First session lab

A first session
worth returning to.

A small experiment in making the first companion conversation feel personal, consistent and worth continuing.

Reflective blue and lilac forms against a black background

From the public job post

The companion
product today.

The public hiring brief describes a companionship platform built around connection, with moderation governing the experience.

Read the public job post

Companionship at the core: The flagship product is described as a companionship platform focused on relationships and connection.
A governed experience: The post names EverGuard as the proprietary moderation system governing the product.
Scale that makes small changes matter: The post reports more than 50 million users and a goal of reaching 100 million next. These are company-reported figures.

Source: Senior Product Manager, remote Europe. Checked 1 October 2026. This is a summary of the public product brief, not an observed walkthrough of the live companion journey.

Working hypothesis

The bottleneck
is memory.

Remembering a detail is only useful if it makes the next moment feel connected.

Third-party source

Recall can be strong.
Continuity can still break.

A hands-on review reports strong recall in shorter conversations, but generic return openings and incorrect stored details in a longer test.

Reported by one independent reviewer. This is an external account, not verified internal product data or a measured retention result.

Read the companion review

Our inference

Make remembering
visible from the start.

If the companion correctly carries a shared detail into its next reply, the opening may feel more personal and less repetitive.

That is the hypothesis for this concept. The review does not establish memory as the primary cause of first-session drop-off.

Share a detailCarry it forward correctlyGive the conversation a next step

A little memory.
A better beginning?

Nova asks three light questions. At message eight, the treatment makes remembering visible—and lets the user decide what stays.

Control: the same three questions, followed by a generic next step.

Treatment: a three-detail review at message eight, with Keep, Edit and Forget.

Proposed primary signal: returning on day seven. Secondary signal: completing the first session. Incorrect recall and unwanted familiarity must not increase.

Fictional companion and answer presets. Replies are scripted and safe for work. Switching condition starts a fresh session. Nothing leaves this tab or survives a refresh.

Nova

Fictional companion · Scripted replies

2 messages

Choose a fictional answer

Day 7 return

Share returning seven days after their first session

0%10%20%30%40%

Control

28%

Treatment

34%

Illustrative difference: six more returning users per 100 people.

Illustrative synthetic cohort counts
Condition	Users	Day 7 returns
Control	1,000	280 (28%)
Treatment	1,000	340 (34%)

The decision we would take

Test with a small real cohort.
Keep the rollout limited.

This synthetic pattern would justify investigating the idea, not declaring a win.

If a real test shows a reliable day-seven lift, with no increase in incorrect recall or unwanted familiarity, expand the test. Otherwise, revise or stop.

Before a live test, set sample size, duration and guardrail limits. Synthetic numbers cannot establish significance or justify shipping.

Enable JavaScript to try the scripted session. The source brief, experiment description and illustrative results remain available.#results-view[hidden]{display:block}#session-view,.lab-toolbar{display:none}

A first sessionworth returning to.

The companionproduct today.

The bottleneckis memory.

Recall can be strong.Continuity can still break.

Make rememberingvisible from the start.

A little memory.A better beginning?