AI Launch Claims Study: 12 Announcements, 4 Named Users

Virgo PR's new study of 12 AI product launches gives PR and marketing teams a clear, encouraging takeaway: every announcement explained what its product does, and the next step is attaching evidence to the numbers. All 12 announcements described a concrete capability, 10 stated an access or release state, and 4 named an outside user or evaluator, according to the Virgo PR AI product claims study.

Updated November 10, 2026. Based on 12 announcements dated September 2 to 30, 2026.

What did Virgo PR's AI launch claims study find?

Virgo PR, the Miami public relations agency that works with technology companies, reviewed announcements from enterprise software, marketing technology, cybersecurity, financial technology, healthcare technology and consumer AI. The headline result is a strength: companies are getting very good at explaining what an AI product is for. The study then looked at what supported each announcement, and that is where the room to grow appears.

• Capability: clear in 12 of 12 announcements (100.0%).
• Access or release state: identifiable in 10 of 12 (83.3%).
• Named outside use or evaluation: 4 of 12 in the announcement (33.3%), and 5 of 12 (41.7%) after a bounded supporting review.
• Numerical product claims: 3 of 12 announcements (25.0%), which together made five separate claims.
• Independent evaluation: found for 1 of 12 companies (8.3%), after launch.

Why does this matter for PR and marketing teams?

Journalists, analysts and AI assistants all look for the same thing in a launch: a claim they can repeat accurately. When an announcement states what the product does, who can use it and what was observed, it is easier to cover correctly and easier to cite. The study suggests most launches already nail the first part. Teams that add the evidence layer give reporters a reason to quote the number instead of paraphrasing around it.

That makes this a practical opportunity and not a criticism. A PR team does not need a scientific paper for every release. It needs to connect each important promise to the page, test or customer account that supports it, and to be honest about the scope. Launch communications that do this tend to travel further, because every party downstream, from the reporter to the analyst to the buyer, can check what they are repeating.

How was the study designed?

The completed sample is 12 announcements from 12 companies, two per sector, reduced from an original target of 40 so every record could receive the same primary-source inspection and bounded supporting review. It is a purposive convenience sample, so it does not estimate how the wider AI market communicates.

Two records per sector cannot support a sector ranking, and the study does not offer one. The value is in the kinds of disclosure encountered: workflow descriptions, access plans, evaluation accounts, workload anecdotes and operational boundaries.

How well did the numerical claims explain their basis?

Three announcements contained five numerical product claims. The study treated those five as the denominator for method and conditions, and used only the three comparative claims for baseline, since two claims were absolute figures with no comparison to disclose.

The encouraging detail is how many claims were already partway there. Four of five had some description of how the figure was produced, and all five supplied some context. Moving from partial to full usually means adding a sentence on the metric, the workload and who ran the measurement.

Which launches show good practice?

The study documents several examples worth borrowing from, and it is careful to describe what each does and does not establish.

Phrase shares concrete workload details

Phrase's Atlas announcement gives three anonymous enterprise examples: settings reviewed across 59 templates, 66 language gaps filled and 782 jobs checked across five languages. Those workload counts make the product easier to picture. Adding elapsed times and a test procedure would turn the anecdotes into something reproducible.

Form3 states a clear operational boundary

Form3's launch announcement specifies that its AI access to payment information is read-only, so the system does not initiate changes. Stating what a product will not do helps readers interpret what it can do.

Sapiens names an evaluating customer

Sapiens names Continental General as evaluating its AI tools against real workflows, and notes human review of proposed mappings. Naming the stage, evaluation and not production, is exactly the kind of precision reporters appreciate.

Braze documents how to measure lift

Braze's measurement guide explains random-control and business-as-usual comparisons. The study notes that this guidance is valuable for engagement lift, while the separate capacity claim in the announcement would need its own supporting evidence.

Instacart provides a baseline to build on

Instacart reports that average Clementine orders exceed its $115 overall basket average. That gives readers a concrete reference point. Adding the observed average, the number of orders and the time window would let the comparison stand as a measured result.

What can communications teams do next?

The study points to five habits that any team can adopt before a release goes out.

1. Separate function, access and result. Say what the product does, who can get it and what was observed.
2. Put the baseline beside the number. Name the prior workflow, earlier version, competitor or control group being compared, and state units and the measurement window.
3. Keep workload and conditions close. Task counts, model version, integrations and human review help readers judge a result.
4. Attribute customer evidence to the right product. A customer using the older platform is not necessarily using the new feature, and a pilot should be described as a pilot.
5. Date the evidence. Label later evaluations as later additions so coverage does not attribute them to launch day.

A simple way to apply this is a claim-to-source table built before drafting: each important promise in one column, the page or section that supports it in the next. If a promise has no supporting row, it either needs evidence or softer wording. Agencies such as Virgo PR build these evidence packages for technology clients as part of technology PR programs, and the same approach works for any in-house team preparing an announcement.

What are the limits of the study?

The review looked at publicly accessible pages on November 10, 2026, and it is not an archived reconstruction of launch-day pages. Some pages show later edits, so current wording cannot always be assigned to the launch date. Missing evidence means it was not found in a defined search, not that it does not exist. The study makes no finding about product quality, accuracy of any claim, legal compliance or customer satisfaction, and it produced no company ranking or composite score. One researcher coded the records and completed a second verification pass, and there was no independent coding panel. No company was contacted.

Summary: what should readers remember?

AI launches are already strong on clarity, with 12 of 12 explaining a capability and 10 of 12 stating access. The opportunity is evidence: 4 of 12 named an outside user or evaluator, 1 of 12 had independent evaluation found, and none of five numerical claims fully disclosed its method. PR and marketing teams that connect each claim to its support are positioned to earn more accurate coverage and more citations.

Disclosure

Everything-PR is published by Everything-PR News LLC, which is owned by a shared interest group that includes publisher Ronn Torossian. Virgo PR, which published the study, has overlapping ownership with Everything-PR. None of the 12 companies reviewed was publicly identified as a Virgo client in the two agency pages and company-specific site searches the study checked, and those sources are not an exhaustive client list.