Conversations are not training data. Your agent never learns from the raw chats themselves. The Conversations tab is a source that helps you build out Q&A: it reads support or sales chats your team already had, drafts suggested question-and-answer pairs, and waits for you to approve them. Only those approved Q&A pairs become training — the conversations stay a pipeline into Q&A, not a knowledge pool the agent draws from.
That makes Conversations the best way to capture the questions you never thought to write down: the small "how do I..." and "what happens if..." asks that come up constantly in support or sales chats but never made it into a knowledge base article — phrased the way your team actually answers, without typing every pair by hand.
What Conversations does with each conversation
What it is
For every question it finds in an imported conversation, eChat decides one of three things: a new or updated Q&A pair should be suggested, a Q&A pair should exist but there's no good answer to draft yet, or nothing needs to change because your agent already handles that question well from its real training (Q&A and other sources). Nothing from a conversation is added to training automatically — every suggestion is something you review and approve yourself. All approval is manual today.
What it's best for
- Surfacing tribal knowledge that lives in your support inbox or sales chats so you can turn it into lasting Q&A
- Matching your team's actual tone and phrasing when you write those Q&A answers
- Filling Q&A gaps after launch, once you can see what visitors are actually asking your human team
How it works
eChat reads each imported conversation, identifies the question a visitor was asking, and checks it against everything your agent already knows before deciding what Q&A to suggest. It does this across every conversation you import, so a busy month of chats can turn into a useful stack of ready-to-approve Q&A pairs with no writing on your part. The chats themselves are never stored as answers the agent can quote later — only the Q&A you approve is. Raw imported conversations and suggestions still waiting for review don't count toward your training storage either; only approved Q&A does.
Imports are streaming: suggested Q&A pairs start appearing in your review queue while the import is still running, so you don't have to wait for the whole batch to finish before you begin approving.
Best practices
Import Conversations after you've set up your core Q&A pairs, not before. When your agent already knows the fundamentals, eChat recognizes what's already covered and won't waste your review queue re-suggesting it — you'll mostly see the genuinely new material worth adding as Q&A.
Importing your conversation history
What it is
You can bring conversations in two ways. For agents linked to an eWebinar webinar, choose Use eWebinar as a connected source (the recommended Import eWebinar conversations card) to pull directly from your eWebinar chat archive — no export or upload step required. Or upload a chat export file (NDJSON, CSV, or JSON) with Upload chat export. Either way, you're feeding the Q&A suggestion pipeline, not uploading chats as training.
How it works
How far back. For an eWebinar import, How far back to import offers the last 7, 30, 60, 90, or 180 days, or the last 1 year; the default is 90 days. After an import finishes you can widen Since on the source card to pull in older conversations — ones already imported are skipped. For a file upload, Include last sets the window (the default is 180 days), then Start Processing begins the import.
Credit budget. Processing conversations uses credits, which cover the cost of reading and distilling them. For an eWebinar import, Max credits to use sets a budget — it defaults to your remaining credits, and the import stops when that budget is used. Conversations are processed newest first, so a tight budget spends on your most recent chats. Before it starts, eChat confirms the import and tells you roughly how many credits it will stop after. See Plans, Credits & Usage.
While an import runs, progress counters show how far along you are, and when it finishes you get a dismissible completion notice.
Four controls matter when an import isn't finished in one pass:
- Stop pauses the current import at a checkpoint.
- Continue resumes that stopped import from the checkpoint — it picks up where Stop left off.
- Resume picks up leftover older conversations after an import shows Budget reached, or imports the extra range when you widen Since. Already-imported conversations are skipped.
- Re-import (shown as Restart when leftover batches are waiting) starts over: archived conversations are fetched again from the start of the window. Q&A you've already approved or answered is kept.
Best practices
If you have a large backlog to bring in as a file export as well as a webinar you plan to connect as an ongoing source, import the file first and let Q&A review settle before you turn on Use eWebinar as a connected source — it's easier to judge suggestion quality in one batch than in a trickle. Start with a modest window and credit budget, review what comes in, then widen Since if the suggestions are useful. Use Stop when you need a break mid-import, then Continue from the checkpoint rather than starting over.
