Microsoft wound Zo down in stages, by its own account

The design

Microsoft positioned Zo as a successor to Tay, describing it on the product's own FAQ page, archived while the bot was live, as 'a social AI chatbot with #friendgoals,' available on Kik, Skype, GroupMe, Twitter direct message and Facebook Messenger, and later Samsung phones through AT&T. The FAQ states plainly Zo is provided 'for entertainment purposes' and that 'Microsoft does not guarantee accuracy, reliability, or appropriateness of Zo's statements.' Unlike Tay, Zo's design built in an explicit consent step for richer data: the first time a user sends an image or voice file, 'she will respond to the file, and ask if it's ok to save image and voice files sent by you for product improvement,' with an opt-out, 'clear me from your memory Zo,' triggering deletion 'within 30 days.'

What the evidence says

Microsoft's own later account, published after the product wound down and archived in November 2019, states in the past tense that 'from late fall 2016 to summer 2019, Zo chatted one-on-one and in groups' across those platforms. An earlier snapshot of the same page, from mid-2019, is more precise about the wind-down's start: 'as of March 1, 2019, Zo stopped posting to Instagram, Facebook, and Twitter and chatting on Kik, Skype, and Twitter DM,' remaining briefly on Facebook Messenger, GroupMe and Samsung's AT&T phones before full retirement that summer. Microsoft states no specific reason for discontinuing Zo in either page, and this entry supplies none.

What it asks of people

Zo's design asked users for the same casual input as Tay, but layered an explicit choice on top for image and voice data, and a documented route to withdraw consent, a difference from Tay's blanket one-year retention suggesting Microsoft built Zo partly in response to Tay's earlier failure, though neither page states that connection directly.

Privacy and safeguards

Microsoft's post-shutdown FAQ tells former users chat logs remain visible only 'in your instant messaging application by scrolling up,' and states 'Microsoft has removed any user-identifying information from the data it has retained,' a company statement about its own research data, not an independently audited claim. Compared with Tay, Zo's live-era FAQ describing per-file consent and a 30-day deletion window marks a documented change in data design between Microsoft's two chatbots.

• Did the product's design change in response to a predecessor's public failure, and does the company say so directly?
• What consent step, if any, does the product require before richer data such as images or voice is retained?
• Does an 'anonymized' or 'removed user-identifying information' statement specify a method, or is it asserted without detail?

Zo's own live-era FAQ shows a more deliberate consent and deletion design than Tay's blanket retention, and Microsoft's own after-the-fact account dates a phased 2019 shutdown without explaining why. Read against Tay, Zo is best understood as Microsoft's documented attempt at a more cautious version of the same experiment, retired quietly rather than dramatically, according to the company's own pages.