---
title: "Where a lead came from"
description: "Open the conversation behind any lead or booking-link click."
source: https://hiy.ai/docs/conversations
---

# Where a lead came from

"You got 3 follow-ups" is a number you can doubt. The exchange that produced
one is something you can read. Every conversation your twin has is stored with
the outcome it produced attached to it — this page is about opening that.

## Three ways in

- **Inbox** — press a person and their conversation opens beside the list, with
  **Open in Activity** for the full review screen. In the older view each row has a
  **See the conversation** link instead. If that conversation also opened your
  booking link, the row says so.
- **Insights → Where your bookings came from** — booking-link clicks from the
  last 30 days that have a conversation behind them, who it was if they left
  details, and a link into the exchange. One row per conversation, so someone
  who clicked twice appears once. The running *total* of booking-link clicks
  lives one panel up in *Clicks from answers*, and it is the larger number:
  it counts every click, including ones with no conversation attached. The
  questions in *Questions that became leads* open the same way.
- **Activity** — every run in your twin's activity log has an **Open**
  button, whether or not it produced a lead or a click.

All three open the same review screen, and **Delete this conversation** is
there whichever way you arrived — the review screen reads what it needs off
the stored exchange, not off which door you opened it from.

**Fix this answer is for a wrong answer as much as a missing one.** Every
answer in the transcript carries its own **Fix this answer** button — not
just the first question in the conversation, whichever one needs correcting.
Pick **Give this answer**, **Say it doesn't know** (prefilled with your own words from When it can't answer, if you set them), or **Hand off**, save, and
you're told straight away whether the fix is live on the next question or
waiting for your next train.

**A correction retires the source it corrects, for that question.** Your
correction wins, and the passage it overruled stops being offered to readers as
a source behind that answer — a reader should never be handed a link to the
claim you just withdrew. It only affects the question you corrected: a source
that also covers something else keeps being cited for that.

## Replay a conversation

Press **Replay** to step through a conversation: on a person in the Inbox, or on a
conversation opened from Activity. It opens in a
side sheet on a desktop and full screen on a phone.

Each step is something that was stored: **Asked**, **Read** *n* pages (the
sources the answer cited), **Answered** with its sources and how long it took
(only when that time was stored), **Said it didn’t know**, **Handed over**,
**Left their details** and **Opened a link**. Nothing is guessed to fill a gap.

- **Play** moves one step every 1.6 seconds and stops at the end; press
  **Replay again** to start over.
- **Drag the track**, press a step in the list, or use the arrow keys (←/→),
  Home and End. Space plays and pauses.
- **Try this again in Test** is in the foot wherever the agent’s Test is open.

What isn’t kept yet: the wording of each search, and pages an answer read but
didn’t cite. The replay says so under its **?**. It follows the same Founding
rule as seeing the conversation, and every conversation is still in your
data export on any plan.

## What you see

The whole exchange, in order, with the outcome marked where it happened:

- **Left their details here** — the moment in the conversation they filled in
  the card. (On a Support Agent it reads *Asked for a person here*.)
- **Followed your link here** — with the link itself, and a *booking link*
  note when it was one.
- **Which of your sources each answer drew on**, as citation chips.

That last one is *your* record of your own twin, not a replay of what the
visitor saw. If you've turned **Show sources on answers** off, visitors never
saw those chips — but your copy of what your twin actually used stays
complete.

**A laid-out answer is the exception: that one IS a replay.** When an answer
arrived to a visitor as steps, a comparison, a list of facts, or an FAQ, the
transcript here shows that same layout, not just the prose underneath it —
so what you're reviewing is what they actually read.

**Only the transcript scrolls.** The title, the "when and how many messages"
line and the close button stay where they are however long the conversation
runs, so on a phone you can close it from the end of a long exchange without
scrolling back up. Tapping outside the panel closes it too, and so does Escape
if you have a keyboard. A short conversation still shows as a short panel — the
box grows with what is in it, up to a height that always leaves the page
visible behind it.