The three things a conversation can turn into
What it is
Every conversation you import lands on the Conversations source's own card, sorted into one of three states depending on what eChat found. You work through them there — approving, answering, or just glancing at what didn't need anything — and once you approve or answer something, it becomes a real Q&A pair, the same as if you'd typed it in yourself. That approved Q&A is what trains the agent. It stays that way even if you later delete or re-import the Conversations source it came from.
How it works
Needs review is for a question where eChat found, or drafted, a good answer — either a brand new Q&A pair, or an update to one you already have. Your job is to approve it, edit the wording, or reject it. During a streaming import, items can show up here before the whole import finishes.
Needs answers is for a real question eChat thinks deserves a Q&A pair, but it couldn't find a good answer for it — not in that conversation, and not anywhere else it's already trained on. It surfaces the bare question and asks you to write the answer yourself. Check training on a card asks eChat to look again against your current training; it uses one credit when it runs.
Not needed is for a question eChat looked at and decided your agent already handles well with what it currently knows, so no new or updated Q&A is necessary. It's shown to you rather than hidden, mainly so you can see the work was actually checked rather than skipped — though there's nothing you need to do with it. If you'd rather lock in an exact wording anyway, you can restore any item from Not needed and turn it into a real Q&A pair.
If two conversations turn out to be about the same question, eChat doesn't make you review it twice — the newer one merges into the same suggestion, with a newer answer taking priority over an older one. If a question first shows up in Needs answers with no answer, and a later conversation actually answers it, that same item moves itself up into Needs review instead of creating a second entry.
Best practices
Work through Needs review first — it's the fastest win, since the answer is already drafted for you. Then work through Needs answers, since those are genuine gaps: real questions your agent doesn't currently have a good Q&A for. You can generally skim past Not needed without acting on it — it's there for visibility, not action.
From creating Q&A to updating it
What it is
Where a conversation's suggestion shows up changes as your Q&A library fills in. Early on, most of what Conversations finds is brand new, so most of your review happens right here on the Conversations card. Over time, as more of your common questions already have a published answer, a new conversation is more likely to be refining something you already have than introducing something new.
How it works
When a conversation's question matches a Q&A pair you've already published, and the conversation suggests a better or different answer, that update shows up as a suggestion on the existing pair over in Q&A's own review queue — not here on the Conversations card, since you're not creating anything new, you're refining something you already own. If the conversation doesn't add anything your published answer doesn't already cover, nothing surfaces anywhere, the same idea as Not needed, just applied to a pair you already have.
Best practices
Don't be surprised if Conversations goes quiet as your Q&A library matures — that's usually a sign it's working, not that it stopped. Check Q&A's own review queue for that ongoing activity instead; that's where most of the refinement happens once the basics are covered.
Ongoing Q&A suggestions for eWebinar agents
What it is
For agents using Use eWebinar as a connected source, importing your history is a one-time backfill (within the window you chose), not a recurring chore you have to kick off by hand. After that initial import, newly archived eWebinar conversations that a human on your team helped answer keep feeding the same Q&A suggestion process — still not as raw training, still as candidates for you to approve into Q&A. Conversations handled entirely by the AI or a bot are skipped.
How it works
Live eWebinar archives don't process one at a time the moment each chat is archived. Instead they queue and are imported together in one batch — after 7 days, once 50 conversations are waiting, or right away if the source hasn't finished its first import yet. Each batch then runs through the same three-way decision as the original import: suggesting a new or updated Q&A, asking you for an answer, or finding nothing needs to change. The Conversations card shows how many conversations are queued and when the next import is expected.
You don't have to trigger each archive yourself. Your Q&A library grows as those queued batches are imported and you approve what lands in review.
Best practices
Keep an eye on both the Conversations card and Q&A's own review queue as a light ongoing habit, rather than treating either as a one-time setup task — together they're what keeps your agent's Q&A current for as long as your team keeps chatting. It pairs well with Chat Logs, where you can see which of your agent's own replies visitors reacted to well or badly.