AI Use Principles

Fidkeep uses AI to help businesses reply to their Google reviews faster, more consistently and in line with their brand, and to see which topics stand out in their reviews. AI is a tool that helps prepare replies; the decision whether to publish a reply, and responsibility for its content, rest with the Customer.

Fidkeep does not use AI to profile reviewers, to score or classify individuals, or to take decisions about individuals that produce legal effects. Review topics and sentiment classifies the content of a review, not people: the labels show which topics a review mentions and with what sentiment; they are not used to profile, score or classify the reviewer or any other person and have no legal effect on anyone.

• Reply drafts: For each review, a reply draft is prepared in line with the brand-voice settings of the relevant location.
• Controlled Automatic Replies: For reviews that match rules the Customer has enabled, a draft is prepared and published to Google without separate approval.
• Review topics and sentiment: Each review is split into topics determined by the location's industry (for example service, food/taste, price, cleanliness, waiting time), and a separate sentiment label (positive, negative or neutral) is assigned to each topic. For example, “The food was great but the service was very slow” is labelled positive for taste and negative for speed. Each industry (for example restaurant, café, hair salon, health, retail) has its own topic set. Tagging works for reviews in Turkish, English, Russian, Arabic and other languages; reviews synchronised earlier are tagged as well. The labels are shown in the review list, the review detail and the dashboard.
• Location setup suggestions: Language, tone and reply preferences are suggested for a new location.
• Report narratives: Summary and commentary text is prepared for the data in daily and monthly location reports.
• Feedback structuring: Product feedback that Authorised Users send to Fidkeep is categorised and passed to Fidkeep's internal issue-tracking tool as a structured record.

Review synchronisation, the inbox, role and access management and the transmission of replies to Google do not use AI. The topic and sentiment filters on the Reviews screen rely on AI-generated labels; other filters do not use AI. Smart rules are evaluated on a rule basis: whether a new review matches a rule, and the sending of e-mails to the team members concerned, are determined by the conditions the Customer sets, without AI; only the rules' topic condition, offered as a beta, uses those labels.

Fidkeep's AI features use language models developed by third-party providers. Which providers and models are used is kept confidential by Fidkeep; to the extent necessary for customers to exercise their rights under the Data Processing Agreement, this information is disclosed on request and subject to confidentiality obligations. The AI providers are subprocessors located in the United States; international transfers to these providers and notice of subprocessor changes are governed by the Data Processing Agreement and the Subprocessors page.

Fidkeep may switch between the models it uses for quality, cost and security reasons. Adding a new AI provider that will process customer personal data is notified as a subprocessor change.

Only the data needed to perform the relevant task is sent to AI models. Depending on the feature, the data sent is:

• Reply drafts and Automatic Replies: the business and location name and category; the review rating and text; the reviewer's display name; the location's language and tone, preferred and prohibited phrases, signatures, example replies and Authorised User instructions.
• Location setup suggestions: business and location information and selected review information for the location.
• Report narratives: rating and review statistics for the reporting period and selected review information.
• Review topics and sentiment: the review text and rating and the topic list for the location's industry.
• Feedback structuring: the feedback text, the selected category and sentiment, the path and title of the page from which the feedback was sent, and the active location.

User passwords, Google OAuth tokens, payment and card details, Authorised Users' e-mail addresses and reviewers' profile photos are not sent to AI providers.

Review texts may contain health or other special-category information published by the reviewer. Fidkeep does not seek, extract or use such information for profiling; however, the review text is passed to the model as it is in order to prepare a reply draft and to determine topic and sentiment labels. The Customer must not enter personal data, and in particular special categories of personal data, into instructions, example replies or brand-voice settings.

Fidkeep keeps records of AI operations (model, token counts, cost, status and request IDs) for 12 months. Generated drafts are kept in the Customer's Workspace for the term of the Agreement and deleted within 90 days of account closure at the latest.

Fidkeep does not use Customer Content, reviews, replies, drafts or brand-voice settings to train its own models or any AI model. Fidkeep works with providers under terms under which data sent through the API is not used for model training. How long and for which purposes (for example abuse prevention) providers may retain data is governed by their own contractual terms.

A location's example replies and instructions are used only to prepare drafts for that location and are not used in drafts for any other customer.

AI output is a draft by default. A draft is not published to Google unless an Authorised User has reviewed and approved it; the only exception is Automatic Replies that the Customer has explicitly enabled. Topic and sentiment labels are not published to Google; they are shown only within the Workspace and used for filtering and smart rules.

Fidkeep recommends that every reply be read, before publication, by an Authorised User who knows the review and its context. The Authorised User can edit the draft, regenerate it or leave it unused. The content of an approved reply is the Customer's statement.