**With a keyboard, the transcript takes focus as the panel opens.** Arrow keys
and Page Up / Page Down move through the conversation straight away, Tab moves
on to the close button, and Escape closes the panel from anywhere in it.

## What is kept

When the owner has turned on [masking](/docs/masking), a card, IBAN, ID or passport number in a conversation saved from then on shows masked, such as `•••• 4242`, in Activity, the Inbox and here, with a `?` that says it can't be shown. Conversations saved before it was turned on stay as they were.

**People** lists the follow-ups your twin captured, and every row on it is
something somebody typed into a conversation: a name, an address or a phone
number, the question they were asking when they left it, and anything a
*Collect details* action gathered. Beside that sits the conversation itself,
plus the status you gave the row and — where the plan includes it — whether
that session followed a booking link in the
[last 30 days](#the-30-day-window).

**And that is the edge of it.** The status on a row is *yours*: hiy sets it
once, to new, and never moves it again, because what happened after somebody
left your page happened somewhere hiy cannot see. A reply you sent from your
own mail, a call you took, a deal that closed — none of it reaches hiy, and
none of it is inferred. That is why the list shows what you set rather than a
stage it worked out for itself, and why nothing here is ever labelled *won*
except by you.

Removing a person takes their row and their follow-up record with it.
[Deleting a conversation](#what-it-deliberately-doesn-t-do) is narrower — it
takes the transcript and leaves the outcome.

## What it deliberately doesn't do

- **It won't link one person's conversations together.** hiy keeps no identity
  across conversations — no cookie that follows someone between visits, no
  profile being assembled. Two visits from the same person are two
  conversations, and hiy cannot tell you they were the same person.
- **It reports clicks, never bookings.** "Opened your booking link" is what was
  observed. Whether a call was actually booked happens on your calendar, where
  hiy cannot see it, and it will never claim otherwise.
- **Deleting a conversation removes the transcript, not the outcome.** The
  review screen's **Delete this conversation** (owner-only) takes the
  exchange itself. A lead or a booking-link click it produced stays on your
  list and still counts in Insights — only the transcript is gone, so if
  someone asks you to erase what they typed, this is narrower than removing
  them from People, which takes their follow-up record with it too.
- **Long conversations are capped at 200 messages**, and when that happens the
  transcript says so rather than quietly stopping.

## The thumbs under an answer

Every finished answer — on your hiy.ai page, in an embed, and on a Support or
Team Agent — carries a thumbs-up and a thumbs-down for the visitor. A click
records **the verdict, and which conversation it came from**: not the question,
not the answer it sat under, not who the visitor is. The link to the
conversation is made only when the visitor's browser proves it holds that
conversation (the same proof a booking-link click needs); a click without it
still counts, with no conversation attached. Thumbs from before this was added
have no conversation. One verdict per answer, taken once: after a click the
buttons stop accepting a second one. Your own preview chat is excluded, the same
way your test questions never spend your monthly messages.

Two places show them:

- **Insights › Overview › Visitor thumbs** shows the share of thumbs that were
  up, as **82% helpful** with **41 of 50 thumbs up** under it. With fewer than
  20 thumbs in the period it says **Not enough conversations yet** and the raw
  count (**3 up · 1 down**), never a percentage. When you pick a channel, only
  thumbs tied to a conversation on that channel count.
- **Unanswered › an open issue** gets a **Thumbs** row, such as **2 down · 1 up in
  these conversations**, once at least one thumb is tied to a conversation in the
  issue.

## The 30-day window

Booking clicks are shown for the last 30 days, on both surfaces — the People
chip and the Insights panel read the same window, so they can't tell you
different things. The **See the conversation** link itself has no window: if
the conversation is still stored, it opens.

## Which plan

The attributed view — the outcome, the conversation, and the moment they met,
in one place — is part of Founding.

