Theme three of three from the 17th Digital Leadership Summit, hosted by Social Beat with Google India and Cheil at the Taj MG Road, Bengaluru, on 31 July 2026.
Most debates about AI in marketing end up in one of two camps. One side says the machines will soon run the whole function. The other says the output is average and the hype will pass. The AI conversations at this year's Digital Leadership Summit were more useful than both, because nobody on stage was predicting. Leaders from NoBroker, Oolka, First Club, Orange Health Labs and UNext were describing where AI actually sits inside a live business, and where they have chosen to stop it.
Across five very different companies, the line landed in the same place. AI now does the middle of the work. A human still makes the call at both ends: the brief that goes in, and the decision that goes out.
Rahul Datta of NoBroker described one of the more advanced conversational AI setups we have seen in an Indian consumer business. An AI voice agent, trained on the company's own best agents, speaks to customers and adjusts its tone as the conversation shifts. A second system listens to live calls and flags what is going well or badly. A third feeds context to human agents while they are still on the phone.
Then he explained where all of it stops. Nobody at NoBroker allows AI to launch a campaign or change one. The system only knows what it has been given. It cannot weigh brand language, connotation, or how a line will read to someone it was never written for. A person signs off, every time.
"You need a neck to catch."
That line from the NoBroker session is the clearest summary of the problem we have heard. The constraint on AI in digital marketing is not capability anymore. It is accountability. Someone has to answer for the decision, so someone has to make it. This is what human in the loop actually means in practice. It is not a compliance checkbox. It is a named person who owns the outcome.
Worth noting, too, how he framed the bigger picture. NoBroker was built to reduce the information gap between brokers and buyers. In his view, AI has not inserted a new middleman. It has extended the promise the brand was founded on. That is a sharper question than most AI commentary asks: before you ask what AI does to your workflow, ask what it does to the thing your brand already stands for.
Sandesh Gupta of Oolka split the growth function into three parts: research and analytics, content to feed campaigns, and execution. His answer was blunt. AI is doing the first two well. The third has broken for him. He has spent months pushing automated execution across Meta and Google, and his honest read is that it is not ready. A human still has to make the tactical calls that produce the result.
Notice what this does to the standard pitch for AI in performance marketing. The usual deck says AI will handle execution and free humans up for strategy. The practitioner running it daily says the opposite: AI handles the inputs to strategy, and humans still have to execute. On current evidence, we side with the practitioner.
One story from that session stayed with the room. Oolka builds for customers in smaller towns, where financial literacy is thin and the lending chain is tangled. Sandesh found a chat in which a customer near Bhubaneswar was planning the finances for her daughter's wedding with their bot, apparently unaware it was a bot, and warmly inviting it to the wedding. Whatever you conclude about AI in marketing, that is a real signal of how quickly conversational AI is being trusted outside metro India.
The most instructive moment of the day came from Lucky Saini of First Club, who took two opposite AI positions inside the same ten minutes. First: no AI creative. When you sell trust in fresh food, machine-made content is the last thing a customer should see. Then, minutes later: in paid media, First Club is automating nearly everything, with humans holding only two decisions, deployment and policy.
Both positions are right, and together they are the answer most teams are looking for. The question is not whether to use AI. It is which layer to give it. Smita Murarka of Orange Health Labs drew the same boundary from diagnostics: AI gets her team seventy to eighty percent of the way on research, thinking and direction, and never produces the final work, because a diagnostics brand sells trust and the story has to be real.
Put simply: automate the layer where the output is a number. Keep humans on the layer where the output is a promise. Most Indian marketing teams currently have it backwards. They generate creative with AI because it is visible and easy to demo, and run media by hand because that is where the team's habits live. Flip it.
The observation with the shortest fuse came from Kiran Kumar R of UNext, almost as an aside. The SEO fundamentals, he said, have not changed. What has changed is how people find you. Users are now arriving through AI models, and UNext can see and track that traffic. He put around seventy percent of the company's demand in Tier 2 and Tier 3 towns, which means this is not an early-adopter story from the metros. It is the main channel shifting under a working funnel.
Set that beside what Kapil Thukral of Tally said about zero-click results turning search into a winner-takes-all market, and the picture gets uncomfortable. Being cited in the AI answer is starting to matter more than ranking on the page. Ranking earns you a slot on a results page. A citation puts you inside the answer itself, which is the only thing many users will read.
This is the ground that generative engine optimization covers, along with answer engine optimization, or AEO. Call it GEO optimization, AI search optimization, or AI Overview SEO; the work is the same: structure your content so that AI engines can find it, quote it, and credit it when they answer a buying question in your category. We built GEO Pulse at Social Beat to measure exactly this, and the honest state of the market is that most brands cannot yet say whether they appear in an AI answer for their own category. That is the gap, and right now it is cheap to close because so few are trying.
Three moves, in the order we would make them.
One. Write down the decision rights. List which calls in your marketing operation a machine may make and which need a named human. Most teams have never put this on paper, which is why the same argument keeps repeating in every review.
Two. Apply the layer rule. Automate wherever the output is a measurable number. Keep people wherever the output is a promise a customer has to believe. If your AI budget is going into creative while your media runs manually, you are automating in the wrong direction.
Three. Check your AI visibility this week. Ask the major AI engines a buying question in your category and see whether your brand is cited. If you have not checked, you do not know, and your competitors' content may already be the answer.
Generative engine optimization is the practice of making your content visible and citable inside AI-generated answers, on platforms like ChatGPT, Gemini, Perplexity and Google's AI Overviews. SEO earns you a rank on a results page. GEO earns you a mention inside the answer itself, which is increasingly where the click would have gone.
Answer engine optimization means structuring content so that answer engines, from AI assistants to featured snippets, can lift a direct, accurate response from your page. In practice, AEO, GEO and SEO overlap heavily: clear structure, direct answers and strong sourcing help you in all three. The terms differ mainly in which surface they target.
Not on the evidence from operators actually running it. AI now does the middle of the work: research, analysis, first drafts, signal sorting. Humans still own both ends, the brief and the accountable decision. As the NoBroker session put it, you need a neck to catch. The marketers at risk are the ones whose entire job was the middle.
Based on what five companies described on stage: use it for research, analytics and content development, where it reliably gets teams most of the way. Be careful with fully automated execution at scale on Meta and Google, which practitioners like Oolka's Sandesh Gupta say is not dependable yet. And keep final creative and brand judgment human wherever the product is trust.
Not today. Even the most automation-forward team at the summit, First Club, holds two decisions back from the machines: campaign deployment and policy. NoBroker will not let AI launch or edit a campaign at all. The blocker is accountability. Until someone can be answerable for a machine's call, a person has to make it.
Write for extraction. Use clear question-based headers, give a direct answer in the first two sentences under each, name entities precisely, and back claims with sources an engine can verify. An FAQ section like this one is itself the tactic: it gives AI engines a clean question and a quotable answer in one place.
You need AI visibility monitoring: regularly asking the major engines the buying questions in your category and recording whether, and how, your brand appears. Doing this manually works as a first check but does not scale, which is why we built GEO Pulse to track citations across engines over time. Most brands that run this check for the first time are surprised by who is being cited in their place.
A closing question for your next team meeting: which decision in your marketing is a machine already making that nobody formally agreed it should make? Write the answer down. That list is where your AI policy starts.