Tüm belgeler Türkçe English

Çiftçi AI — Artificial Intelligence Disclaimer

Data controller / Service provider: HNK Tech

Application: Çiftçi AI (com.ciftciai.app)

Effective date: 14.08.2026 · Version: 1.0


1. What this document covers

Some features in Çiftçi AI are powered by artificial intelligence (AI). This

document explains:

> The short version: AI output in this application is **for information

> purposes only. It does not replace** the advice of an agricultural

> engineer, a veterinarian, the Ministry of Agriculture and Forestry, or your

> provincial/district agriculture directorate. For pesticide dosage, quarantine

> periods and pre-harvest waiting periods, **always follow the product label and

> official sources.**


2. Which features use AI

FeatureWhat the AI doesHow reliable it is
AI Advisor chatProduces free-text answers to farming questionsGeneral knowledge only; it does not see your field or know your soil
Plant disease detectionExamines an uploaded photo and suggests a possible disease, severity and treatmentNot a definitive diagnosis — based on visual similarity
Satellite index interpretationTranslates satellite-derived numbers (NDVI etc.) into plain languageThe numbers are measurements; the interpretation is an AI estimate
Irrigation adviceSuggests an irrigation plan from the weather forecast and field dataDepends on the forecast; if the forecast changes, the plan is wrong
Content moderationAutomatically screens community posts and listingsMay block wrongly or miss violating content

No other feature in the application uses AI. Weather data, fire alerts, the

satellite index calculation itself, laboratory results, timesheet records and

marketplace listings are not AI-generated — they are measurements, calculations

or user input.


3. How the AI works

3.1 Tiered model flow

Every AI request is handled in this order:

1. The local model on our own server (Ollama) is tried first. At this stage

your data does not leave our server.

2. The local model returns an answer together with its own confidence score

(0–1). If the score is above the threshold (default 0.65), the answer is

shown to you as is.

3. If the score is below the threshold, or the local model cannot produce an

answer, the request is sent to an external provider located abroad:

Google Gemini, OpenAI or Anthropic. The order in which providers

are tried is set from the administration panel; if one fails, the next is tried.

4. If no model can respond, you are told "analysis is currently unavailable."

The system never fabricates a result and never fills the gap with a guess.

3.2 An important difference for photos

Our local model runs on limited hardware for image processing. For this reason

**plant disease photos are, by default, sent directly to an external provider

abroad** — the local step may be skipped entirely. Read Section 4 before

uploading a photo.

3.3 A confidence percentage is not a guarantee

The confidence percentage you see on screen (e.g. "82% confidence") is a

score the model assigns to itself. It is not an independent accuracy

measurement, a laboratory result, or a statistical guarantee.

Treat the confidence percentage as a rough hint about how carefully you should

proceed — not as a basis for a decision.


4. Cross-border transfer warning

> WARNING — please read this.

> Your questions and the plant photos you upload **may be sent to servers

> located outside Türkiye** so that an answer can be produced.

4.1 What goes where

ProviderWhat is sentWhere it is processed
Google (Gemini)The text of your question, chat history, uploaded photoAbroad
OpenAISameAbroad
AnthropicSameAbroad

What is transmitted is the text you write. Your account name, phone number,

Turkish national ID number, IBAN and field boundaries are not sent separately

to external providers — but if you type them into the chat box, they travel as

part of the text.

4.2 Do not type these into the chat

Do not enter the following into the AI chat, a disease photo description, or

an irrigation note:

When uploading a photo, make sure the frame contains **no faces, licence plates,

signs, addresses or documents**. Capture only the plant, leaf, fruit or soil.

4.3 Legal basis and withdrawing your consent

This transfer relies on your explicit consent under Article 9 of Law No.

6698 on the Protection of Personal Data (KVKK). ("Explicit consent" means

consent that is informed, freely given and related to a specific subject.)

app, or write to [email protected].

(field records, marketplace, weather, community) continues to work.


5. AI makes mistakes

5.1 What a "hallucination" is

Language models sometimes produce **completely fabricated information in a

confident tone**. This is called a "hallucination." A model may invent a disease

name, a non-existent pesticide, an incorrect dosage or a regulation article that

does not exist — and present it as fact.

This is not a malfunction of the application; it is a known limitation of

today's AI technology and cannot be fully eliminated.

5.2 Common types of error

Error typeExampleWhat you should do
Wrong diagnosisSaying "rust" when it is powdery mildewShow a sample to an agricultural engineer or laboratory
Invented product or doseA product name that is not on the market, a wrong dosageFollow the product label only
Outdated informationRecommending an active substance whose licence was revokedCheck the Ministry's current list
Region-inappropriate adviceAdvice valid for a different climate or soilConsult your provincial/district agriculture directorate
Unit or scale confusionMixing up decares and hectares, litres and millilitresVerify the calculation yourself
Misreading an imageWrong reading due to light, shadow or blurTake a new photo from a different angle
Losing contextForgetting the crop mentioned in your earlier messageWrite the full question in a single message

5.3 Limits of photo-based detection

Plant disease detection is based on visual similarity. The result is clearly

unreliable when:

frost, chemical burn or water stress,

Most viral, bacterial and soil-borne agents **cannot be distinguished from a

photograph**; laboratory analysis is required for a definitive result.


6. Non-binding advice

**No AI output in this application constitutes binding advice, an expert report,

a prescription, a licence, a certificate or an official opinion.**

AI output does not replace:

6.1 Pesticides, dosage, quarantine and pre-harvest waiting periods

> This section is critical.

