Perspective AI news and product updates

Perspective changed its analysis sessions so that anything in quotation marks is exactly what a participant said, while paraphrases appear without quotation marks and still link to the moment in the conversation. It also removed the Generate Infographic feature, stating that it could show findings participants never said.

Buyer relevance. Removing a feature because it could misstate evidence is rare, and it matters when interview findings feed a legal or compliance decision. Teams relying on Perspective's quotes can now treat them as verbatim.

People invited to a Perspective conversation can now answer it inside an AI app such as Claude or ChatGPT through the Perspective MCP server, and new users can set up an account or join a team workspace from the same app. Voice only conversations opened this way send the person to the link to answer by voice.

Buyer relevance. For legal intake, a prospective client's answers can now be given inside an outside AI assistant rather than on the firm's own page, which changes where that intake data passes through. Firms should check whether this route can be switched off for confidential matters.

Perspective added GPT Live 1 to its opt in Live (Beta) voice conversations, using full duplex audio so the agent keeps listening while it speaks and can stop to follow a participant's correction; answers stay paired with audio for playback. Live (Beta) also adds 12 voices, and GPT-6 Sol became the default model for analysis sessions and scheduled insights. The existing Realtime voice engine remains the default unless an owner or admin switches.

Buyer relevance. Buyers tracking which models process client conversations should note that analysis now runs on GPT-6 Sol by default and that voice intake can move to a new beta engine, both set at the workspace admin level.

Perspective changed how its Form Data extraction treats questions a participant never answered: missing text, number and yes or no fields now stay empty instead of being filled with an empty string, 0 or No. A conversation that ends before qualification is decided is no longer filed as disqualified. The same release makes GPT-6 Sol the default model for agent design, with GPT-6 Luna also selectable.

Buyer relevance. For a firm running client intake through Perspective, an invented No or 0 in a structured field can misroute a matter or wrongly screen out a prospective client, so this directly changes how the system behaves when it has no answer.