What Is AI Visual Audit?
AI Visual Audit is an Audit Now feature that analyzes photos or short videos captured by auditors in the field and automatically answers the questions on a checklist page.
Here is how it works:
The admin enables a checklist page as AI Visual Audit and tells the AI how to evaluate each question.
When the field team reaches that page in the mobile app, they capture a photo or video.
The AI reviews the media, answers each question, provides a confidence score, and explains its reasoning.
The auditor confirms the answers, makes corrections if needed, and adds notes or actions.
The report is generated automatically with AI answers, reasoning, detected objects, and visual evidence.
In short: Your field team shifts from "performing the audit" to "collecting evidence for the audit." The AI makes the call; the human approves it.
When to Use It
AI Visual Audit is built for situations that can be verified visually. It is especially valuable in scenarios like these:
Hygiene and food safety audits — staff uniforms, hairnets/beard covers, open or closed containers, jewelry, and similar items.
In-store visual standards — window display, product placement, labeling, cleanliness.
On-site safety checks — fire extinguisher visibility, PPE usage, emergency exit signage.
Routine facility walkthroughs — repeatedly evaluating the same checkpoints to a consistent standard.
Multi-location quality consistency — verifying that the same standard is applied the same way across dozens of branches.
Part 1: Setting Up AI Visual Audit on Web
All AI Visual Audit configuration is done in the web app, from the checklist designer. Nothing needs to be set up on mobile; the moment your settings are published, they appear in the field team's app.
Page-Level AI Settings
A checklist page is either an AI Visual Audit page or it isn't — this choice is made per page. AI Visual Audit pages and standard pages can sit side by side in the same checklist.
Steps:
Open the checklist from the Checklists menu.
Click the purple gear icon (AI Visual Audits) next to the page you want AI to evaluate.
In the panel that opens, switch on Enable AI Visual Audits.
From here, you set the basics for the page:
Input Type (Required) — The field team will capture Photo or Video on this page. Use video for dynamic processes (food prep, assembly) and photos for static checks (window displays, storage areas).
Photo Scope — Two options:
Per Question: A separate photo is captured for each question. Best for independent checkpoints.
Single Capture: A single photo set is enough for all questions on the page. Use this when several things are checked in the same scene (kitchen hygiene, for example).
Maximum Photo Count — The upper limit your field team can capture (e.g., 5). Setting it too low risks insufficient evidence; setting it too high creates unnecessary media bloat.
Video Duration — Maximum recording length in video mode (default 30 seconds). Short, focused clips deliver more accurate analysis.
Capture Guide and Reference Content
These are the most critical settings for making sure your field team records from the right angle and with enough detail.
The Capture Guide field shows the auditor a short set of instructions the moment they switch to the camera in mobile. Keep it short, sequenced, and action-oriented, like the example below:
"Start by showing an overall view of the area being recorded. Move the camera slowly and steadily to capture the entire space and key checkpoints. Make sure staff, equipment, and work areas are clearly visible. Pause briefly on important details to highlight them. Keep the footage clear, well-lit, and uninterrupted."
In the Reference Content (Optional) section, you can upload a sample Reference Video or Reference Photo for the field team. This appears under the "Reference Visuals" button in the mobile app and shows the auditor what "a good capture looks like." It cuts training time significantly.
Question-Level AI Instructions
Page settings build the frame; the real driver of analysis quality is the AI Inspection Prompt at the question level. For every question, define how the AI should evaluate the visual — clearly and measurably.
Steps:
Click a question on a page where AI Visual Audit is enabled.
The Question Settings tab opens in the right panel.
Write what the AI should focus on in the AI Inspection Prompt field.
A good prompt example:
"Verify that the product label is clear and readable, that the barcode is intact, and that the expiration date is clearly visible. The label must not be torn, wrinkled, or faded."
A weak prompt: "Check the labels." (Too vague — produces inconsistent results.)
Three principles for writing strong prompts:
Be specific — instead of "clean," write "no grease stains, food debris, or marks."
Be measurable — instead of "appropriate," write "label readable from a 1-meter distance."
State the positive/negative case clearly — spell out what counts as "Yes" and what counts as "No."
Object Detection: Required and Forbidden Objects
If you want the AI to verify the presence or absence of specific objects in the image, use Object Detection. Instead of relying on text reasoning alone, the AI physically searches for the object — which significantly increases accuracy.
Toggle on Question Settings → Object Detection and fill in two lists (each with a maximum of 4 objects):
Required Objects — Objects that must be present in the image. Examples: "fire extinguisher," "hairnet," "hand sanitizer." If absent, the AI marks the answer as No / Fail.
Forbidden Objects — Objects that must not be present in the image. Examples: "watch," "ring," "open bottle." If detected, the AI marks the answer as No / Fail.
