At a glance
Where farm assistance is enabled, recommendations require review by default; eligible replies can follow a separately configured automatic path. AI responses can be wrong. Advice-content sharing for improvement has separate opt-ins and does not automatically train a model.
Information used for an answer
Where enabled and consented, Openfarms Lab combines your question with relevant information about the selected farming season: crops, varieties, planting dates, stage, activities, observations, administrative area, language and conversation context. The selected season matters; tell your organisation if the wrong farm or season is associated with a conversation.
Supported photos may be sent to a model capable of analysing images. Voice notes may be processed by an audio-capable model or a transcription service. Media compatibility and provider availability affect what the AI can assess. A description of a photo is not a reliable diagnosis.
Models and research tools
The service can use more than one model or provider, including a fallback if a preferred route cannot answer. A research tool can retrieve publications or search the web. A weather tool can request information about a named place. Search queries are generated for the agricultural question and filtered for contact details; do not assume they are guaranteed anonymous.
Model calls and processing traces can contain personal content. Working sessions let the AI continue a conversation, and their lifecycle is separate from the longer-lived platform record. See AI processing in our Privacy Policy and Service providers.
Automation, review and escalation
Where farm assistance is enabled, the default pilot policy holds AI recommendations for an authorised person to review. Automatic recommendation approval requires staff to make each applicable category eligible, the organisation’s settings to permit it, and the draft to satisfy confidence and other checks. A general clarification without detected risk can pass automatically under its own checks. An automated reply therefore does not mean a person has reviewed it.
A response can be held for review, blocked or referred to a person. Failed or timed-out processing can also be referred for follow-up. These controls do not prove that an answer is correct or that a qualified agronomist has checked it. Review staffing and response times depend on the organisation’s service arrangement.
Ask for a person through the available conversation or your organisation’s support contact if you need human help. Proposed changes to a farm activity or observation require the relevant confirmation before becoming established records; an AI statement alone is not evidence that an activity happened.
AI assistance is not presented as a land-title decision, credit decision, insurance assessment or guarantee of eligibility for another service. A new use involving decisions with legal or similarly significant effects needs its own assessment, notice and safeguards.
Practical limitations
- Information can be incomplete, outdated, mistranslated or unsuitable for your local conditions.
- References, confidence scores and photographs do not establish accuracy on their own.
- Confirm chemical labels, permitted uses, rates, protective equipment and waiting periods with a qualified professional.
- Do not rely solely on an automated answer for a consequential farm or safety decision.
Our Terms of Service explain agricultural advice limitations. The public website chat answers company questions and is separate from the farm-assistance workflow.
Optional improvement and your choices
Consent for AI advice permits processing needed to answer a question. Making eligible question, recommendation or correction text available for optional improvement requires both the person’s advice-improvement consent and the organisation’s training opt-in. An organisation that has not opted in is treated as having declined.
These controls govern optional advice-content sharing. They do not start a training job or mean that a conversation automatically trains a model. Working conversation sessions and processing traces support the operation of the AI service; recording them is separate from model training. Ratings, approval outcomes, identifiers and category codes can also be recorded separately for service evaluation and may still relate to a person.
You can ask your organisation or contact us to withdraw the relevant consent. Withdrawing advice-improvement permission stops further consent-dependent content sharing; it does not withdraw your separate permission for AI advice. Withdrawing AI-advice consent can stop the assistance that depends on it. Neither action automatically deletes existing records. Use Data rights & deletion to request a records review.