The CoStar Clause That Breaks Claude and ChatGPT Workflows

CoStar does not publish a self-serve API for CoStar Suite, LoopNet listing data, or its comps and analytics products. As of this writing PSV finds no publicly documented developer program, and access beyond the logged-in portal runs through commercial conversations with CoStar. That absence is not an oversight. CoStar’s business is licensed data, and its Terms of Use, the license agreement displayed at costar.com, reserve every path to that data. The operative clause prohibits licensees, without express written permission, from using “any data mining, gathering or extraction tool, or any robot, spider or other automatic device or manual process, to monitor or copy any portion of the Product or the data generated from it,” alongside prohibitions on using the product to build any database or product and on providing any portion of it to a competitor.

Two words in that clause deserve more attention than they get: “manual process.” The workaround that comes up most often in PSV’s rooms, teaching an AI agent to click through the portal the way a person would, log in, run a search, export five hundred records at a time, is automation of exactly the kind the sentence reaches, and the browser-driving part does not launder it. CoStar also enforces. In a 2016 press release the company announced a settlement in which RealMassive paid $1 million and acknowledged, in CoStar’s words, systematic misappropriation of CoStar research. Whatever one thinks of the terms, they are the terms, and an AI workflow built on violating a data license is not an asset. It is a liability with a run rate.

The license restricts how data leaves the product, not whether a licensed professional can use the product as built. CoStar’s portal ships export functions for licensed users in the ordinary course of use, subject to the caps and conditions in each firm’s license schedule, and those schedules differ by agreement, which is why the only reliable statement about what your seat may export is the one your CoStar representative and your counsel give you. PSV’s point is narrower and holds across schedules: there is a real difference between a person using the product’s own export features for the firm’s internal analysis, and a bot monitoring or copying the product. The first is how the product is sold. The second is what the clause prohibits.

That difference defines where AI legitimately enters: downstream. Once an export produced by a licensed seat is sitting in the firm’s own folder, on the firm’s own systems, the work AI is actually good at begins. Normalizing five hundred rows of inconsistent property records, deduplicating against the firm’s existing pipeline, cross-referencing a region list from an internal Excel file, flagging which sites match a buy box and which fail it and why. None of that touches CoStar’s systems. The product produced a report for a licensed user; the firm’s AI reads the firm’s copy of it. This is the same boundary PSV recommends for Yardi when the connector is unavailable, and it has the same virtue: the AI reads what the system of record already produced under existing controls, and writes nothing back.

Start with the decision the data feeds, which on most calls is site selection or comp support. The inputs are the export files a licensed analyst produces in the normal course of work, plus the firm’s own deal list and criteria. The workflow normalizes each export into one schema, deduplicates against everything previously exported so the running dataset stays clean, applies the firm’s screening criteria with the arithmetic done in code rather than by the language model, and returns a shortlist memo: which properties cleared, which failed, on which criterion, with every row traced to the export file and date it came from. The reviewer is the analyst who owns the territory. The approval gate is that no owner gets contacted and no LOI moves on a property that a person has not looked at in the portal, because the export is a snapshot and the listing may have moved.

The comparison worth internalizing is between the four paths people actually consider. PSV’s position is the third column, with the fourth as the complement for market data.