name: AI Disclosures Notice slug: ai-notice version: 1.0.0 effective_date: [EFFECTIVE_DATE] last_updated: 2026-05-09

AI Disclosures Notice

Plain-English Summary

We use AI to moderate the photos you upload and the messages you send, to suggest listing copy when you ask for it, and to enhance ad copy when you tap "Enhance." The provider is Google Gemini 2.5 Flash, accessed primarily through OpenRouter, with a direct-to-Google fallback. We do not send your account credentials, payment information, pickup addresses, precise geolocation, or other users' messages to AI providers. AI output is often wrong — you are responsible for reading, correcting, and verifying anything AI suggests before you publish it. Image moderation on uploads and message scanning are required as a condition of using the Service for safety reasons. AI suggestion features ("Fill with AI," "Enhance Ad") are entirely optional and only run when you tap the button.

This Notice supplements our Privacy Policy, Terms and Conditions, Acceptable Use Policy, and Advertising Policy. Where this Notice and the Privacy Policy describe the same topic, this Notice provides the AI-specific detail and the Privacy Policy provides the broader data-handling framework.

1. Where We Use AI

ShareFree integrates AI in five places. The following table summarizes each, what triggers it, what is sent to the AI provider, and what comes back.

Feature	What It Does	Trigger	What's Sent	What's Returned
`analyze-and-moderate-image`	Moderates uploaded listing photos and analyzes their content for listing suggestions	Automatic on listing photo upload	Image bytes, listing context (category and listing type if you have selected them)	Moderation verdict (approve / flag / reject), suggested category, suggested title fragments, content tags
`analyze-listing-image`	Generates listing-copy suggestions only (does not gate publication)	"Suggest with AI" user action	Image bytes, listing context	Suggested title, description, category
Fill with AI (listings)	Generates a listing title, description, and category from analyzed image data plus the context you have already entered	You tap the Fill with AI button in the listing creation flow	Analyzed image data plus user-supplied context (any text fields you have already filled in, type/category selection state)	Suggested title (≤ 80 characters), description, category, condition
`enhance-ad-copy`	Generates or improves an ad headline, description, and call-to-action	You tap Enhance in the ad creation flow	Existing ad fields (headline, description, CTA) and the linked listing context if any	Suggested headline (≤ 100 characters), description (≤ 125 characters), CTA (≤ 40 characters); 25-second timeout
Message moderation	Scans chat messages for policy violations	Automatic on message send and receive	The single message text and minimal context (claim ID for grouping)	Flag / no-flag verdict for human review

ShareFree does not today use AI to:

Issue sanctions, suspensions, or bans
Determine appeal outcomes
Set ad pricing dynamically
Personalize content recommendations or feed ranking
Match users to listings or to each other
Score, rank, or otherwise profile users

If we expand AI usage into any of the above or into new feature areas, we will update this Notice and notify users in accordance with our Privacy Policy change-notification process.

2. AI Providers

We rely on a fallback chain managed in our callLLMWithFallback server-side helper:

Primary provider: OpenRouter, routing prompts to Google Gemini 2.5 Flash.
Retry tier: OpenRouter is retried automatically on timeout or HTTP 5xx errors.
Fallback provider: Google Gemini API, Gemini 2.5 Flash, called directly when OpenRouter is unavailable or returns a 429 rate-limit response.
Hard-stop conditions: Content-policy violations from the model, 4xx client errors, and parse errors stop the chain immediately rather than retrying.

The provider that handled any individual request is recorded in our internal ai_audit_logs table along with the model name, latency, and outcome. We do not currently use OpenAI, Anthropic, Meta, or any other LLM provider in the production AI pipeline.

When you use an AI feature, the provider's terms apply to their handling of the prompt and response in addition to ours. The relevant provider terms are:

OpenRouter Privacy Policy: https://openrouter.ai/privacy
Google Gemini API Terms / Generative AI Privacy: https://ai.google.dev/gemini-api/terms and https://policies.google.com/privacy

3. What We Send to AI Providers

We send the minimum data necessary for each feature to function:

Image moderation and image analysis features receive: the image bytes you uploaded, plus listing context (category if you have chosen one, listing type if you have chosen one). They do not receive your name, email, account ID, or other listings you have authored.
Listing copy features ("Fill with AI") receive: the structured output of the prior image analysis, the text fields you have already filled in on the listing form, and your selection state (type, category).
Ad copy enhancement receives: the existing headline, description, and CTA you have entered in the ad form, plus the linked listing context if the ad references a listing.
Message moderation receives: the text of the single message being sent and a claim ID used internally to group related messages. It does not receive your message history with any other user, the recipient's identity beyond what's required by our system, or any account metadata.

