Skip to content

AI systems

AI in marketing for small Indian businesses: what works, what is hype, and the rules

The useful version of AI in marketing is less exciting than the pitch. Controlled studies show it makes people faster at drafting and helps less experienced staff the most. The bigger claims, soaring sales and teams replaced, mostly come from the companies selling the tools. Meanwhile India has put rules in place on AI-made images and video, and the old rules on misleading ads apply to everything AI writes for you. This guide covers both.

What does AI in marketing mean for a small business right now?

For most small businesses in India, it means five things:

  • drafting text: social captions, emails, ad copy, product descriptions, first drafts of blog posts;
  • translating and adapting that text into the languages your customers use;
  • letting Google Ads and Meta’s ad tools choose audiences, placements and even the words in your ads;
  • answering routine customer questions on WhatsApp or your website;
  • making or editing images and short videos.

Each of these works to a different degree, and each carries a different rule. The rest of this guide takes them in turn.

Which uses of AI have real evidence behind them?

Three studies stand out because they measured real professionals against a comparison group: two controlled experiments on realistic tasks and one study of a real workplace.

In a study published in Science in 2023, economists Shakked Noy and Whitney Zhang gave 453 college-educated professionals short writing tasks of the kind their jobs involve; their working paper lists marketers among them. Those given a commercial chatbot assistant took 40 per cent less time, and the quality of their work, judged by other professionals, rose 18 per cent. The authors are careful about what that covers: the tasks were “relatively short, self-contained, and lack a dimension of context-specific knowledge, which may inflate our estimates”.

Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied 5,179 customer support agents (5,172 in the published version) at one company as an AI assistant was rolled out. In their 2023 working paper, agents resolved 14 per cent more issues an hour on average, and new or less skilled agents 34 per cent more. The most experienced agents gained very little. The published version in 2025 puts the average gain at 15 per cent and finds that the top performers saw small gains in speed and small declines in quality.

A Harvard Business School study of 758 Boston Consulting Group consultants found both sides of the coin. On tasks the AI handled well, consultants using it completed 12.2 per cent more tasks, 25.1 per cent faster. On a task chosen to sit outside what the AI could do, the full paper reports they were 19 percentage points less likely to get the right answer than consultants working without it.

Put together: AI is good at producing a competent first draft quickly, it helps people who are newer to the work the most, and it can lead people astray on work that needs judgement or context it does not have. That is exactly the shape of a small business’s marketing. The drafting is real help. The deciding is still yours.

Where do the claims outrun the evidence?

Watch for four kinds of claim:

  • Platform numbers. Google says advertisers who turn on its AI Max feature in Search typically see 14 per cent more conversions at a similar cost. The footnote reads “Google internal data, 2025” and limits the figure to non-retail advertisers whose campaigns already get most conversions from exact or phrase match keywords. An independent analysis of more than 250 campaigns by the agency SMEC (Smarter Ecommerce), reported by Search Engine Journal, found a median 13 per cent rise in conversion value but a median 16 per cent rise in cost per acquisition. Return on ad spend was unchanged at the median, but results ranged from 42 per cent above each campaign’s baseline to 35 per cent below it. It works for some accounts and not others.
  • Founder quotes. Announcement pages for AI products often include a business owner saying conversions “skyrocketed”. These are chosen by the company and come with no method.
  • “AI will replace your marketing team.” A large study in Denmark that linked surveys to official records found no measurable effect on earnings or hours worked two years after chatbots arrived, ruling out effects larger than 2 per cent. That measures jobs, not marketing results, but it is a useful check on the bigger predictions.
  • Adoption statistics. A Dun & Bradstreet survey of 210 Indian MSME leaders in August 2026 found 24 per cent not using AI, 35 per cent experimenting informally, 37 per cent automating some tasks and only 5 per cent describing AI as core to their operations. That is the most useful picture we found, and it is a small sample of business leaders, not all Indian small businesses.

Will Google penalise content written with AI?

Not for being written with AI. Google’s guidance on AI-generated content says its focus is “on the quality of content, rather than how content is produced”. What it does penalise is scaled content abuse: “many pages … generated for the primary purpose of manipulating search rankings and not helping users”, which applies “no matter how it’s created”. Its examples include using generative AI to produce many pages without adding value.

