Search has been the internet’s front door for two decades. It has also been one of the most reliable business models in tech. In 2023, global digital ad spend reached $455 billion, and nearly half of it came from search. The mechanics have not changed much since Google’s rise: crawl, index, rank, serve ten blue links, monetise the click.

The cracks are visible. Search is optimised for generating ad impressions, not solving problems. Users are drowning in results, wading through tabs, and still ending up on content that has been SEO-gamed to death. Generative AI is the first genuine chance to rebuild discovery around context, intent, and resolution rather than pages and keywords.

From keywords to context

Traditional search was built to retrieve something that already exists. GenAI shifts the goal: understand what I actually want, then produce it or orchestrate the steps to get there.

Instead of bouncing between ten tabs, context-aware systems parse intent from history, behaviour, and environment, aggregate relevant data, and return a finished output: an answer, itinerary, product bundle, or claim submission.

Perplexity AI points to the future: conversational, cited responses that skip the detour through blue links. In travel, Kayak’s AI blends past searches, live pricing, weather patterns, and reviews to deliver book-ready itineraries. The search step dissolves into the result.

Curation as the moat

In a world where LLMs can churn infinite copy, the competitive edge is no longer in having content. It is in filtering ruthlessly. Pinterest’s 450 million monthly active users come for a pre-filtered world that feels personal. Spotify’s Discover Weekly built a data flywheel: better recommendations lead to more listening, richer data, and even better recommendations.

The non-obvious insight: LLM horsepower alone is not a moat. Defensibility comes from proprietary datasets, feedback loops, and the taste layer. The “why this, not that” logic that users come to trust. Shopify’s recommendation engine is not just similar products. It is optimised for conversion across millions of storefronts.

Agentic discovery

The next leap is not better answers. It is AI that skips the question entirely. Viv.ai’s model is “tell me your constraints and I will book it.” Hopper watches travel pricing and nudges you to buy when it is optimal. Glean searches your company’s Slack, Drive, and Docs to pull a relevant answer before you type the query.

This is the shift from query to result into context to action. In consumer products, it means an AI that plans and books your trip while you are still mulling dates. In enterprise, it means project briefs assembled from live documents without anyone searching for them.

Horizontal incumbents, vertical openings

Google, Microsoft, and Apple will protect horizontal search at all costs. The openings are in verticals where:

• Data is proprietary or hard to scrape.

• Accuracy and trust decide adoption.

• The user’s intent is inherently actionable.

Hopper in travel, Glean in enterprise, and Kalendar AI in B2B lead generation all prove that vertical depth can beat horizontal reach when context outweighs coverage.

Monetisation follows the decision

Search is shifting from pages to decisions. AI answers compress the funnel into a single surface where intent gets resolved- compare, book, claim, refund. That shift pulls money away from link clicks and into paid presence inside the answer and the action.

You can already see it forming. Google is testing ads in conversational results. Perplexity is experimenting with sponsored follow-ups and perks. Microsoft is building copilot-native ad formats. The spend moves to the moment the user decides.

For brands, SEO alone will not carry you when the interface is an agent. You need to be machine-addressable and executable: clean product feeds, real-time pricing and availability APIs, verified policies, answer-ready content the model can cite.

New metrics will emerge: position inside the answer, action conversion rate, refund friction, policy compliance. In this world, brands will win by being legible to machines and trusted by agent versus persuasive to humans.

Multi-modal by default

The next search interfaces will not make you choose between typing, speaking, or showing. They will accept whatever you offer: a voice prompt, a snapshot, a sketch: and adapt output to the medium you prefer.

In commerce, that means finding a product from a photo. In productivity, snapping a whiteboard and getting a structured project plan. The opportunity is in fusing these modes seamlessly so the user never thinks about switching inputs.

India’s wedge

Vernacular and voice search already outperform typed English queries in conversion. Large verticals like jobs, education, and government services remain fragmented, with poor discovery UX and no integrated agentic layer. Local proprietary datasets (from agricultural market prices to logistics) are untapped inputs for domain-specific retrieval that horizontal players will not prioritise.

What to build

• Agentic vertical search with action completion in high-intent, messy categories.

• Proprietary corpus and curation loops as the real moat, not just the model.

• Trust-first monetisation that improves the decision, not just rents the space.

• Multi-modal interfaces that make text, voice, image, and context fluid.

Bottom line

Search is moving from “find” to “understand and do.” The money will follow the decision point, not the link. The winners will own the context, resolve the intent, and close the loop — in the same surface.

If you are building the next layer of discovery, especially in verticals where context is king, we want the first call.
→ build@boundlessvc.com