Where context fields contain user-authored text you have entered yourself (e.g., a draft listing description), that text is part of the prompt. We do not strip user-supplied free text before sending it.

4. What We Do Not Send

We never include the following in any prompt to any AI provider:

Account credentials — passwords, OAuth tokens, magic-link tokens, session cookies, or refresh tokens
Email address or phone number — these are not part of any AI prompt envelope
Other users' messages — message moderation processes one message at a time and does not receive the recipient's prior messages or the broader conversation history
Payment information — Stripe identifiers, IAP receipts, card data, payout details
Pickup addresses or precise geolocation — addresses agreed for pickup or meetup are never sent; location precision in any prompt is no finer than city/community level
Reports, sanctions, appeals data, or moderator notes — none of this is exposed to LLMs
Trust scores, reporter trust state, or community standing data — never sent
Private demographic attributes we do not collect and do not infer (race, religion, sexual orientation, disability, immigration status, etc.)
Children's data — minors are not permitted on the platform per our Terms; we do not knowingly send minors' content to AI providers

If you discover content in a prompt envelope that you believe should not have been sent, contact privacy@sharefree.org.

5. Provider Retention and Training

OpenRouter

OpenRouter's Privacy Policy permits limited retention of prompts and responses for abuse prevention and operational purposes. Our integration is configured with API keys and request parameters that disable training-on-prompts where the provider exposes such a control. We rely on OpenRouter's published terms; we do not have a direct contractual data-processing agreement that overrides them. You should review OpenRouter's current Privacy Policy for the authoritative statement of their retention practices.

Google Gemini API

Google's Gemini API terms commit that prompts and responses sent to the Gemini API are not used to train Google's general-purpose models without separate, explicit consent. This is distinct from the consumer Gemini app, where different terms apply. ShareFree uses the Gemini API exclusively, not the consumer app, and we rely on Google's API-tier commitment.

Our Own Audit Logs

We retain our own AI audit logs in the ai_audit_logs table for 12 months. Each log entry contains:

The prompt envelope (the structured payload we sent)
The response envelope (what the model returned)
The provider used (OpenRouter or Gemini direct)
The model name and version
Request latency and outcome (success / error class)

These logs are used for debugging, abuse detection, capacity planning, and compliance review. They are not used to train any model — neither ours, nor any third party's. Access is limited to engineering and trust-and-safety personnel with a legitimate need.

If you exercise a deletion right under the Privacy Policy, the user-authored content within your audit-log entries (your message text, your draft listing copy) is purged on the schedule described in the Privacy Policy. Provider-side retention is governed by the provider's terms and is outside our direct control.

6. Opt-Out and User Control

Not all AI features are equally optional. We distinguish between safety-required AI and convenience AI.

Required (no opt-out)

Image moderation on listing upload. When you upload a listing photo, it is sent to the moderation pipeline. This is a condition of using the Service. If you do not consent to AI moderation of your uploaded photos, do not upload photos.
Message scanning. Outbound and inbound chat messages are scanned in real time for policy violations. Consent is obtained at first message in accordance with Privacy Policy §15. If you do not consent, do not use the messaging feature.

These are required because the Service cannot meet its safety obligations to other users without them. They are limited in scope to what is necessary for moderation.

Optional (opt-in by user action)

Fill with AI (listing creation): runs only when you tap the button.
Suggest with AI (analyze-listing-image): runs only when you tap the button.
Enhance Ad (enhance-ad-copy): runs only when you tap the button.

These features are entirely optional. The listing and ad creation flows work end-to-end without ever invoking them. You can use ShareFree indefinitely without ever triggering an opt-in AI suggestion.

Deletion

You may delete listings, messages, ads, and your account at any time using the in-app controls described in the Privacy Policy. Deletion removes the content from our systems within the retention windows specified there. Provider-side traces are governed by the provider's retention terms.

7. Accuracy and User Responsibility

AI output is often wrong. Generative AI systems, including the Gemini family, are known to:

Fabricate facts that sound plausible (commonly called "hallucination")
Misidentify objects, brands, conditions, or features in images
Suggest categories that do not fit the actual item
Produce ad copy that overstates the value, condition, or capabilities of an item
Produce ad copy that understates required disclosures or safety information
Generate text that incidentally infringes third-party copyright or trademark
Produce output reflecting biases present in training data

You are responsible for everything you publish. Before submitting any AI-suggested listing, ad, or message:

Read every word the AI produced
Verify the description matches the actual item in front of you
Verify any factual claim ("works perfectly," "originally $200," "still under warranty") is accurate and substantiable
Confirm the suggested category and listing type are correct
Ensure the content complies with our Acceptable Use Policy and, for ads, the Advertising Policy
Remove or correct any output that misrepresents the item

The submission flows for listings and ads include an attestation that the final content is accurate and lawful. AI assistance does not reduce or transfer that responsibility. It is your post and your ad; you are accountable for it.