• Opt-in: Automatic Replies are off by default. They are enabled only explicitly, by an authorised Workspace user.
• Rules: The Customer determines, using the settings the Service provides, which reviews Automatic Replies apply to, for example by rating.
• Switching off: Automatic Replies can be switched off at any time; no new automatic publications are made after they are switched off.
• Responsibility: The Customer is responsible for the decision to enable Automatic Replies, for the rules and for the content of all replies published this way.
• Recommendation: Fidkeep recommends using Automatic Replies only for positive reviews with limited content, having low-rated reviews and reviews containing complaints answered by a person, and considering Automatic Replies carefully for healthcare providers and public institutions.

Automatic Replies also count towards the daily AI quota. Fidkeep may stop Automatic Replies where there is a suspected security issue or breach of the Acceptable Use Policy.

AI output is generated probabilistically and can be wrong. In particular, it may:

• misunderstand a review or miss a complaint in it;
• invent services, campaigns, opening hours, prices or facts that do not exist;
• make commitments to refunds, compensation or actions that should not be promised;
• use a tone that does not suit the situation, such as overly formal, overly familiar or defensive;
• make language, spelling or grammar errors, or reply in a language different from that of the review;
• produce very similar replies to similar reviews.

Report narratives may also contain incorrect interpretations or incomplete assessments; business decisions should be made together with the data in the report.

Topic and sentiment labels are also AI output and can be wrong: a topic may be missed or assigned incorrectly, and irony, slang or mixed feelings may be misread. This can also affect filter results based on labels, topic breakdowns on the dashboard and matches of smart rules that use a topic condition. Labels should not be used as the sole basis for business decisions or for assessments of employees; they should be considered together with the review itself.

AI output is not legal, medical or other professional advice.

For each location the Customer can set the language, tone, preferred and prohibited phrases, signatures, example replies and free-form instructions. These settings directly affect the content of drafts and are the Customer's responsibility.

• Instructions and examples must not aim to produce content that breaches the Acceptable Use Policy (for example belittling the reviewer, making misleading statements or keyword stuffing).
• Personal data of reviewers or employees and special categories of personal data are not entered into instructions, examples or signatures.
• If a personal name is used in signatures, informing the employee concerned is the Customer's responsibility.
• Settings should be reviewed regularly, and changing information such as campaigns, prices or opening hours kept up to date.

Whether to disclose that replies are prepared with AI assistance is the Customer's decision; where legislation or platform rules require such a disclosure, compliance is the Customer's responsibility.

For healthcare providers (hospitals, clinics, practices, pharmacies and similar):

• instructions should be set so that replies use general, neutral language that does not confirm the reviewer is a patient;
• if a draft contains diagnosis, treatment, medication, appointment, physician or department information, it should be removed before publication;
• drafts should not contain medical advice or promises of outcomes, and complaints should be directed to official patient communication or patient rights channels;
• not using Automatic Replies for low-rated reviews and reviews containing complaints is recommended.

Review topic and sentiment tagging also applies to reviews in sensitive sectors such as health. The labels only classify the content of the review against the topic list for the location's industry; they are not used to draw conclusions about the reviewer's health or other characteristics or to establish that the reviewer is a patient.

Public institutions assess AI-prepared replies in accordance with their own legislation, institutional communication rules and approval procedures; replies should not contain information about applications, files, proceedings or individuals. Detailed rules are set out in section 4 of the Acceptable Use Policy.

The AI features may not be used to:

• produce fake or incentivised review text or write reviews to be posted on other platforms;
• produce replies that threaten, harass, humiliate or discriminate against reviewers;
• extract personal information about reviewers, identify them or build profiles of them;
• produce replies containing untrue advertising, health, price or legal claims;
• steer the model to bypass its restrictions, reveal system instructions or produce content for purposes outside the Service;
• use AI Output in bulk to develop a competing product or train a model;
• open multiple accounts or use automation to exceed daily quotas.

Each Plan's daily AI reply quota is shown on the pricing page and the order screen. The quota resets every day; unused quota does not carry over to the next day and is not refundable. When the quota is used up, no new AI drafts or Automatic Replies are generated until it resets; during that time Authorised Users can write and publish replies themselves. Additional quota can be purchased as an add-on. Fees and add-on terms are governed by the Subscription and Payment Terms.

If you encounter AI output that is wrong, inappropriate or harmful, you can use the in-app feedback tool or write to [email protected]. We review your reports to improve brand-voice settings and system instructions.

If you notice an error in a published reply, you can update the reply through the Service or remove it through Google Business Profile. Reviewers and other data subjects first direct requests about the processing of their personal data to the business that is the data controller; requests that reach Fidkeep are handled under the Data Subject Request Form and the Data Processing Agreement.

These Principles are part of the Terms of Service. Personal data processed on the Customer's behalf within the AI features is governed by the Data Processing Agreement; data Fidkeep processes as data controller is governed by the Privacy Policy and the KVKK Privacy Notice; prohibited uses are governed by the Acceptable Use Policy. In case of conflict, the order of precedence in the Terms of Service applies.

These Principles may be updated as AI technology, providers and the relevant legislation change. Material changes are announced by e-mail and in-app notice at least 30 days in advance, in accordance with the Terms of Service. When a new AI feature is made available, these Principles are updated to show the data that feature uses.