Google’s current advice for anyone using AI to write is short: “It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing.” It explains why in one line: “generative models don’t retrieve facts, but predict a likely sequence of words.” A hundred AI-written blog posts about your city and your service will not help you. Ten posts that answer your customers’ real questions, checked by someone who knows the answers, might.

What do Google Ads and Meta’s AI features do on your behalf?

More than many advertisers realise. In Google’s Performance Max campaigns, the system can write headlines and descriptions from your website and send people to a different page from the one you chose. Google’s help page is clear about who answers for the result: “you remain responsible for reviewing and ensuring compliance and accuracy of landing page content, and all dynamically generated assets.”

AI Max, added to Search campaigns in 2025, works the same way: it widens the searches your ads match, rewrites ad text and can choose the landing page. Google’s help page says new Search campaigns have it switched on by default. You can exclude words and pages, and turning off text customisation also turns off landing page expansion. If you run a clinic, a legal practice or anything where a wrong claim matters, read what is switched on before you launch.

Meta labels ads that were made or significantly edited with its own AI tools, and in June 2026 said it will also begin automatically detecting ads made or edited with other companies’ AI tools. It notes the experience may vary by region.

Can an AI assistant answer your WhatsApp customers?

For routine questions, yes, and Meta now offers one itself. In May 2026 it began rolling out Business AI on WhatsApp to eligible small businesses in India, in Indian languages, to answer questions, capture leads and help with bookings. Meta says the owner can step in at any time, adjust it or turn it off.

The question to settle first is who is responsible when it gets something wrong. The best-known answer comes from Canada. In 2024, a tribunal ruled against Air Canada after its website chatbot gave a customer wrong information about a fare. The airline argued it could not be held liable for information given by its agents, including a chatbot. The tribunal’s decision replied: “It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.” That is a Canadian decision, not Indian law, and we know of no Indian ruling on a chatbot yet. The safe assumption is that a customer, a consumer commission or a regulator will treat your chatbot’s answers as your business’s statements.

So keep an assistant to the questions your own approved information answers, and have it hand everything else to a person. Our guide to AI receptionists goes through the tests to run before launch.

Do you have to label AI-made ads in India?

It depends on what the AI made, and who is publishing it.

The IT Rules: realistic images, audio and video

Amendments to India’s IT Rules, notified on 10 February 2026 and in force from 20 February 2026, cover “synthetically generated information”: audio, images or video that look real and depict a person or event in a way that could pass for the real thing. Platforms that offer tools to create such content must label it, and large social media platforms must ask users to declare it. Routine editing, colour correction, translation and similar uses are excluded. MeitY’s FAQ says plainly that “pure text / written outputs, by themselves, are not SGI”. Platforms must also block AI content that deceptively depicts real people or events.

These duties fall on platforms, not directly on advertisers. In practice you will meet them as a user: a platform may ask you to declare that an image or video is AI-made, and will label it.

ASCI: AI in advertising

The Advertising Standards Council of India, the advertising industry’s self-regulatory body, published guidelines on labelling AI-made content in ads in September 2026. They come into effect three months after publication, so they are not yet in effect as of this guide. Under them:

  • AI-written or AI-improved ad copy needs no label.
  • Where AI-made content could change what a customer believes about the product, and leaving it unlabelled would mislead, a label is needed. ASCI’s examples: synthetic influencers and ambassadors, fabricated events or settings that affect how the product appears to perform, demonstrations of products that do not yet exist, sound effects central to the product, and consented replicas of a real person’s face or voice. Paid AI product suggestions must be labelled as sponsored.
  • Using AI to fabricate an endorsement or testimonial breaks the ASCI code “even if an AI label is used”.
  • Also banned, label or not: AI images that exaggerate results (ASCI’s example is a weight-loss before-and-after with an AI-made after picture), depicting places or facilities that do not exist, and using anyone’s likeness without consent.
  • A label does not rescue an ad that is otherwise misleading.

ASCI’s influencer guidelines already require a virtual influencer to disclose upfront that people are not dealing with a real human. Our guide to influencer marketing rules covers labels for creators and virtual influencers in detail.

What can you not do with AI at all?

The rules that matter most here are not AI rules. They are consumer protection rules that apply to every ad, however it was made.

