What you'll build
By the end of this guide, you'll have a form with one or more Agent Fields that automatically generate AI-powered analysis, like a lead quality score, a sentiment summary, or a risk explanation, the moment someone submits your form. The output is stored in the response row and shows up as its own column in your Responses table, but the submitter never sees it. You'll write the AI prompt using the same answer piping menu you already know from ending pages and email templates, so the analysis reads based on exactly what that person submitted.
ℹ️ The Agent Field is available on the Business and Enterprise plans.
When to use this
You want to score or summarize a submission for your team, but the ending page is visible to whoever fills out the form, so you can't safely show your internal read on them there
You've been using AI Content in email templates to run an AI prompt on new submissions, but the output never gets stored as row data you can filter, sort, or export
You want a quality or risk score sitting next to every lead, application, or support request without exposing that judgment to the person who submitted it
You want more than one AI take on the same submission, for example a lead score and a sentiment read and a suggested next action, without duplicating your form
Your team currently copies submissions into a separate AI tool to get a read on them, and you'd rather have that analysis appear automatically in the Responses table
Adding the field in the form editor
The Agent Field lives in the form's Add field panel like any other field type, but it behaves differently once it's on the canvas: the submitter never fills it in. Instead, it sits quietly in the background and generates its value only after the form is submitted.
1. Open the Add field panel: In your form editor, click Add field and look for Agent Field. If you're starting from scratch, the Form Creation Agent can build the rest of your form's fields first, so you can drop the Agent Field in afterward.
2. Add the field to your form: Click Agent Field to drop it onto the canvas. It arrives admin-only by default, with no visible position on the form respondents will fill out.
3. Write a starting prompt: Open the field's prompt box and write what you want the AI to analyze, for example "Summarize this lead's intent and urgency." You can pipe submitted answers straight into the prompt using the @ menu, the same way you would on an ending page or in an email template. The next section covers writing the full prompt and answer piping in detail.
4. Give it a clear title: Name the field for what it's actually analyzing, for example "Lead quality analysis," "Sentiment summary," or "Suggested next action." This title is what you'll see as the column header in the Responses table, so make it specific.
5. Repeat for each analysis you need: A form can hold multiple Agent Fields. Add one for each distinct read you want on the same submission, for example one field scoring lead quality and a separate field summarizing sentiment, rather than trying to fit everything into one prompt.
6. Publish your form: Once your prompts are written, click Publish in the island bar at the top of the editor to make the field live.
Field settings and writing the AI prompt
Each Agent Field opens into the same AI box you already use on ending pages and in email templates, so the writing experience is consistent even though the output behaves differently here.
Field type, ID, and visibility: the top of the settings sidebar shows the field type and an editable field ID. Unlike other fields, Admin-only is locked on for the Agent Field and can't be turned off. There's no toggle to expose the output on the form itself.
The AI prompt box: write the instructions that tell the AI what to analyze, for example "Analyze this submission and summarize the user's intent, urgency, and quality as a potential lead." This is the same prompt editor used on the ending page and in custom email templates, so if you've written an AI prompt in either of those before, this will feel familiar.
Answer piping inside the prompt: click the @ menu inside the prompt box to insert any submitted field by reference, instead of typing field names from memory. A prompt like "Based on @company_size, @industry, @budget, and @use_case, decide whether this is a high, medium, or low quality lead. Explain your reasoning in one short paragraph" pulls in the respondent's actual answers at submission time.
Analyze uploaded files: turn this on if you want the Agent Field to read files uploaded to the record, like a CV, an invoice, or a contract, along with the rest of the submission. It's off by default and works for the first run, logic-triggered runs, and manual re-runs. It's available on the Enterprise plan. On other plans the setting shows as locked, with a Contact sales option. Each run can analyze up to 10 files, 5 MB per file, and 20 MB in total. Files over the limit are skipped, and the rest of the run still goes ahead.
Where the output shows up: once generated, the Agent Field's output behaves like any other field's data. It appears in your project's data block alongside the rest of the response, downloads with everything else when you export to Excel, syncs over if you've connected the form to Google Sheets, and is included in any webhook payload triggered by that submission.
