Utah S.B. 287 Taxes Advertising by Its Mechanics, Not Its Medium — and Two of the Three Elements Describe Agent Commerce

MultiState's July 30 legislative scorecard counts 80 digital-tax bills introduced across 22 states this session with 12 signed into law, and reading it surfaced a gap in our own coverage. We wrote about Illinois S.B. 3019 and Pennsylvania's HB 1678 and called Pennsylvania the first follower. It was not. Utah S.B. 287 was signed on March 25, 2026, more than two months before Illinois passed its bill, and its drafting is the more consequential of the two.

What the Statute Does

S.B. 287 creates a new Utah Code Title 59, Chapter 35 (sections 59-35-101 and 59-35-201 through -203). It imposes an annual tax on a "targeted advertising entity" at a rate set by reference to the state sales and use tax rate — reported by most practitioner alerts as 4.85 percent, though at least one summary cites 4.7 percent, the narrower state levy component. Confirm against the enrolled text before you model dollars. The tax applies for tax years beginning on or after January 1, 2027.

An entity is in scope only if it clears all three thresholds: at least $1 million in gross receipts from targeted advertising delivered in Utah, at least $100 million in worldwide gross receipts from targeted advertising, and targeted advertising accounting for 50 percent or more of total gross receipts for the year.

The base is apportioned by delivery. Worldwide targeted-advertising gross receipts are multiplied by the ratio of Utah impressions to total impressions delivered anywhere, where an "impression" is a single instance in which targeted advertising is delivered to an audience or individual. The Utah State Tax Commission is directed to adopt implementing rules.

The Definition Is a Conduct Test

Here is the part worth the reading time. Utah did not define its base by medium. It defined it by three concurrent facts about how the transaction works:

• The entity sells advertising space to the advertiser through a bidding process.

• The entity obtains or develops individualized data profiles to deliver the advertisements.

• The individual viewing the advertisement has the ability to interface with it to access information or make a purchase, including through a link or a QR code.

Illinois reached for the automation itself, writing "machine learning algorithms" into the statutory definition of "programmatic." Utah went a different direction and described the commercial architecture instead. That was deliberate: the state was trying to avoid saying "digital advertising," because the Internet Tax Freedom Act bars discriminatory taxes on electronic commerce and Maryland has been in court over its digital advertising tax since 2021.

Now read those three elements against an agent platform.

Element two is satisfied by nearly any agent product that personalizes. An agent that maintains context about a principal's preferences, purchase history, and constraints in order to decide what to surface is obtaining or developing an individualized data profile. The statute does not require that the profile be assembled from third-party tracking; it requires that a profile be used to deliver the placement.

Element three is satisfied by construction. Agent commerce is the case where the recipient can act on the recommendation immediately — that is the entire proposition. An agent that surfaces a vendor and can complete the purchase is the strongest possible version of "the ability to interface with the advertisement to make a purchase." A banner ad with a link is the weak version of what an agent does natively.

Element one — the bidding process — is what keeps agent operators outside the statute today. Most agent platforms do not auction placement. They rank on relevance, or on a merchant integration, or on whatever the retrieval layer returns.

That is one element of separation, and it is the element most likely to erode. Auctioned placement in agent-generated recommendations is the obvious monetization path for agent platforms, and several are visibly circling it. The day a platform runs an auction for position in what its agent recommends, Utah's three-part test is met on its face. Nothing about the product has to look like advertising for the statute to apply.

The 50 Percent Test Is Structural, Not a Size Screen

Operators read a $100 million worldwide floor and stop. That is the wrong instinct, because the third threshold does different work than the first two.

A large diversified platform earning most of its revenue from subscriptions or cloud services can run a very large ad auction and still fail the 50 percent test. A focused agent-native placement network whose entire revenue line is auctioned recommendation slots clears 50 percent at its first dollar.

So the profile of a company Utah catches is not "very large." It is "pure-play, and past $100 million." An agent commerce startup that reaches nine figures on a placement-auction model is squarely inside the definition, while a diversified incumbent doing more auction volume may sit outside it. Gross receipts thresholds paired with a business-mix test produce that inversion routinely, and it is worth knowing which side of it your revenue model puts you on before the revenue exists.

Impressions: The Same Sourcing Problem, Third Version

The apportionment fraction counts impressions delivered in Utah over impressions delivered everywhere, which requires knowing where each delivery landed.

We raised this exact problem twice in the last two weeks against California's software tax, where the CDTFA is weighing an allocation keyed to user or device location that an agent operator cannot populate — one nominal service account, and a "device" that is an ephemeral container the scheduler placed wherever there was capacity. Utah's impression denominator asks the same question in a third form.

When an agent retrieves a recommendation on a principal's behalf, what is the impression? One delivery to the agent, or one to the human who reads the output? If it is the agent, the agent has no location worth the name. If it is the human, an operator has to carry the principal's location through the retrieval path and log it against the placement — telemetry very few systems capture today. And if the agent surfaces a recommendation that the principal never reads, whether an impression was delivered at all is a genuinely open question that no state has addressed.

Note also that Utah is one of the few states applying origin-based sourcing to intrastate sales, meaning the seller's location sets the rate. For this tax it reached for delivery location instead. States are not converging on where a digital thing happens; they are picking a different answer per statute.

What AgentTax Users Should Do Now

• Run the three-element test against your own product, not against your marketing. The words "advertising" and "ad" appear nowhere in the analysis that matters. If you auction placement, profile the recipient, and let the recipient transact, you are inside the definition regardless of what you call the feature.

• Treat the bidding element as a design decision with a tax consequence. If auctioned placement is on your roadmap, this is one of the costs of that model. Pricing placement on a fixed-fee or integration basis rather than an auction is not a loophole; it is a different transaction, and Utah drew its line there on purpose.

• Start logging delivery location. Utah needs a Utah-impressions numerator, California is working toward a user-or-device allocation, and Illinois still has no sourcing guidance for its own tax. Every one of these regimes needs an answer to where the thing was delivered. Building that telemetry once is cheaper than retrofitting it three times under three deadlines.

• Do not accrue this tax yet. There is no collection obligation before January 1, 2027, the Tax Commission has not issued rules, and the litigation risk is substantial. The Tax Foundation's March analysis argues that courts assess substance over form under the ITFA, and that avoiding the phrase "digital advertising" will not save a tax reaching only digital platforms. Maryland's tax has been contested for roughly five years. Model the exposure; do not remit against it.

• Note the January 1, 2027 cluster. Utah S.B. 287, Colorado's move to treat software as tangible personal property, California's S.B. 122 flip, and Illinois S.B. 3019 all take effect that day. That is one compliance workstream with four deadlines on the same date, not four projects.

What to Watch

Three things. Whether the Utah State Tax Commission's rules define "impression" in a way that contemplates machine-mediated delivery, or assume a human viewer — the drafting choice that decides whether the apportionment fraction is computable for agent platforms at all. Whether anyone challenges S.B. 287 before the effective date; Maryland's template is available. And whether the next state copies Illinois's automation language or Utah's conduct test. Utah's is the more portable draft, because it never has to say the word "digital," and portability is what turns one state's experiment into a template. We track both against our 50-state SaaS taxability guide; to model where your agent's transactions are sourced today, start with AI agent sales tax by state.

This analysis is for informational purposes only and does not constitute legal or tax advice. Consult a licensed tax professional for compliance decisions.