Generative AI Content Without Losing Brand Voice or Creating Legal Risk

You can use generative AI for brand content safely if three things are in place: a written brand voice system the model and editors both follow, a risk-tiered human review before anything is published, and clear rules on synthetic people, voices and claims. In India the rules tightened twice in 2026. Amended IT Rules made platforms label realistic synthetic media from 20 February, and the Advertising Standards Council of India (ASCI) published AI labelling guidelines for ads on 29 September. This guide turns both into a working system.

What are the main risks of using generative AI for brand content?

The risks fall into four groups: voice drift, accuracy, rights and likeness, and disclosure. None is a reason to avoid AI; each is a reason to put a system around it.

• Voice drift. Models default to a smooth, generic register. Over months, a distinctive brand starts sounding like everyone else.
• Accuracy. Models confidently produce invented statistics, misattributed quotes, outdated prices and features you do not offer.
• Rights and likeness. Output may resemble copyrighted work, a competitor's trade dress or a real person's face or voice.
• Disclosure and regulation. Realistic synthetic media now carries labelling duties in India, and misleading advertising rules apply however content was made.

What do India's 2026 IT Rules say about AI-generated content?

Since 20 February 2026, platforms must prominently label realistic synthetic audio and visual content, embed provenance metadata, and remove unlawful content much faster.

The Ministry of Electronics and Information Technology notified amendments to the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules on 10 February 2026, effective 20 February, as Hogan Lovells summarises.

What is synthetically generated information?

Synthetically generated information (SGI) is audio, visual or audio-visual content created or altered artificially or algorithmically so that it appears real and could be mistaken for a real person or event. Khaitan and Co's analysis on Mondaq notes the definition is technology-neutral and excludes text-only content, routine or good-faith editing, and tools that improve accessibility, clarity, quality or translation, provided they do not manipulate a material part of the content.

What must platforms do?

• Visual SGI must carry labels that are prominent and easily noticeable; audio SGI needs a prominently prefixed audio disclosure. The draft rule requiring labels to cover 10 percent of the frame was replaced by this qualitative standard.
• Platforms must embed permanent metadata or provenance markers, including a unique identifier, and must not allow them to be removed.
• Significant social media intermediaries, those with more than 50 lakh registered users in India, must ask users to declare whether uploads are synthetic and verify those declarations.
• Compliance with lawful takedown directions dropped from 36 hours to 3 hours, per Bar and Bench, and complaints about non-consensual nudity or morphed imagery must be acted on within 2 hours.

These duties fall mainly on platforms, but brands publish through them, so expect declaration prompts on upload. As King Stubb and Kasiva note, labelling is not by itself a defence for otherwise unlawful content, and agency contracts should say who is responsible for labelling and records.

What do ASCI's AI advertising guidelines require?

ASCI's guidelines require an AI label where synthetic content could mislead a consumer's decision, exempt routine edits and ad copy, and prohibit fake endorsements outright.

ASCI released a draft in May 2026 and issued the final Guidelines for Responsible Labelling of Synthetically Generated Content in Advertising on 29 September 2026. They take effect three months from publication, which places the start around the end of December 2026. The final version sorts content into three groups:

As ASCI's leadership put it at launch, the mere use of an AI label may not make an otherwise misleading advertisement acceptable.

Who owns AI-generated content, and can it infringe copyright?

Ownership of purely machine-generated output is unsettled in India, and policy on AI training is still being drafted, so protect yourself through human authorship and careful inputs.

On 8 December 2025 a DPIIT committee released a working paper proposing a hybrid model: blanket access to copyrighted works for AI training, with royalties payable on commercialisation, as Storyboard18 reported. It is a consultation draft, not law. Practical guidance meanwhile:

• Your position is strongest where people make substantial creative choices: concept, selection, arrangement and editing. Keep records of that work.
• Never prompt for work "in the style of" a named living artist, photographer or competitor brand, and never feed a competitor's assets in as reference.
• Read each tool's terms for commercial use rights and any indemnity.
• Reverse image search key visuals, and clear any AI-assisted name or logo using our guide to brand naming and trademarks in India.

How do you keep your brand voice when using AI?

Write the voice down as rules a model can follow and an editor can check. A useful brand voice system has five parts:

1. Voice principles with contrasts. Three to five traits written as "this, not that": "plain-spoken, not casual; confident, not boastful".
2. Vocabulary lists. Words you use, words you never use, product names and how they are written, and banned clichés.
3. Annotated examples. Ten to twenty pieces of your best real content, with notes on why they work. Models imitate examples far better than adjectives.
4. Channel rules. Length, structure and tone per channel: Instagram captions differ from LinkedIn posts and WhatsApp templates.
5. Claims register. Approved facts, figures, certifications and promises, with sources. Anything outside it needs sign-off.

Load this into a shared, reusable prompt or project so every writer starts from the same base. For Indian brands working in Hindi, Tamil or Malayalam, add a short voice note and examples per language, reviewed by a native speaker. If positioning itself is fuzzy, start with our brand positioning framework.

What review workflow should brands use for AI content?

Match review effort to risk, with three tiers and three rules that never bend.

• Low risk (internal drafts, alt text, replies from approved snippets): author self-check against the voice checklist.
• Medium risk (blogs, captions, email, AI-edited images): editor review for voice, with every claim checked against the claims register.
• High risk (ads with performance or health claims, synthetic people or voices, comparative ads, content for children): editor plus legal sign-off, consent records for any likeness, and an ASCI-compliant AI label where required.

The non-negotiables: a named human is accountable for every published piece; every statistic, quote and price is traced to a source; and no realistic depiction of an identifiable real person, including employees and customers, without written consent for that use.

Common mistakes we see

• An AI-generated "customer" in a testimonial-style ad. Under ASCI's guidelines that is a fabricated endorsement.
• Reusing a founder's cloned voice for a new campaign without fresh written consent.
• Agency contracts silent on who labels, who keeps records and who fixes a takedown.

What should an AI content policy include?

• Approved tools, and data that may never be entered (customer personal data, unreleased financials, client confidential material). See our DPDP compliance guide for marketers.
• The brand voice system and claims register, and where they live.
• Review tiers and who signs off at each.
• Rules for synthetic people, voices and likeness, with consent forms.
• Labelling rules per platform and per ASCI category.
• Responsibilities in agency and freelancer contracts.
• An incident process: who removes content, who responds and how fast.

How we approach AI content at Old Port

We treat AI as a production tool inside a creative direction process, not a replacement for one. We document voice and claims first, build the shared prompt library second, and only then scale output, with the review tier decided when the brief is written rather than after the draft exists. Our creative direction and content production and brand strategy teams work this way in India, and our Dubai studio offers the same through its content production service in the UAE. For the wider picture, read how Indian businesses are using AI in marketing in 2026.

This article is general information, not legal advice.