As technology co-founder of Marketing Kernal, Shrivastava is working on one of the harder engineering problems in marketing today: measuring what AI chatbots say about a business when nobody is watching.
Pratyush Shrivastava can describe the moment he realised his first startup was in trouble, and it was not a crisis. It was a quiet afternoon, looking at a product that worked exactly as designed, with almost nobody using it.
“The code was fine,” he says. “I was proud of it. That was the problem. We had built something good that nobody had asked for yet, and you cannot debug that.”
The company closed. Shrivastava, who is from Jaipur, did what many engineers in his position do. He went back to building, but he says he went back with a different rulebook.
Today he is the technology co-founder of Marketing Kernal, a company formally established in 2026 that works on AI visibility, the emerging discipline of understanding how businesses are represented inside answers generated by tools such as ChatGPT, Gemini and Perplexity. He leads its architecture, infrastructure and software development. His co-founder, Vaibhav Palhade, an engineer who spent years as a screenwriter for Hindi and Marathi film and television before turning to fiction and then marketing, leads marketing and product.
The rulebook
The first rule, Shrivastava says, is to be slow at the beginning. Before the company was set up, the two founders spent roughly two to three years researching how AI systems choose which businesses to mention, and where existing marketing and sales tools were blind to that process.
“In my first company we wrote code to find out if we were right,” he says. “This time we tried to find out if we were right before writing much code. It is less exciting. It is much cheaper.”
Palhade, who counts three or four failed ventures of his own, shared the instinct. “Neither of us needed convincing that speed can kill a company,” Shrivastava says. “We had both watched it happen.”
A difficult thing to measure
The problem the company has chosen is technically awkward. Search rankings, for all their complexity, are relatively stable and can be observed. AI answers are not. Ask the same question twice, with slightly different wording, on a different day or a different model, and the businesses named can change.
“People want a single number,” Shrivastava says. “Am I visible or not? But one check tells you almost nothing. You need repeated sampling, consistent questions, a record of which sources were cited, and a way to compare all of that over time without fooling yourself. It is closer to running an experiment than running a report.”
The scale of the audience explains why businesses are starting to care. In October 2025, OpenAI chief executive Sam Altman said ChatGPT had passed 800 million weekly users. Research published by the Pew Research Center in July 2025 found that when Google showed an AI-generated summary, users clicked on links inside that summary in only about 1 per cent of visits. More and more of the decision happens inside the answer itself.
What he is building
Marketing Kernal is developing a family of software products under the Kernal name. The first to reach the market will be Outreach Kernal, a B2B prospecting tool focused on finding and researching the right buyers with verified data. Shrivastava expects a beta in November, opened first to a limited group of users. Shrivastava declines to name the rest, saying only that they cover visibility in AI answers and the sales conversation itself, “the two ends of the same journey.”
Shrivastava’s interest is less in the product names than in what sits underneath them. “Sales and marketing software has rewarded volume for a long time,” he says. “More contacts, more emails, more dashboards. I think the next generation of tools will be judged on a duller question. Is the data right?”
He is equally blunt about privacy. Some of the markets the company is targeting, including the United Kingdom and the European Union, have strict rules on personal data and artificial intelligence, among them the GDPR and the EU AI Act.
“If you get the data model wrong at the start, you spend years apologising for it,” he says. “I have no interest in apologising for years.”
Asked what actually makes a business more likely to be named by an AI assistant, he lists factors that sound mundane. “Consistency is underrated,” he says. “If your company is described one way on your website, another way on a directory and a third way in an old article, a model has to guess which version is true. Machines are not good at giving you the benefit of the doubt.”
Structured data, the behind-the-scenes labelling that tells software what a page is about, matters too, as do references from credible, independent publications. None of it is new to search specialists. What is new, he says, is that the consequences are harder to see. “With search, you could watch your ranking fall. With AI, you are simply not mentioned, and nothing in your analytics tells you.”
Building global from Jaipur
Jaipur has spent the last decade acquiring a reputation as a startup city, producing companies in payments, travel and consumer technology, some of which later moved to larger hubs. Shrivastava sees no reason to leave.
“A customer in London does not ask for your postcode,” he says. “They ask whether the system works at three in the morning their time.”
The working relationship with Palhade, he says, is built on an argument they have most days. “He describes a problem from the buyer’s side. I describe it from the database’s side. The product usually ends up somewhere in between.”
Second chances
The company is young, and Shrivastava is careful not to talk about it as though it has already arrived. Its products are at different stages, and the field it works in is still settling its own vocabulary.
What he will say is that the failure changed him more than any success could have. “The first time, I thought the hard part was the technology,” he says. “The second time, I know the hard part is being honest about whether anyone needs it.”