Visibility: what respondents actually see
This is the part that makes the Agent Field different from every other field type: respondents never see it, and you can't change that.
The field does not appear anywhere in the live form. It's not on the last page, it's not on a hidden page, it's simply not part of what a respondent fills out or scrolls past.
Because Admin-only is locked on, there's no setting anywhere that exposes the Agent Field's output to the person who submitted the form.
The output is accessible to your team based on your existing response access permissions in the workspace, the same permissions that already govern who can see other admin-only fields.
Adding an Agent Field doesn't change how your ending page behaves. If you're already running an AI prompt there, both can exist on the same form independently.
💡 Want internal fields, like status, assignee, or scores, invisible to the person filling out the form? What are admin-only fields and how they help is the foundational guide to that mechanism, and it's exactly what the Agent Field builds on, just permanently on instead of optional.
How the Agent Field runs after a submission
Every Agent Field on a form runs automatically, right after someone submits. You can also run one again later, either by hand or with logic.
The moment a respondent submits, Formaloo creates the response row first, then triggers every Agent Field on that form.
If a form has more than one Agent Field, Formaloo sends all of those AI requests at once instead of working through them one at a time, so you're not waiting on field two because field one is still generating.
Each field's output is saved to its own column independently. If one Agent Field takes longer or fails, it doesn't hold up or break the others, they still complete and save normally.
Because outputs can land at slightly different times, the Responses table updates each column as its result comes in rather than waiting for every Agent Field to finish before showing any of them.
Run an Agent Field again manually: open the record in the Edit Row sidebar and click Run again next to the Agent Field. Change whatever you need, save the record, then run again. It uses the latest saved data and runs only that field, and the button shows Running… until it finishes. If the record has unsaved changes, the button is disabled until you save. A field that is already running can't be started a second time.
Responses table and error states
Each Agent Field shows up as its own column in the Responses table, right alongside your regular fields.
While generating: if a respondent just submitted and the output isn't ready yet, the cell shows a Generating… state rather than sitting empty with no explanation.
If generation fails: the cell shows a Failed to generate state along with an internal error indicator, so you know to investigate rather than assuming the AI simply had nothing to say.
Once complete: the cell displays the plain-text output exactly as the AI generated it, ready to read, filter, or export like any other field's data.
💡 Want to see only the submissions that matter right now? How to sort and filter your submissions data covers filtering and sorting on any field in the same Responses table, including Agent Field columns.
Using Agent Field output in email templates and PDFs
Because the Agent Field's output is stored as real row data with its own field ID, it's fully supported in answer piping wherever piping already works, not just inside the Responses table. Set an ID on the Agent Field, then recall that same value anywhere with @field_id.
Custom email templates: reference the Agent Field's ID in the body of a custom email template using @field_id. Since the Agent Field is always admin-only, use it in an internal notification you send to your own team, not the confirmation email that goes to the respondent.
PDF templates: reference the same field ID in the PDF template using @field_id, so an internal review PDF can display your Agent Field output right next to the raw answers it was generated from, useful for a reviewer packet or an internal case file.
💡 See how to create and send custom email templates and how to create PDF templates to turn responses into documents for the full setup on either.
Allow AI to update records: let the Agent Field edit your form's own fields, not just analyze them
On top of generating its own private analysis, an Agent Field can also update other fields on the same row directly. Turn this on when you want the AI's read to become the record itself, not just sit next to it.
1. Add an Agent Field: set it up as covered above, with its own prompt describing what to read and analyze.
2. Turn on "Allow AI to update records": in the field's settings, enable this option, then choose which fields it's allowed to write to, either all fields on the form or just the specific ones you pick. That's the whole setup.
3. Let it run after submission: once someone submits, the AI reads the full submission and, because the option is on, writes its result straight into the fields you selected on that same row, no extra step needed.
4. Trigger logic off the update (optional): if you want something to happen once those fields change, like a notification, reassignment, or status change, add an On Update logic rule on your form. Because the AI's update goes through the normal update path, On Update logic fires on it exactly like it would if a person had made the edit.
