Overview
The AI Extractor is a Journeys step that reads your candidate's conversation from an earlier step and tags it with five signals: sentiment, engagement, job interest, contact and data consent, and AI interview consent. You can then branch on these signals or use them in your messages.
With AI Extractor, you can route candidates automatically based on what they've told you, so your team no longer has to read every transcript. For example, you can send interested candidates straight to scheduling, or create a task for your recruiter when a candidate asks to speak with a person.
Before you start
Before you add the step, please make sure you have the following in place:
- Your journey already has a conversation step. AI Extractor reads a conversation from an earlier step in the same journey, so please add that step first. You can find the supported steps in How it works.
- You have permission to edit journeys. You'll need access to create and edit journeys in the workflow builder.
- Your journey has steps that act on the result. AI Extractor only labels the conversation. If you'd like to opt a candidate out, delete their data, or schedule an interview, you'll need to add the steps that do this.
How it works
Each AI Extractor step reads one conversation from an earlier step in your journey. It can read any of these:
- Voice call
- Chatbot over email, SMS, or WhatsApp
- SMS replies
- WhatsApp replies
- Survey responses
For each of the five signals, AI Extractor gives you exactly one value. Because every value comes from a fixed list, the same conversation always takes your candidate down the same branch.
| Signal | What it tells you | Possible values |
|---|---|---|
| Sentiment | How the candidate feels about the role or interaction | Positive, Negative, Mixed, Neutral |
| Engagement Level | How much effort the candidate puts into their replies | Engaged, Disengaged |
| Job Interest - Role Specific | Whether the candidate wants this specific role | Interested, Not interested |
| Contact & Data Consent | Whether the candidate agrees to be contacted and have their data stored | Consent granted, Consent denied, Erasure requested |
| AI Interview Consent / Intent | Whether the candidate agrees to an AI-led interview | Accepts AI interview, Declines AI interview, Requests human interview |
Any signal can also come back with one of these values:
- Unclear: The candidate replied, but the conversation doesn't give a clear answer. For example, they replied "Maybe" when asked about consent.
- No User Response: The candidate didn't reply.
Add an AI Extractor step
To get started, add the AI Extractor step right after the conversation step whose replies you'd like to analyze.
- Open your journey in the workflow builder.
- Below the conversation step you'd like to analyze, click Add a node.
- Under Advanced Automations, select AI Extractor. You'll find it right next to AI Generator.
- In AI Extractor Setup, under Input, add the transcript from your earlier conversation step. It appears as a variable, for example {{#A1.Your step name.Transcript}}.
- Under Output, review the five signals. They're grouped under Sentiment and Engagement and Consent and Compliance, and are included for you automatically, so there's nothing more to set up.
- Click Save.
Your AI Extractor step now appears in the journey, showing the transcript it reads and the signals it returns.
Test with a sample transcript
Before you publish your journey, we recommend checking which signals AI Extractor returns for a sample conversation. This helps you feel confident your branches will work as intended.
- In AI Extractor Setup, click Test Outputs.
- Under How do you want to add transcript?, choose how you'd like to add one:
- Paste as Text to paste a transcript.
- Upload a Transcript to upload a .txt file.
- Sample Transcripts to pick one of three ready-made examples, such as a forklift operator screening.
- Click Extract Outputs and review the five signals you get back.
- Click Done to return to the setup.
If a result isn't what you expected, try a transcript where the candidate states their intent more directly.
Use the results
Once the step has run, you can use the signal values in later steps of your journey:
- Branch on a signal using a Path node, so each candidate follows the route that fits their reply.
- Insert a signal into a message by including a signal value in the content of a later message.
Branch on a signal
- Below the AI Extractor step, add a Path node.
- In the condition, select the signal you'd like to check, for example {{#A6.AI Extractor.Sentiment}}.
- Choose is, then select a value from the list, for example Positive. You'll also see Unclear and No User Response here.
- Click Save. To route other values differently, click Add OR Block or Duplicate as new path.
Once saved, your Path node shows the condition, and candidates who match it continue to the next step, such as an SMS.
Example
Let's say your journey sends a chatbot over SMS to screen candidates for a warehouse role. You add an AI Extractor step to read that chatbot conversation, followed by these branches:
| If the signal is… | Route the candidate to… |
|---|---|
| Job Interest = Interested | Interview scheduling |
| AI Interview Consent / Intent = Requests human interview | A recruiter task |
| Contact & Data Consent = Erasure requested | A step that removes the candidate's data |
| Any signal = Unclear or No User Response | A follow-up message or a recruiter review |
Expected result
When your candidate reaches the AI Extractor step, you'll see one value for each of the five signals, and your candidate moves to the branch that matches.
If you're running a large batch, results may take a few minutes to arrive, as candidates are processed gradually.
Important notes and limitations
- Signals are labels only. "Consent denied" and "Erasure requested" don't opt the candidate out or delete their data on their own. Please add a step to your journey that takes care of this.
- The five signals are fixed. You can't add, remove, or rename signals or their values just yet.
- Each step reads one conversation. If you'd like to analyze two conversations, simply add a separate AI Extractor step for each one.
- Each step reads a conversation from earlier in the same journey. It can't read conversations from other journeys.
- You can't yet see which part of the conversation a signal came from.
- For intent and consent checks, we recommend AI Extractor over AI Generator. AI Generator writes free-form text, so its wording can change from run to run. AI Extractor always returns a value from a fixed list, which keeps your branching reliable.
Troubleshooting
| Problem | Likely cause | What to do |
|---|---|---|
| The candidate's step shows AI extraction failed | The AI service was unavailable and the automatic retries also failed | Check the candidate's journey activity. If the error keeps happening, please contact Support with the journey name and candidate, and we'll be happy to help. |
| Results haven't arrived yet | Large batches are processed gradually | Please wait a few minutes, then check again. |
| A signal returns No User Response | The candidate didn't reply in the conversation step | Check that your AI Extractor step reads the right conversation step. You may also want to add a branch for candidates who didn't reply. |
| A signal returns Unclear more often than expected | Your conversation doesn't ask the candidate directly about that topic | Try updating your conversation step to ask a direct question, then test again with a sample transcript. |
| AI Extractor doesn't appear under Advanced Automations | The feature isn't enabled on your account yet | Please reach out to your CSM, who can check this for you. |
FAQs
Does "Erasure requested" delete the candidate's data? No. AI Extractor only labels the conversation. To remove the candidate's data, please add a step that does this whenever this value comes back.
Can I extract other details, such as availability or pay expectations? Not yet, but more signals are planned for a later release. We'll update this article as they become available.
What's the difference between Unclear and No User Response? Unclear means the candidate replied but didn't give a clear answer. No User Response means they didn't reply at all. We recommend handling them separately: follow up to clarify with candidates who replied, and gently re-engage those who didn't.
Should I replace my existing AI Generator steps? For intent and consent checks, we recommend it. AI Extractor always sends the same conversation down the same branch. You can keep using AI Generator wherever you need free-form text.