8. No Medical, Legal, Financial, or Tax Advice

AI features in ShareFree exist to assist with listing creation, ad enhancement, and content moderation. They are not designed for, validated for, or appropriate for:

Medical advice, diagnosis, treatment recommendations, or symptom triage
Legal advice or interpretation of contracts, statutes, or regulations
Financial, investment, tax, accounting, or insurance advice
Mental-health counseling, addiction-recovery guidance, or crisis intervention
Safety-critical determinations (electrical code, structural, automotive mechanical)

If an AI suggestion strays into any of these domains — for example, suggesting therapeutic claims for a wellness product, or implying a legal status for an item — disregard the suggestion and consult a licensed professional. ShareFree expressly disclaims any reliance on AI output for any of the above purposes. To the maximum extent permitted by law, we are not liable for harm arising from such reliance.

9. Biometric Data and Sensitive Attributes

ShareFree's posture on biometric identifiers and sensitive-attribute inference:

We do not extract or store biometric identifiers (faceprints, voiceprints, retinal patterns, fingerprint geometry, or hand geometry) from images you upload. Our image pipeline does not perform face recognition, face matching, or any one-to-one or one-to-many biometric comparison.
We do not instruct LLM providers to extract biometric identifiers, and where the provider exposes a configuration to disable face-detection or facial-feature outputs, we have set those parameters to off.
We do not profile users by inferred protected-class characteristics (race, ethnicity, religion, sexual orientation, gender identity, disability, national origin, age beyond adult/minor for eligibility) using AI or any other means.
Our compliance posture under the Texas Capture or Use of Biometric Identifier Act (CUBI) and the Illinois Biometric Information Privacy Act (BIPA) is set out in Privacy Policy §17.

If we ever expand AI usage into anything that would constitute the capture, use, or storage of a biometric identifier under CUBI, BIPA, or any analogous state or federal law, we will obtain prior, affirmative, written consent before doing so, and we will update this Notice and the Privacy Policy.

10. Automated Decision-Making and the Right to Appeal

The following actions in ShareFree are taken automatically in whole or in part and may affect your account or your content:

AI moderation flagging or rejection of a listing photo at upload time, or of a chat message at send time
5-pending-reports auto-disable trigger, which temporarily disables a listing once five distinct reports are pending — this is a rule-based threshold, not an AI decision
Reporter Trust auto-status changes (probation, suspension, revocation), which adjust how your reports are weighted based on accuracy history — also rule-based, not AI

Under the EU General Data Protection Regulation Article 22, the California Consumer Privacy Act / CPRA, and analogous state laws, you have the right to:

Be informed that an automated decision has affected you. We notify you at the time of action (in-app banner, email for account-level actions).
Obtain meaningful information about the logic involved. The descriptions in this Notice and the Privacy Policy provide that information at the level required by law; we will provide additional detail on request.
Appeal any automated decision through our standard appeal process. Appeals are submitted via the in-app appeal form. Appeal cooldowns are 7 days for community-level actions and 30 days for platform-level actions, as set out in Terms §19.4.
Request human review of any AI-driven moderation outcome by emailing appeals@sharefree.org. A human moderator will re-review the content and the decision.

ShareFree does not use AI to issue final sanction or ban decisions today. All sanction-level decisions — account suspension, account termination, community removal — involve human review by a member of the trust-and-safety team. AI flags content; humans decide consequences.

11. Changes to This Notice

We will update this Notice when:

We change AI providers or add a new provider to the fallback chain
We add a new AI-driven feature, or remove an existing one
A provider materially changes its retention or training terms
Regulatory developments — including the EU AI Act, Colorado AI Act, California ADMT regulations, Texas Responsible AI Governance Act, or analogous laws — require additional disclosures
We expand AI usage into any of the areas listed as "not currently used" in §1

Material changes will be announced through the in-app notification system and via email, on the timeline described in the Privacy Policy. The version and last_updated fields at the top of this Notice are authoritative.

Contact

Privacy and AI questions: privacy@sharefree.org
Appeals of automated decisions: appeals@sharefree.org
Trust and safety: safety@sharefree.org