☰

StreetSmart AI Policy

Last updated: 21 September 2026

StreetSmart uses artificial intelligence to help people see and understand the streets and public spaces around them. This policy explains how, why, and with what limits. If anything here is unclear, don't hesitate to get in touch.

What we're using AI for

We use AI to improve public space, support better decision-making, and build public demand for change. We don't ask AI to decide for itself what makes a street good or bad. StreetSmart's people-centred criteria, rules and interpretation framework are developed by our urban experts. AI applies that framework to what it detects in a street photo and helps communicate the resulting insights. StreetSmart's insights aren't just for app users, either: they're meant to be useful to communities, planners, and advocates working on public space, including people who never open the app but benefit from the audits, reports, and conversations it helps generate.

How we use AI

Limited AI use. StreetSmart uses AI at two specific points in the photo analysis process:

  1. A computer vision model detects the individual elements in a streetscape photo, such as crossings, trees, seating, and so on.
  2. A language model follows instructions and rules written by StreetSmart's urban planning and design professionals to generate a short title and summary for each insight, and to determine which pre-written “Dig Deeper” section is linked to it.

Human-defined framework. StreetSmart's criteria, rules and “dig deeper” content are developed by our urban planning and design professionals. AI applies that framework to individual street images. AI is one part of our process; it is not our source of judgement.

Your experience matters most. We ask you directly about your own experience of the street because that lived experience is central to the evidence StreetSmart brings to communities and decision-makers. Your subjective feedback is not shared with any AI model or used to influence the AI-generated insights. Our AI-assisted analysis exists to build urban literacy and spark discussion, not to replace or attempt to capture your lived experience.

What guides our AI use

No profiling. StreetSmart doesn't identify who you are, and we don't build a profile of you or attempt to personalize our automated analyses to your location or reported experience.

Not neutral. StreetSmart does not claim that its analysis is neutral. It is built on a people-centred view of streets and deliberately challenges car-dominant planning assumptions, including the idea that vehicle movement and parking should automatically take precedence over safety, accessibility, health, public life and climate resilience. What we choose to measure, how we interpret street design and the insights we surface all reflect those priorities.

Limits on accuracy. AI-generated insights can be incomplete or incorrect, and they're limited to what's visible in a single photo. Their goal is to build urban literacy and spark discussion, not to serve as a professional site assessment. If you're making a real planning, safety, or accessibility decision, treat our insights as a conversation starter, not a substitute for a qualified, in-person assessment.

Bias. Computer vision and language models can reflect biases in what they were trained on, including which streets, contexts, and geographies they've seen most. We're committed to testing StreetSmart's AI outputs across a range of different geographies and contexts, not just the easiest ones to test in, and to reviewing and correcting biases we find. This is ongoing work, not a one-time task.

Privacy by design. Faces and licence plates are blurred before permanent storage, unblurred images are only briefly processed by our AI provider, and the image analysis is limited to detecting elements, not identifying people. See our Privacy Policy for the full picture.

Which AI we use, and how we decide

Which AI we use. We currently use Google's Gemini models to power the detection and AI-written text described above. We choose providers based on privacy safeguards (including contractual guarantees our data isn't used to train the provider's public models), reliability and accuracy for this task, and appropriate data transfer and security safeguards for GDPR compliance. AI moves quickly, and we expect our models and providers to change over time. When they do, we'll update this policy.

Checking our work. We test new AI model performance two ways: against expert-reviewed examples, checking its output against a set of streetscape photos our team has already reviewed and analysed; and on tone and framing, using a second AI system to review whether the AI-written titles and summaries match the tone and framing guidelines our experts set. Together, these checks help us evaluate whether model changes will result in AI-assisted analyses closer to what an urban expert would provide themselves.

Questions

We're glad to answer questions about how StreetSmart uses AI. For the wider legal context our AI use sits within, the EU AI Act is a good place to start; we follow its development and aim to continue aligning our practices as its requirements take effect.

This policy may be updated as our AI use, providers, or the regulatory landscape evolve. Significant changes will be reflected in the ‘last updated’ date above.