Ben Molin • Updated Aug 7, 2025
About our SNAP Chatbot
SNAP Screener now offers a SNAP Eligibility Chatbot to help people better navigate SNAP eligibility and our SNAP screening tools.
Our screeners are designed to be straightforward to complete and direct those interested to the application afterwards. However, for people with more complex situations or less common income types, it can be helpful to have a Chatbot that can help you figure out how to best answer questions on the screener, for example, "Does _____ count as income?" or "Who should I include in my household?"
Here's some information about how our Chatbot works. We may release our full instructions as open source. If you'd be interested in this, or would like to work together on SNAP AI technology, please email me. I'd like to see what's possible in ensuring benefit and social safety net information can benefit from new artificial intelligence technologies.
Chatbot Model
The Chatbot currently runs on Claude 4 Sonnet. I found Claude 4 Sonnet's ability to follow instructions to be better than Claude 3.7 Sonnet.
A major opportunity for error in SNAP guidance is applying one state's rules to another. It can be surprising which SNAP policy options vary between states. For this reason, every state has its own Chatbot, specifically trained on its state-specific data.
Chatbot Data Sources
The Chatbot is trained on the following data
-
Federal SNAP Policy: The majority of general SNAP policy, such as countable income, allowable deductions, household composition, and the SNAP benefit equation comes from the federal SNAP rules. Specifically, 7 CFR 273.
-
State SNAP Handbook: The state SNAP handbook provides some additional information about state policies. There is a lot of variance in state SNAP handbooks, which is the biggest reason of any quality differences between state chatbots.
-
Custom SNAP Screener Data: The main focus of snapscreener.com is providing information about income limits, asset limits, deductions, and the SNAP benefit formula. Custom data and instructions were compiled specifically for the Chatbots to be able to understand concepts such as Broad-Based Categorical Eligibility. This is the main strength of the Chatbot compared to using a regular AI tool, as our Chatbot is able to provide a higher level of quality in terms of providing correct and up-to-date income limits and coverage of multiple pathways to eligibility.
Chatbot Actions
The Chatbot links to resources that can be helpful for clients when applicable:
-
State Application: If a client wants to apply, they will be linked to the state's application (e.g. snapscreener.com/go/pa/apply)
-
State's Website: If a client wants official information about SNAP, they'll be linked to the state's SNAP website (e.g. snapscreener.com/go/pa/website)
-
State County Office: If a client wants to get in contact with someone at the state for help, they'll be linked to the state's SNAP County Assistance Office (e.g. snapscreener.com/go/pa/cao)
-
SNAP Screener Calculator: If a client wants to see if they're eligible for SNAP, or calculate a benefit amount, they're linked to our SNAP calculator (e.g. snapscreener.com/screener/pennsylvania)
-
SNAP Work Requirements Screener: If a client has questions about work requirements, they're linked to our SNAP work requirements tool (e.g. snapscreener.com/work-requirements/pennsylvania)
Chatbot Scope
SNAP Screener primarily focuses on pre-application topics, answering questions such as "Am I eligible" and "How much will my benefit amount be?". Accuracy is a core priority of the Chatbot, so these scope limitations help us ensure that the Chatbot is only attempting to answer questions where it can provide a high quality answer.
In Scope Topics
- What are the income limits?
- Does ______ count as income?
- Who should I include in my household?
- Can you help me understand this deduction?
Out of Scope Topics
The Chatbot is designed to not answer questions about anything after the application is started, including but not limited to the following topics:
- Issues with documentation
- Reporting requirements
- Recertification questions
Chatbot Errors
If it's determined the guidance of the Chatbot did not meet our quality standards, we update the data sources or instructions to ensure that the Chatbot either answers the question correctly or designates it as out of scope (so that it won't attempt to answer it in the future). The Chatbot will only continue to improve as we improve our data and the AI companies release better models.
We welcome any feedback via email, or you can send it directly to the Chatbot.