Object detection is especially powerful in hygiene and OHS audits. For instance, on the question "Does staff wear watches, rings, or bracelets?" — adding watch, ring, bracelet to the Forbidden Objects list lets the AI flag a fail the moment it detects one and highlight the marked frame in the PDF report.
Evaluation Examples
If you want the AI to be more consistent in borderline situations, add Evaluation Examples (up to 4 visuals per category):
Pass Examples — Visuals you consider "this counts as Yes / Pass."
Fail Examples — Visuals you consider "this counts as No / Fail."
These examples are particularly useful when you are setting up a new AI Visual Audit and want to clarify the quality threshold. If you have multiple auditors in the field who interpret the same situation differently, this is the first tool to reach for.
Note: Evaluation Examples are optional; the AI Inspection Prompt alone is enough in most cases. But for hard-to-judge questions (e.g., "Does the window display meet our standards?"), examples noticeably improve accuracy.
Part 2: Running an AI Visual Audit on Mobile
Once configuration is complete on the web, the workflow on mobile becomes very simple: capture, submit, review.
Starting an Audit
The field team can start an audit in one of three ways:
Ad-hoc start — Pick a checklist that contains an AI Visual Audit page from the Checklists screen.
Scheduled task — Open an assigned scheduled task from the notification or the tasks screen.
The first screen is the Start Page. Required fields are filled in here:
Location / Department — Where the audit takes place (selected from the location hierarchy).
Auditor Name — Pre-filled automatically from the logged-in user.
Audit Date — Pre-filled automatically; can be changed if needed.
Tapping Continue moves you to the AI Visual Audit page.
Audit Preparation Screen
This screen prepares the field team for capture. Three pieces of information are shown:
The mode selected for the page (Photo/Video Mode) and the duration or limit.
The number of questions the AI will answer on that page (e.g., "5 Questions").
Capture Guide — The instructions you configured on the web are shown here.
Optionally, the Reference Visuals button lets the user review the reference video or photo. Once ready, tapping Start opens the camera.
Capturing Photo or Video
The camera screen is single-purpose by design:
The page name and a short Capture Guide snippet (expandable) stay visible at the top.
At the bottom: mode selector (Photo Mode / Video Mode), zoom options (1x/2x/3x), and flash control.
A large capture button in the middle.
In video mode, recording stops automatically when it reaches the maximum duration. In photo mode, every captured image is listed in the gallery — you can delete, retake, or add new ones (up to the configured maximum).
Once capture is complete, tapping Continue starts the upload. After upload, the AI analysis runs on the server side; a progress indicator is shown while it runs.
Reviewing AI Answers
When the analysis is complete, the AI's answer is shown for each question with color coding:
Green selected button (e.g., Yes) — the AI considers the situation compliant.
Red selected button (e.g., No) — the AI has detected a non-compliance.
AI Confidence Score — Shown as a percentage below the answer (e.g., 95%). When the score is low, the AI answer is flagged with a yellow/orange warning; these answers should be reviewed manually.
Reason — A short, expandable text explaining the AI's reasoning.
Detected Objects — Highlighted as labeled boxes overlaid on the image.
If the AI couldn't answer any of the questions, the questions are shown with the standard manual answering UI; the field team can fill them in manually.
Editing Answers, Adding Notes and Actions
Three quick actions sit below each question card:
Edit Note — Modify the AI's Reason text or add your own note. The report shows both the AI answer and the user note.
Add Action — Create a corrective action for the finding. The question, location, and date are pre-filled.
Media — Attach additional photos or video.
To change the AI's answer, tap a different answer button. The system records the change; linked findings and action rules are updated automatically (Pass↔Fail transition).
Automatic action creation: If an action rule is defined for the question (e.g., "if the answer is No, an action is required"), the system opens an automatic action when the AI returns a Fail answer (for example, ACT-115 — Covered Storage of Food Items). This action appears immediately on mobile, with its priority and assignee.
When all questions are answered, tap Complete Audit to submit. The report is generated automatically — AI answers, confidence scores, reasoning, detected objects, created actions, and all media attachments appear in the PDF.
Tips for Best Results
Use specific and measurable criteria in the AI prompt. Instead of "clean," write "no grease stains or food debris."
Use object detection to guarantee the presence or absence of critical objects. It is the strongest tool for hygiene and OHS questions.
Use clear reference photos for required objects. Make it explicit what the AI should understand by "fire extinguisher" with example visuals.
Define forbidden objects explicitly to prevent errors. Anything not in the list isn't treated as forbidden.
Keep the capture guide short, sequenced, and action-oriented. Guides longer than three to five sentences won't be read.
Start with a pilot location. Compare AI answers against human review at a single branch, optimize the prompts, then roll out broadly.
Always manually review answers with low confidence scores. AI is a tool; the auditor remains accountable.
Keep historical AI answers. Tracking how the same question is answered for the same scene over time shows where your standard is evolving.