About
Jump to:
Talk to the City (T3C) helps large groups of people coordinate by understanding each other better.
It's an open-source tool for public consultation that solves the trade-off between depth and scale, using LLMs for comprehensive analysis of large response data. Whether you're soliciting general feedback from thousands of constituents, or untangling subtle details of a complex concern, T3C preserves the nuance of individual views while making sense of the collective conversation.
Previous versions of Talk to the City have been used by the Taiwanese government and the Taiwan AI Assembly, unions, policy makers and more. See our case studies for more details.
Built by the AI Objectives Institute. All code is open-source on GitHub.
Have a question we didn't answer here, or interested in direct support in using Talk to the City? Reach out at hello@aiobjectives.org.
“Ten years ago, the vTaiwan scope was limited, because stakeholder groups needed to adapt to the technology at the time. Today, this is being bridged with the advent of language models that can adapt to their needs. Back in 2014, it was impossible with the capacity of g0v contributors to interview a mini public of people and aggregate their ideas while preserving the full nuance. But now, with Talk to the City's help, that cost has been reduced to essentially zero. It's broad-listening and it can change the nature of this recursive public.”
– Audrey Tang, Taiwan's 1st Digital Minister and co-author of Plurality.net -
“Given the vast amount of qualitative data we collected, Talk to the City's analysis was crucial in identifying gender-specific challenges in the lives of young Australian women and men. This analysis helped us craft our policy recommendations, focusing on the experiences of marginalised demographics. Having the data organised under specific themes, then being able to go through the interviewees actual statements, was powerful.”
– Rizina Yadav on behalf of Young Women's Alliance, a social initiative for improving the life outcomes of Australian women -
How it works
We use Large Language Models (LLMs) to analyze broad themes from large datasets of free-text responses, summarize specific claims, and link those claims back to exact quotes. We use LLMs from OpenAI (default: gpt-4o-mini), and are adding more options later this year.
We create an interactive report from the results, combining all scales of analysis, stored for you on the talktothe.city site. See examples in the Case studies section.
This alpha version of T3C accepts input of any unstructured text, including free-form survey responses, interview and meeting transcripts, and collections of social media posts. We have experimental features to process audio and video datasets, and include video in reports – reach out if you're interested in using these features while they're under development.
For this alpha launch, all reports are temporarily public to anyone who has the exact URL; we're adding password protection to reports soon.
Case studies
AI Assemblies
Heal Michigan
FAQ
What inputs do I need to create a T3C report?
- A correctly formatted CSV file
- A title and description for your report to help viewers understand the context for the data analyzed
How do I sign into T3C?
We currently allow signing in with accounts managed by Google:
- Click the "Sign in" button in the top right corner
- Choose the Google account you want to sign in with and complete authentication
- You'll then have access to your T3C dashboard
How secure is the data I send through T3C?
See our Privacy and security section for detailed information on our policies.
Who can I contact for support?
Reach out to hello@aiobjectives.org with questions or feedback.
Report creation FAQ
How do I create a new report?
- Click the "Create a report" button in the upper right corner
- Fill in required fields:
- Report name (e.g. "Sentiment Analysis June 2025")
- Report description (including context and purpose)
- The dataset to analyze, as a CSV
- Optionally, modify the prompts for any stage of the processing pipeline to better fit your use case/data.
What data format does the CSV file need to follow?
- Required column: "comment" (contains the text of the responses to analyze with T3C)
- Optional column: "interview" (respondent name, which can be a pseudonym, or blank for anonymous speakers)
- Optional column: “id” (for linking to any external data, will be generated if blank)
A sample CSV template is available in Google Sheets.
How do I customize the analysis prompts?
You can modify prompts to:
- Specify the number of topics/subtopics, or suggest possible themes to include
- Adjust length of direct quotations from respondents
- Focus on specific response types
What are the common errors and how can I avoid them?
- CSV formatting: Verify required columns are present
- Empty prompts: Ensure default prompts remain in place
- Refresh page and re-upload CSV if formatting issues occur
Privacy & security
Data handling
- Data transmission: When you use "Talk to the City," your text input (specifically the “comment” column of a CSV) is sent to OpenAI's API for processing.
- No data retention by OpenAI: OpenAI does not store the data sent to its API for long-term use, in accordance with their API usage policies. However, your inputs may be temporarily processed to improve their AI models unless you explicitly opt out.
What we collect
- User input: text or queries entered into "Talk to the City" are transmitted to OpenAI's API for generating responses. Raw input text, intermediate stages of processing, and final report data are stored in our cloud infrastructure so we can serve the report.
- Technical Information: We automatically collect certain technical information such as your IP address, browser type, and device information for security and service optimization purposes.
- Cookies and local storage: We use essential cookies and local storage to maintain your session and basic preferences. No third-party tracking cookies are used.
- No sensitive data: We discourage users from sharing personal, sensitive, or identifiable information in their input.
How we protect your data
- Encryption: All data transmitted to OpenAI is encrypted in transit using HTTPS.
- Error monitoring: We collect anonymous error logs to maintain service quality, excluding any user input or personal information.
- Data minimization: We collect only what is necessary to provide the service.
- Data deletion: You may request deletion of any data associated with you by contacting us (hello@aiobjectives.org).
Third-party sharing
No additional sharing: Your data is sent exclusively to OpenAI for processing. We do not share, sell, or distribute your data to any entity.
Your rights
- Access and control: You have the right to access, correct, or delete your technical data.
- Data portability: You may request an export of any data we hold about you.
- Opt-out rights: You can opt out of any optional data collection or processing.
- EU/UK rights: Additional rights under GDPR/UK GDPR include the right to object to processing and the right to lodge a complaint with supervisory authorities.
Content ownership
- User inputs: You retain rights to your input content.
- AI responses: Outputs generated by an external AI (currently only OpenAI models) are subject to OpenAI’s terms of service and usage policies.
Age Restrictions
This service is not intended for users under the age of 13. Users between 13-16 years old may require parental consent depending on their jurisdiction.
Your Responsibilities
- Safe use: Avoid submitting sensitive personal information or confidential data through the service.
- Compliance: Ensure your use of the service complies with applicable laws and regulations.
Changes to This Policy
We may update this Privacy and Security Policy from time to time. Any changes will be reflected here with an updated effective date. Significant changes will be communicated directly to users.
Policy last updated: 9 Feb 2025