When using plant protection products (agricultural pesticides):

timing is the product's licensed label.** If a figure given by the AI conflicts

with the label, the label prevails.

quarantine / re-entry period. These vary by product, active substance and

crop; applying them incorrectly may lead to your produce being **destroyed

because of residues, rejected for export, and to administrative fines.**

published by the Ministry of Agriculture and Forestry. Unlicensed or off-label

use is subject to sanctions under Law No. 5996 on Veterinary Services, Plant

Health, Food and Feed.

prescription** is mandatory and may be issued only by authorised persons. This

application is not a substitute for a prescription.

6.2 Animal health

This application is primarily aimed at crop production. If you ask a question

about animal health, the answer you receive is general information.

immediately.**

be used under a veterinarian's prescription and supervision. The AI **cannot

issue a prescription.**

report it to your provincial/district directorate of agriculture and forestry.

The application does not make this notification on your behalf.

a veterinarian and the product leaflet.

6.3 Official procedures and subsidies

For subsidy applications, farmer registry records, licences, certificates,

insurance and incentives, AI output should not even be treated as guidance.

Legislation changes frequently and the model may not be current. Rely on the

announcements of the competent authority.


7. Satellite, weather and alert data

Satellite indices, weather forecasts and fire alerts in the application come from

third-party sources and carry their own limitations. AI interpretations built

on this data inherit the same limitations.

DataSourceLimitation
Satellite indices (NDVI etc.)Google Earth EngineThe image may be days old; cloud, shadow and snow distort results
Weather forecastOpen-MeteoA forecast, with no guarantee of occurrence
Fire hotspotsNASA FIRMSDelayed data tied to satellite overpass times; small fires may be missed
Map tilesStadia MapsBoundary and parcel drawings are not cadastral documents

> Frost, hail, flood, storm and fire alerts do not replace official warnings.

> Where life or property is at risk, rely on official sources — in particular the

> Turkish State Meteorological Service (MGM), AFAD and the emergency

> number 112. The absence of a notification in the app does not mean there is

> no danger, and receiving one does not mean the event will occur.

When satellite imagery cannot be retrieved or cloud cover is high, the

application tells you so explicitly — it does not substitute an estimated number.


8. Moderation AI

Community posts, comments, listings and messages pass through an automatic

filter. This filter also uses AI and can err in both directions:

An automated decision is not final. If your post was blocked unfairly, you

may request human review by writing to [email protected]. You can

likewise report violating content using the in-app report and block user

tools. Reports are reviewed by a human.

The operation of the automatic filter does not remove our notice-and-takedown

obligations under Law No. 5651.


9. Transparency: which model produced it, what is recorded

9.1 What you are shown

displayed in the interface.

factors relied upon.

9.2 What is recorded

RecordContentsWhy
Chat messagesYour question, the answer text, provider and model nameSo you can see your history and continue the conversation
Disease analysisDetected disease, severity, confidence score, recommendation text, sourceSo your field history can be tracked
AI call logTask type, provider, model, confidence, duration, token and cost data — the question and answer text are NOT part of this recordAuditing, cost control and abuse prevention
Usage counterDaily usage timeTracking the free usage quota

Your chat history and disease analyses are tied to your account; they are

deleted when you delete your account (see the Account Deletion document). The

user link in AI call logs is also removed on account deletion.

How long external providers retain your data on their side is governed by

their own policies; HNKTech cannot determine that period unilaterally. This

is precisely why the "do not type personal data" warning in Section 4 matters.


10. Limitation of liability

AI features are provided "as is." HNKTech:

suitable for your specific conditions;

fertiliser costs, rejection of produce due to residues, animal loss,

administrative fines or similar damages arising from actions you take based on

AI advice;

satellite, weather and map services;

missed content by automatic moderation.

The final decision is yours. It is your responsibility to consult a qualified

specialist before acting on decisions concerning spraying, irrigation, planting,

harvesting and animal health.

These limitations do not cover HNKTech's liability arising from its **intent or

gross negligence** (Turkish Code of Obligations No. 6098, Article 115), and they

do not limit the mandatory provisions of Law No. 6502 on Consumer Protection that

operate in favour of consumers. Your statutory consumer rights are reserved.

The application is intended for users aged 18 and over.


11. Emergencies

In the following situations, do not wait for the application — contact the

authorities directly.

SituationContact
Fire, flood, earthquake, danger to life112 / AFAD
Forest fire112
Suspected pesticide poisoning (human)112 / National Poison Information Centre 114
Sick or dead animal, suspected outbreakVeterinarian / provincial-district agriculture and forestry directorate
Notifiable plant disease or pestProvincial/district agriculture and forestry directorate

12. Reporting bad output and giving feedback

If you see an answer from the AI that is wrong, dangerous or fabricated, please

report it — these reports are our main way of correcting the system.

1. Use the in-app Feedback screen. You may attach a screenshot; the app

version, platform and device model are attached to the feedback automatically.

2. Or write to [email protected]. Where possible, include a screenshot of

the answer and the date and time.

Output that poses a serious safety or health risk is reviewed as a priority, and

the relevant feature may be temporarily disabled.

You may exercise your rights under KVKK Article 11 (to learn whether your data is

processed, to request rectification or erasure, to request restriction of

processing, and to object where an outcome to your detriment arises from analysis

by automated systems) via [email protected] or —.

Applications are concluded within 30 days at the latest, pursuant to KVKK

Article 13.


13. Changes

The AI providers, models and confidence threshold we use may change over time. If

a change materially affects the scope, this document will be updated and you will

be informed in the application. If the scope of cross-border transfer expands,

your explicit consent will be requested again.

The effective date and version of this document appear at the top.


Contact

E-mail: [email protected]

Registered electronic mail (KEP): —

Address: Akyurt / Ankara, Türkiye

VERBİS: kayıt yükümlülüğü bulunmamaktadır

MERSIS: — · Tax office / number: —


Sürüm 1.0 · Çiftçi AI