AI email tools have had a rough reputation since they started showing up a few years ago. Most of them produce replies that sound like a chatbot wrote them after one too many cups of optimism. Superhuman’s latest take on the problem is different enough to actually be useful, at least some of the time.
The email client has launched a new version of its auto-draft feature that reads your inbox, figures out which messages need a reply, and writes one in your tone. It also gives you two alternative versions to choose from in case the first one misses the mark.
Superhuman has tried this before with instant replies and follow-up drafts, but those earlier attempts often sounded like a overly eager sales pitch. The difference now comes down to the models powering the feature. Co-founder Rahul Vohra confirmed the company has moved away from older models like GPT-3.5 and is now using a combination of frontier models from both Anthropic and OpenAI, which gives the system more intelligence and a much bigger context window to work with.
The feature does more than just generate a generic reply. It learns from your past conversations to match your tone, and it picks up on patterns from how you respond over time. After the feature mistakenly suggested agreeing to a meeting scheduled after midnight, it later recognised a similar request from someone else and drafted a response saying the timing didn’t work. That kind of correction matters.
During the testing phase, the results were promising in a measured way:
- 40% of auto-generated drafts were sent within one day
- 60% of those went out without any manual editing at all
In practice, the feature handled things like confirming embargo details on a pitch, agreeing to meeting times with minor edits, and even correctly responding to requests for authored posts by saying that isn’t something the recipient handles. That last one is particularly impressive, since it requires understanding a person’s role and limits, not just their tone.
There are still real limitations. The default responses can be too agreeable or miss context entirely, which means you still need to check before hitting send. But having two alternative drafts ready makes it easy to swap to something more appropriate without much effort.
Users can also give the feature more to work with by going to Settings, then Personalization, and adding information about themselves, their role, and any files or links that provide extra context. The more it knows, the better it gets.
This update comes at a time when AI-generated outreach has made inboxes harder to manage for anyone who receives a high volume of messages. PR and communications professionals, for example, are already using AI to write first drafts of pitches faster than ever, which means the volume hitting journalists and other recipients keeps climbing. A tool that helps people respond more efficiently without sounding robotic fills a real gap.
The broader context here is also worth noting. Last year, Grammarly acquired Superhuman and later rebranded the combined company under the Superhuman name. The company is now working on a product called Superhuman Go, an assistant designed to work across multiple platforms while carrying context from one app to another. The auto-draft feature looks like a building block for that bigger vision, where AI handles more of the repetitive communication work without losing the personal touch that makes replies worth reading.
