PathwAI's Google Gemini integration lets AI Scan steps analyze respondents' photos with Gemini vision models. Gemini 3 Flash is the default model for a new AI Scan step, and Gemini 3.1 Pro and Gemini 2.5 Flash are also offered. Scans run only on your workspace's own Google AI API key, and the model returns the output fields you define, which are saved on the submission.
What data moves
- Respondent to PathwAI: one or more guided photos, taken with the camera or uploaded, stored in your workspace's assets.
- PathwAI to Google: the photos, the step's prompt and the output fields, read on the server from the published flow.
- Google to PathwAI: the outputs, saved with the photo links, the model and the time of analysis.
Where it is configured
The AI Scan step's Model setting picks the model. A Connect button there saves the workspace's Google AI key to Integrations, where every AI Scan step uses it.
Use cases
Wrinkle screening for aesthetics
The Wrinkles use case takes one front-on face photo and returns the most visible area, a severity and a summary.
Hairline photos for hair loss programs
The Hairline use case takes a front hairline and a top-of-head photo and returns stage, pattern and density fields.
Show the result to the respondent
A result card shows the outputs before the respondent continues, or the step moves on with the result hidden.
How to set it up
- Create an API key in Google AI Studio.
- Add an AI Scan step and choose a use case.
- Pick a Gemini model and click Connect to paste the key.
- Adjust the photos, outputs and disclaimer, then test on a phone before publishing.
Good to know
- The built-in AI Scan prompts tell the model to describe what is visible and never to diagnose, name a disease or recommend a treatment. The result is a screening description for a provider to review, and the step shows a disclaimer you can edit.
- AI Scan is in beta and available in design-system flows.
Google Gemini integration FAQ
Whose API key do scans use?
Your workspace's own Google AI key. PathwAI never runs scans on its own platform keys.
Can respondents change the analysis instructions?
No. The prompt and output fields are read on the server from the published flow.
Does the scan diagnose anything?
No. The built-in prompts describe what is visible and never diagnose or recommend treatment. A provider reviews the result.
Want to see Google Gemini in a flow?
We can walk through where Google Gemini fits in your funnel and exactly which data it would receive.