AI personas

A Digital Twin is an agent built only from what one account, segment, persona, or behavioral group has said in your workspace. You ask it a question in Chat and it answers in character, citing the interview, ticket, or call the answer came from. Every answer cites its source, so you can open that source and judge whether the twin represented it accurately. A twin also draws on the Channels and projects attached to it, so it reflects new feedback as it arrives, rather than describing customers as they were when a persona document was written. The Digital Twins page covers building one. This page covers using them across a team.

---

​

Decide which twins to build

Build twins for the questions your team asks repeatedly, such as whether enterprise buyers would accept a price change, why particular accounts left, or whether a flow makes sense to a new user. Each recurring question identifies one voice worth representing. A twin scoped this way gets used, because it answers a question the team already has. Build several narrow twins rather than one broad one. Three twins asked the same question show where segments disagree. A single twin for “our customers” combines those differences into one answer.

---

​

Write instructions that produce useful answers

Instructions have the largest effect on how a twin responds. Three instructions do most of the work:

• Speak in first person. Ask the twin to use “I” and “we” rather than “customers in this segment tend to”. First-person answers read as a customer response rather than as a summary of one.
• Push back. Include an instruction such as “push back when concepts don’t land—don’t hedge to be helpful.” A twin that agrees with every proposal produces no useful signal.
• Cite, and state gaps. Ask for the specific quote behind each claim, and instruct the twin to say so when nothing in the attached data covers a question. A twin that reports missing data lets you distinguish an absence of evidence from a weak answer.

Start from a ready-made prompt in Twin prompts. If you edit it substantially, see Writing instructions.

---

​

How teams use twins

• Testing a concept before building it. Paste a spec or a screenshot into Chat and ask what the twin responds to. Remove the weak directions, then test the remainder with customers, so live sessions cover the options still under consideration.
• Rehearsing objections. Ask an account or segment twin what it would object to before a renewal, a price change, or a QBR, then ask what would change its position. The second answer is the one to prepare for.
• Checking assumptions during a review. One person queries the twin in Chat during a roadmap review or design critique. When the discussion turns to what customers want, the claim can be checked during the meeting rather than afterwards.
• Testing language. Ask the twin what it would call a feature, or which of two descriptions matches a real problem. Customers’ own terms are usually clearer than internal ones.
• Preparing for customer interviews. Questions the twin answers with citations you trust don’t need a live session. The remaining questions form the discussion guide.

---

​

Share a twin with the team

Share the agent and teammates can enter its persona in Chat. Sharing controls who can invoke a twin, not who can change it. Only the creator can edit its instructions or attached sources, so a twin the team relies on doesn’t change unless the creator changes it. See Sharing agents. You talk to a twin in Chat. Type @ and the twin’s name to include it in a single message, or click the Digital Twin icon to stay in its persona for the whole conversation.

---

​

Results

Questions about how a customer segment would react can be answered during a discussion rather than deferred to follow-up research, and the answer cites the call or ticket it came from. Because the twin is shared, the team works from the same source rather than from each person’s recollection of the customers they spoke to most recently.

---

​

Where to go next