measurement