Two things that are worth being plain about. **Your conversations are yours on
any plan**: every one of them is in your data export, free, whenever you want
it. And **nothing about honesty is ever behind a plan** — citations, the
honest "I don't know" and its search count, the AI label, the Report link and
the Verified badge are free forever. What Founding buys is having the outcome
and the exchange in the same place, not access to your own data.

One carve-out: on a **Support Agent**, a conversation that handed a question to
your **Inbox** always opens, on any plan. The promise there is that nothing
dead-ends, and a handover you can't read has dead-ended.

## Who can open it

The account that created it, and anyone it has invited to that agent. An invite
names the agents it opens, so a colleague invited to one Support Agent cannot
read another's conversations. People invited before agents were named still
read every agent on the account until you limit them (their row in
**Workspace → Team → Members** says **All agents**). A **Team Agent** keeps no
per-person trail for anyone — see below.

An [invited colleague](/docs/team-agents) signs in, opens `/app`, and the twin
or agent appears in the twin switcher under **Shared with you** — marked
*Read-only*, with the Inbox as its only screen. From there they can read every
person who left details and every conversation behind a handover, exactly as the
owner sees them.

What an invited colleague cannot do is almost everything else. They cannot
change its knowledge, voice, rules or appearance; they cannot publish it, wire
it into a site or see a Support Agent's token settings; they cannot change a
follow-up's status, remove a person from the list, or delete a conversation; and
they cannot see or change the plan. Those stay with the account that created it.
Inviting someone is a key to the Inbox, not a seat at the controls.

There is exactly one thing they can write, and it is below.

## Who took it

Open a conversation from the Inbox and, if you are an invited colleague, there
is one button under the transcript: **I replied**. Press it and the row in the
list reads *"Replied by Sam Rivera · 2 hr ago"* for everyone who can see that
Inbox.

That is recorded **after** the fact, by the person who did the thing, and it
answers a question a shared inbox otherwise cannot: was this handover actually
answered, or only read?

**A different mark from being assigned one.** The owner can also say who a
conversation is *for*, from an **Assignee** picker beside its status — see
[Who it's assigned to](/docs/handover#who-it-s-assigned-to). The two can
disagree, and both show: a conversation can read *"Assigned to Sam"* and
*"Replied by Priya"* at once. "I replied" is the one mark ANY invited colleague
can make, about themselves, after the fact; assigning is owner-only, decided
in advance, about someone else.

- **One mark per conversation.** If a second colleague presses it, the mark
  moves to them rather than stacking up. It says *someone did*, and names the
  most recent.
- **No note, no text, no copy of your reply.** The reply happened in your own
  mail; hiy holds a timestamp and who pressed the button, and nothing else.
- **The owner isn't offered it.** They are the person a handover was handed
  *to*, not somebody who picked it up on their behalf.
- **A colleague with no display name yet** — invited people don't have one
  until they set one — reads as *"Replied by a colleague"*. We would rather say
  the true vague thing than invent a name.
- **Delete the conversation and the mark goes with it.** So does the name, if
  that person's account is removed.

**Not on a Team Agent.** There is no Inbox to mark and no per-person trail to
add one to — see below.

## Team Agents

Team Agents have no attribution view, on purpose. A Team Agent answers your
colleagues from internal material, and hiy's default for an organisation is
that admins see usage and topics — not a per-person trail of who asked what.
Building one for Team Agents would contradict that, so the People rows for a
Team Agent stay exactly as they are.

The same line holds for invited colleagues: a Team Agent **never** appears under
*Shared with you*, and no colleague can read its conversations. A read surface
for one would be a per-person trail of the people they work with, which is the
exact thing that promise rules out.

## What visitors are told

Nothing here changes what a visitor is promised. [The visitor
page](/docs/for-visitors) says their conversation is stored and visible to the
creator, and [the privacy policy](/privacy) says the same. This surfaces what
was already true — it doesn't store anything new.