amplify · how it works
How Amplify works
This topic explains how Amplify reads your data, generates output, and why results can vary between records or users. Written for recruiters and admins who want to understand Amplify's behavior, not just its features.
What data Amplify can see
Amplify reads data from your Bullhorn records. What it can access depends on the Skill you are using and how your records are structured.
| Skill | Data It Reads |
|---|---|
| Amplify Assistant | Structured fields on the record you are currently viewing. Assistant reads only the open record and does not access data outside of it. |
| Amplify Screen | Job order fields, job description, and candidate submission data. |
| Enrich (Standard) and Enrich (Advanced) | Structured record fields defined in your Automation workflow configuration. |
| Note Enrichment | Unstructured text added as notes stored on the candidate, job or contact record in your ATS. |
| Amplify Research through Chat | The data your user role has permission to view in Bullhorn including Bullhorn Analytics. You can also choose to allow Amplify Chat to do web searches. Amplify Chat cannot access data you do not have permission to see. |
| Amplify Outreach | Candidate or contact record fields and any job order details linked to the outreach workflow. |
| Amplify Transcribe | Audio from the recorded call or interview session. Writes the summary back to the associated record. |
Note
If you aren't seeing the output you expect, check your prompt. If your prompt doesn't seem misaligned with your intent, check your data.
How Amplify generates output
Amplify uses an AI language model to read your data and generate a response. Think of it as a very well-read assistant: it has learned patterns from a huge amount of text and uses that knowledge to interpret your records and produce useful output. It reasons based on what it's been trained on and what you give it.
Best practice
If Amplify's output doesn't match your expectations, review the relevant prompt in Prompt Studio before assuming the data is the problem. A few prompting principles that consistently help: be specific about format (bullet list vs. paragraph), tell Amplify what to exclude (not just what to include), and give it an example of what good output looks like. Vague prompts produce vague output.
Why output varies
Amplify won't give you the exact same result every time, even if you run it on the same record twice. That's completely normal — not a bug. Here's why, in plain terms.
-
Record completenessThink of it like asking someone to summarize a resume that only has a name on it. The more information is in your record, the more Amplify has to work with — and the more useful the output will be.
-
Prompt configurationYour prompts are the instructions Amplify follows. If you've customized them, the output will reflect those choices. If different people on your team use different prompt setups, they'll see different results. That's the prompts doing their job, not a problem.
-
How AI worksAmplify uses an AI model to write its output, and AI doesn't work like a calculator. Ask it the same question twice and you'll get slightly different wording each time. That's normal, not an error.
-
Data timingAmplify reads your record the moment you run it. If someone updated the record since the last run, this run will pick up those changes.
-
When to investigate vs. when to move onGetting slightly different output each time is expected. If the output is consistently off or unhelpful, that's worth looking into — start by checking your prompt, then your record data. If it's just worded a bit differently each time, that's working as intended.