ℹ️ You decide exactly which fields the AI can update. Leave it set to all fields if you want the AI to potentially touch anything on the row, or select only specific fields if you'd rather it stay limited to a few, like a status, a score, or a handful of contact fields.
Use cases
Job applications: the applicant only uploads their CV. The Agent Field reads the file and fills in the rest, name, phone number, email, years of experience, so those fields can stay admin-only instead of asking the applicant to retype what's already on their resume.
Healthcare intake: a patient describes their symptoms in their own words. The AI reads that description and updates an Assignee or Department field so the case automatically routes to the right doctor or team.
Dynamic price calculators: a customer fills in project details, like scope, quantities, or add-ons. The Agent Field updates a price or quote variable based on those answers, which then gets piped into the email or PDF the client receives.
Re-run an Agent Field automatically with logic
An Agent Field normally runs once, when the form is submitted. If the record changes later, for example a lead becomes urgent, you can make the Agent Field run again on its own. You do this with an On Update logic rule, so the analysis always reflects the latest saved data.
1. Open advanced logic: in the form settings panel on the right side of the editor, open Advanced logic.
2. Go to the On Update section: add a new rule under On Update. The Run Agent action is only available here, not on submission.
3. Set your condition: choose the change that should trigger the re-run, for example "Priority is changed to Urgent."
4. Choose Run Agent: add the Run Agent action and pick the Agent Field you want to run. Add more rules if different changes should re-run different Agent Fields.
5. Save your form: from now on, when a record meets the condition, only the Agent Field you selected runs again, using the record's latest saved data. Your other Agent Fields stay unchanged.
Note: to prevent endless loops, for example when an Agent Field updates a field that triggers the same rule again, Formaloo stops after 3 Agent-triggered runs in a row on the same record.
💡 Want a rule to fire whenever an existing record changes, like a status flipping to Approved? On Update is the trigger for that. What is On Update logic and how it works covers it in full.
What you now have
You now have a way to run AI-generated analysis on every submission that stays entirely internal, without touching the meaning of your ending page or relying on AI Content in emails that never stored its output as real data. Every response gets its own AI read, or several, sitting right in the Responses table where you can sort, filter, and export it. Your team gets that analysis automatically on every submission, and the person who filled out the form never sees the internal judgment you formed about their answers.
What's next
Route submissions based on your Agent Field output Once you have a lead score or risk read stored on every response, you can use advanced logic to notify the right person or update a status field when that output crosses a threshold you check for manually. → How to add advanced logic to your form
Keep your ending page AI output separate from your internal analysis If you're also running an AI prompt on your ending page for the respondent's benefit, it's worth understanding how that box works independently from the Agent Field so the two don't get confused. → How to use answer piping in Formaloo
See a full AI-scored workflow end to end For a complete example of AI analysis driving a real workflow, from scoring through routing, this walks through the pattern in detail. → How to build an AI-powered lead qualification system
Ready-to-use templates
Don't build this from scratch, we've already put it together for you. None of our current templates are built specifically around the Agent Field, but these are the closest starting points if you're building an AI-scored intake or review workflow.
AI lead qualification & scoring form — Scores incoming leads with AI and routes hot and cold leads differently. Swap its scoring logic for an Agent Field to keep that quality read fully internal instead of feeding it through the ending page.
AI-powered job application pipeline with candidate screening — Screens candidates with AI as applications come in. Add an Agent Field for a private hiring-manager summary that never surfaces to the applicant.
Patient intake & consent workflow for healthcare — Uses AI-powered email templates as part of intake. An Agent Field is a natural fit for a private triage note on each intake that only clinical staff should see.
Learn more how-tos
See AI scoring drive a full lead workflow This walks through the surrounding pieces, hot and cold routing, conditional emails, and Slack notifications, that pair naturally with a private Agent Field score. → How to build an AI-powered lead qualification system
Screen applicants with AI before a human ever looks Another workflow where AI reads a submission before your team does, useful context for framing what an Agent Field prompt should ask for. → How to build an AI-powered job application pipeline with candidate screening
Pull the same answers into a document instead of an internal field If you also need submission data turned into a contract, invoice, or certificate rather than an internal AI read, this covers the document side of answer piping. → How to create PDF templates to turn responses into documents