How Investigators Turn an AI Story into Evidence
The Seismic–Highspot question became meaningful only when the Antitrust Division could investigate it: not whether AI sounded transformative, but whether AI-native firms were entering sales enablement software, winning customers, and constraining established providers. This chapter examines that evidence problem. It explains why contemporaneous business records and market behavior usually matter more than litigation assertions, how the timing agreement focused the review, and what the public closing statement establishes without revealing.
Why ordinary-course evidence matters
A merger investigation often contains several kinds of statements about the future. A company may describe its strategy in a deal presentation, an expert may model likely effects for litigation, and a customer may describe what they might do under a hypothetical change. These materials can be relevant, but they are not interchangeable. Their value depends partly on when they were created, what decision they were meant to support, and whether other evidence corroborates them.
The Division’s public statement specifically says that AI claims are tested against ordinary-course documents and data from the merging companies and third parties. Ordinary-course evidence is material developed while the business is operating, rather than created solely to defend a transaction. It can reveal what executives actually monitored, which rivals they treated as threats, where they planned to invest, what customers they tried to win, and how the business described its own constraints before an investigation began.
That does not make every internal document true or every contemporaneous forecast accurate. A business plan can be wrong. A strategy document can reflect ambition rather than capability. A customer can change its mind. Ordinary-course status improves the evidentiary starting point; it does not eliminate interpretation. The investigator still has to ask what the document measures, what comparator it uses, and whether its claim is consistent with observed behavior.
The same principle applies to data. A customer-win record can show adoption, but not necessarily durable substitution. Growth can show momentum, but not the scale of competitive pressure. A product roadmap can show intent, but not successful execution. The evidence becomes stronger when different sources converge: company records point to a planned expansion, customers demonstrate willingness to buy, third parties recognize the entrant as a threat, and incumbent behavior changes in response.
The Guidelines describe several ways of examining competition, including customer substitution, win/loss reports, switching data, surveys, product characteristics, strategic decisions, and prior entry or exit events. These are not a mandatory checklist. They are ways to connect a claim to the market. For Seismic and Highspot, the relevant connection is between AI-native capability and the competitive constraint that the merger could remove.
A useful evidence record also separates observation from interpretation. “An AI-native firm won customers” is an observation if the underlying sales record supports it.