How Governly uses AI.

01 Overview

Governly uses third-party large language models (LLMs) to help board members, commissioners, and staff prepare for and understand meetings. Governly is model-agnostic and, as of June 2026, uses two providers — OpenAI and Google (Gemini).

02 What the AI does

For the meeting and policy material a Customer provides, the AI:

• summarizes agenda items and meetings;
• drafts strategic questions a member might raise;
• identifies the specific action a body is being asked to take;
• analyzes fiscal and stakeholder impact;
• checks agenda items against the organization's own adopted policy documents;
• drafts suggested comments; and
• powers a chat assistant that answers questions about a meeting, its attachments, and policy documents.

AI features are tailored by organization type (for example, school board, city commissions, or county commissions). The chat assistant is designed and instructed to answer only from retrieved source material and to avoid fabricating answers it cannot find in that material.

03 What the AI does not do

Governly's AI does not make decisions, cast votes, or take any action on its own. It does not provide legal advice or authoritative legal interpretation, and it is not a substitute for a body's deliberation, for staff judgment, or for professional legal or financial advice. It should not be relied on for final decisions without human review against the source documents.

04 Models, providers, and how your content is handled

As of June 2026, Governly routes work to two providers — OpenAI and Google (Gemini) — and sends them the meeting and organizational material to be processed: agenda item text and descriptions, attachment and PDF contents, policy documents, and the questions users type into the chat assistant. We do not train or fine-tune any model on client content.

We also use each provider under terms that prohibit them from training their models on the content we send. Because these are the providers' own policies and can change, we link to the current versions rather than restate them as permanent guarantees. As of the date of this writing:

• OpenAI (API) OpenAI does not use data submitted through its API to train its models, and retains API inputs and outputs only briefly for abuse monitoring, currently up to 30 days before deletion, unless a zero-data-retention arrangement is in place or a legal requirement compels longer retention. Current terms: OpenAI — Your Data.
• Google (Gemini) Governly uses Google's paid Gemini API tier, under which Google does not use submitted content to train its models and processes it as a data processor, logging it only briefly to enforce its usage policies and meet legal obligations. (Google's free tier does permit such use; Governly does not use the free tier for customer content.) Current terms: Google — Gemini API Paid Services data use.
• Embeddings On the data processing side, Governly converts meeting content and uploaded policy documents into numerical representations ("embeddings") that enable semantic search. These embeddings, along with the underlying text, are stored in Governly's own database (not with the AI providers). AI features are also tailored by organization type (e.g., school board, city commission, or county commission).

05 Accuracy and verification

AI summaries and analysis can be incomplete or wrong. Outputs are starting points, not authoritative records. Users should verify AI output against the underlying source documents before relying on it for any decision. Governly grounds outputs in the client's documents and makes them traceable to source material to support that verification — grounding reduces errors significantly.

06 Human oversight

Everything the AI produces — summaries, drafts, questions, analysis — is decision-support for humans. The board, commission, or governing body deliberates and votes; staff exercise their own judgment. Governly is built to inform those people, not to act for them.

07 Bias and fairness

Some features involve judgment calls: stakeholder-impact analysis, strategic-question generation, and any simulation of stakeholder or constituent perspectives. Outputs from these features can reflect biases present in the underlying models. They are intended to broaden, not narrow, the human's view, and a human should remain in the loop for anything that affects constituents.

• Non-discrimination Governly does not design its features to produce, and instructs its systems against producing, outputs that disadvantage individuals or groups on the basis of race, color, national origin, ethnicity, sex, gender, religion, age, disability, or other protected characteristics. Where a feature involves judgment — such as stakeholder-impact analysis or strategic-question generation — outputs are grounded in the Customer's own documents, are intended to broaden rather than narrow the human's view, and remain subject to human review and final decision by the governing body.
• Independent review As of June 2026, Governly's AI features have not undergone a formal independent bias or fairness audit by an outside party. We plan to commission such a review.

08 What you control (for Customers)

• Inputs You decide what to load into Governly and what to type into the chat assistant, and you remain responsible for whether that content is appropriate to process (see the Privacy Policy, "Customer responsibilities"). Content you submit is handled as described in Section 04.
• Records status Where you are a public body, you determine, as records custodian, whether AI-generated summaries, drafts, and chat exchanges are public records subject to your retention and disclosure obligations. Governly takes no position on that determination.
• Public disclosure You decide whether and how to tell board members and the public that AI assisted in preparing materials.