AI Usage Policy

Last updated 6 September 2026  ·  Varnan Digital (OPC) Private Limited  ·  CIN U62090KL2025OPC096808

AI is not a garnish on what Varnan does; it is in the infrastructure. That makes it something we owe you an honest account of. This policy names the models we use, what we let them do, what we refuse to send them, and where a human still has to sign off.

1. The models we use

Models change quickly. When we move to a materially different model or provider, this page is updated. The current list is accurate as of the date at the top.

2. What AI actually does here

On this website

Articles in our journal are produced by an AI-assisted pipeline: a model researches the topic, drafts the piece in structured blocks, and proposes titles, meta descriptions and schema. Nothing publishes itself. Every post passes a quality gate and human review before it goes live, and we correct or remove anything that turns out to be wrong. Where a piece was substantially drafted by a model, we are happy to say so on request.

AI does not read your form submissions, does not personalise what you see, and does not decide whether we reply to you.

In client work

• Research and strategy — market, keyword and competitor analysis, audience research, campaign structure proposals.
• Content and copy — first drafts of articles, ad copy, landing pages and email sequences, always edited by a human before delivery.
• Creative — concept generation and variant testing for ad creative.
• Automation — the workflow systems we build for clients, which may classify, summarise, route or draft on the client’s behalf.
• Analysis — reading performance data and surfacing what changed and why.

Where a deliverable was AI-assisted, we will tell you if you ask, and we will always tell you before an automation we build starts producing content that goes out under your name.

In our product

Entuer, our WordPress plugin, uses Claude to generate and optimise content inside a customer’s own site. Customers on managed AI have their requests routed through our Cloudflare Worker proxy to Anthropic under our commercial account; customers on bring-your-own-key call their chosen provider directly with their own credentials.

3. What we never send to a model

Prompts are not a safe place for other people’s secrets. We do not send:

• Credentials, API keys, tokens or passwords.
• Payment card data, bank details or identity documents.
• Special category data such as health, biometric, religious, caste or political information.
• Personal data belonging to a client’s customers, unless the engagement specifically requires it and the client has instructed it in writing.
• Material a client has marked confidential, where sending it is not necessary for the task.

Where a task genuinely needs personal data, we minimise and pseudonymise it first, and we prefer sending a description of the data to sending the data itself.

4. Training and data retention

We use commercial API access, not consumer chat products, for anything touching client work. Under Anthropic’s and OpenAI’s API terms, inputs and outputs are not used to train their models. We do not grant any AI provider permission to train on our content or our clients’ content, and we will not accept a provider that requires it.

Providers may retain API traffic briefly for abuse monitoring under their own terms. Our own logs of AI-assisted work are kept with the relevant project record and deleted on the schedule in the Privacy Policy.

5. Human accountability

A named person at Varnan is responsible for every AI-assisted output that reaches a client or the public. In practice that means:

• Facts get checked. Models fabricate confidently. Statistics, quotes, citations, legal and medical claims and anything about a named company are verified against a primary source before publication.
• Claims get evidence. We do not publish a performance claim on a client’s behalf that we cannot substantiate.
• Nothing ships unreviewed. Automated publishing without a human gate is not something we run for ourselves or set up for clients without explicit written agreement and a rollback plan.
• Errors get fixed. If an AI-assisted output is wrong, we correct it, tell whoever it affected, and look at why the gate missed it.

6. Automated decisions

We do not use AI to make decisions that produce legal or similarly significant effects about a person without human involvement. Ad platform bidding algorithms decide which impressions to buy, which is an automated decision about an auction rather than about a person’s rights. Where a client asks us to build a system that scores, ranks or filters individuals, such as lead qualification or applicant screening, we require a human decision-maker in the loop, a documented basis for the criteria, and a route for the affected person to ask how a decision was reached.

7. Bias, accuracy and limits

Language models reproduce patterns in their training data, including unfair ones. We review AI-generated audience definitions, creative and copy for discriminatory framing, and we apply advertising platform policies on sensitive categories such as housing, employment, credit and health. We also accept a plain limit: a model does not know your business, your customers or your market. It accelerates the work. It does not replace the judgement.

8. Client control

You can tell us to limit or stop AI use on your account. Options include restricting AI to research only, requiring disclosure on every AI-assisted deliverable, excluding your data from any model call, or opting out entirely. Ask and we will record it in your engagement documents. Excluding AI usually means longer timelines and different pricing, and we will tell you what that looks like before you decide.

9. Intellectual property

Deliverables we produce for you, whether AI-assisted or not, transfer to you on payment as set out in the Client Service Terms. AI-generated material may not attract copyright protection in some jurisdictions, and identical or similar output can in principle be produced for someone else. We check work for originality before delivery, but we cannot guarantee exclusivity over an idea, and we do not warrant that AI-generated output is free of third-party rights.

10. Governance

This policy is owned by the Director and reviewed at least every six months, or whenever we adopt a new model, provider or use case. Team members and contractors are required to follow it, and any AI use outside it needs approval.