  • Make claims you cannot prove. The Central Consumer Protection Authority’s 2022 guidelines cover “all advertisements regardless of form, format or medium” and require factual claims to be “capable of substantiation”. They also bar exaggerating a product’s “capability or performance”. Calling your product “AI-powered” when it is not would fall foul of this.
  • Invent reviews or testimonials. Endorsements must reflect a genuine, reasonably current opinion based on adequate information about, or experience with, the product. An AI-written testimonial is neither.
  • Manufacture urgency. The Dark Patterns Guidelines, 2023 bind advertisers and sellers as well as platforms, and list false urgency (“Only 2 rooms left!”) among the banned practices. The CCPA penalised two companies for dark patterns in June 2026.

Per the government’s 2022 press release, penalties for misleading advertisements can reach Rs 10 lakh, and Rs 50 lakh for repeat offences. AI makes it very easy to produce a confident claim. It does not produce the evidence for it.

What happens when customer data goes into an AI tool?

Under the Digital Personal Data Protection Act, 2023, once its main duties commence on 13 May 2027, your business will remain responsible for personal data processed on its behalf, and may only use a processor under a valid contract. Uploading a customer list to an AI tool to write personalised messages is processing. Two practical points now: read what the tool’s terms say about using your data, and do not use AI to target advertising at children, which the Act will prohibit, with narrow prescribed exceptions.

Until then, rules already apply. The IT Act’s 2011 rules on sensitive personal data cover health conditions and medical records, among other things. Clinics in particular should not upload patient details to a general AI tool. Our DPDP guide for small businesses sets out the duties that start in May 2027.

How do you test an AI tool before you trust it?

  1. Pick one repeated task, such as weekly social captions or answers to the ten most common enquiries.
  2. Write down what good looks like before you start, so you are not grading the AI against its own output.
  3. Run it alongside your current process for two to four weeks. Keep the human version going.
  4. Check every factual claim it makes. Count the ones that were wrong.
  5. For ads, keep a comparison: an ad set or campaign without the AI feature, at a similar budget.
  6. Decide on the numbers you recorded, not on how impressive the output looked.
UseOur reading of the evidenceThe rule to remember
Drafting copy with a person editingWorks: faster, often betterYou own every claim it makes
Translation and summariesWorks for routine materialRoutine translation is outside the IT Rules’ AI label
Ad platforms’ AI featuresMixed: good for some accounts, worse for othersYou remain responsible for generated ads
WhatsApp assistant for routine questionsUseful for narrow, approved answersNeeds a person to hand over to
Hundreds of AI pages for SEOHype and a spam-policy riskScaled content abuse
AI reviews or testimonialsNeverMisleading under CCPA; breaks the ASCI code
Synthetic people in adsAllowed, with careLabel under ASCI once in effect

QUESTIONS

Common questions

Do I have to label my social media posts as AI-generated in India?

Text does not need a label: MeitY says pure text outputs are not synthetically generated information under the IT Rules. Realistic AI images, audio or video may need to be declared on large platforms, which will label them. ASCI’s guidelines, once in effect, require labels where AI-made content such as synthetic people, fabricated settings or non-existent products could mislead customers, but not on AI-written copy.

Can I use AI to write my ad copy and blog posts?

Yes. Google judges content on quality, not on how it was made, and ASCI needs no label for AI-written copy. You remain responsible for every claim, so check the facts before publishing.

Can I use AI to write reviews or testimonials to get started?

No. Under the CCPA’s 2022 guidelines, endorsements must reflect a genuine opinion based on adequate information about, or experience with, the product, and ASCI says AI-fabricated testimonials break its code even with a label.

Will Google penalise my website for AI-written content?

Not for using AI. Google penalises pages made at scale mainly to manipulate rankings, however they were produced, and advises fact-checking all AI-generated content before publishing.

Is it safe to put my customer list into an AI tool?

Check the tool’s terms on how your data is used first. Once the DPDP Act’s main duties commence on 13 May 2027, your business stays responsible for data a processor handles on its behalf, under a valid contract. Until then, the IT Act’s 2011 rules on sensitive personal data, which cover medical records, already apply.

If an AI chatbot gives a customer wrong information, is my business responsible?

Assume so. In a 2024 Canadian case, a tribunal held Air Canada responsible for its chatbot’s wrong answer. That is not Indian law, and we know of no Indian ruling on a chatbot yet, but the safe assumption is that your chatbot’s answers will be treated as your business’s statements.

REFERENCES

Sources and further reading

Each link was opened and checked on .

← All field notes