The Invisible Buyer

Agents Buy from Agents: Why Your GTM Motion Needs to Change

Something quietly crossed a threshold in 2025. A ChatGPT agent doing research on your behalf does not visit five websites the way you would. It visits five thousand. Cloudflare CEO Matthew Prince put a number on it at SXSW 2026: for every task a human performs with five web visits, an AI agent performs the same task with roughly 5,000 page requests. That is not a metaphor. That is measurable load, and every company serving web traffic is now dealing with it.

Gartner forecasts that by 2028, AI agents will outnumber human sellers by 10x. Forrester found that 74% of B2B organisations are already adopting AI agents in purchasing workflows, and 61% of purchase influencers say their organisation has or will use a private AI engine for buying decisions. G2's survey of 1,169 enterprise decision-makers confirmed that 29% now start vendor research via LLMs more often than Google, a number that was essentially zero three years ago.

The critical shift: agents do not respond to persuasive copy. They parse structured data. An agent evaluating your product does not read your "Why Choose Us" page. It queries your OpenAPI spec, checks schema markup, counts your G2 reviews, and synthesises a recommendation from structured signals, not narrative ones. Amazon's Rufus AI already makes users 60% more likely to complete a purchase. Forter recorded an 18,510% day-over-day surge in agentic traffic within hours of ChatGPT Agent's launch.

The non-obvious implication for Indian B2B founders: Your content strategy was built to persuade humans. Your GTM infrastructure needs rebuilding to be discoverable by machines. Add schema markup, publish an OpenAPI spec, implement llms.txt, and ensure your G2 presence is active because Reddit and G2 are the two sources AI tools cite most when recommending software.

70% of Your Buyer's Journey Is Invisible, and Your Analytics Are Lying to You

The buyer is doing their homework. They are just not doing it anywhere you can see. The 6sense 2025 Buyer Experience Report, drawn from 4,000+ respondents, found that 61% of the B2B buying journey completes before any vendor contact, and only 3% of website visitors self-identify through form fills. The other 97% are researching, comparing, and forming opinions about your brand in places your CRM will never record.

Where are they? Reddit threads that appear in Google AI Overviews. Private Slack communities where someone asks "has anyone used X?" and three people respond. LinkedIn DMs between a VP and an analyst she trusts. WhatsApp groups of sector peers. ChatGPT conversations that leave no cookie. Cognism's data shows 77.5% of buyers share links through dark social channels, traffic that registers as "direct" in your analytics and tells you nothing about where the conversation started.

Reddit has become the structural backbone of B2B dark funnel. Reddit pages are cited 5.3 million times by Google AI Overviews, 5.5 million times by Perplexity, and 4 million times by ChatGPT, making it the number one most-cited source across all major AI platforms, three times more than Wikipedia. 75% of Reddit's B2B user base say they will use the platform to inform future software purchases. Cost per lead on Reddit for B2B SaaS runs $45–85 versus $120–200 on LinkedIn.

Chris Walker at Refine Labs spent years proving this with data: social media influence on pipeline is under-reported by attribution software by approximately 70%. The people who became your best customers this quarter were shaped by something they read in a Slack channel three months ago, and your Salesforce dashboard credits the Google Ad they clicked at the end.

For Indian B2B founders: Add a free-text "How did you first hear about us?" field to every demo form. Run it for 60 days. You will find 3–5 channels your analytics have never shown you. Do not sponsor any dark channel before you listen in it first. In India specifically, WhatsApp peer groups are the dominant dark channel, senior buyers compare tools in ways that leave zero digital trace.

Brand Is Now a Measurable Moat, Not a Vanity Metric Finance Ignores

The argument that brand is unquantifiable has always been a convenient excuse for underfunding it. In 2025, that excuse ran out. Les Binet's research established Share of Search, your brand's search volume as a proportion of total category search, as a leading indicator of market share with an 83% correlation coefficient and a 6–12 month lead time. Brand search goes up today, revenue follows in six months. Track it. Report it. Fund it accordingly.

Forrester is now explicit: "41% of B2B buyers begin their purchase journey with a single preferred vendor already in mind. Over 90% have a shortlist. Where do those preferences come from? Brand." Forrester calls the separation of brand and demand "a falsehood." Cognism CMO Alice de Courcy put the budget split clearly: 60–70% into demand creation (brand, awareness) and 30–40% into demand capture (intent-based, retargeting). Most Indian B2B companies run this ratio in reverse.

The Ehrenberg-Bass 95:5 rule explains why this matters. At any given moment, 95% of your potential market is not actively buying. Performance channels reach only the 5% who are. Brand channels build preference with the 95% so that when they eventually enter the market, you are already on the shortlist. Chargebee's G2 intent data strategy generated 280+ top-funnel leads in one year: the "European Leader" badge converted buyers who had already formed their preference on a platform Chargebee did not control.

The Founder Is Now the Most Valuable Content Asset the Company Has

When AI generates most web content and discovery moves to dark channels, the signal that cuts through is a recognisable human voice with a point of view. Not a brand voice. A human voice. Zerodha reached 16 million users with zero advertising spend. Nithin Kamath built India's largest retail brokerage by publishing genuine, sometimes uncomfortable opinions about investing, opinions that people trusted because Nithin was visibly accountable for them. Zerodha Varsity, free and comprehensive, generated more inbound leads than any campaign the company has ever run.

Sridhar Vembu at Zoho built a "barefoot billionaire" persona that earned a Padma Shri and helped position Zoho as the principled alternative to Silicon Valley SaaS. Girish Mathrubootham at Freshworks turned a competitor's condescension into a brand narrative that carried them to NASDAQ. Krish Subramanian at Chargebee addressed the international credibility gap by deeply embedding in SaaSBOOMi and hosting the "Second Acts" podcast, building authority in a peer community first, then projecting it outward.

The common thread is not reach. It is accountability. Each of these founders said things that were specific, sometimes contrarian, and always personally accountable. That accountability is what AI cannot replicate. A model can produce content that sounds like your founder. It cannot produce content that your founder is willing to stand behind in a public forum, respond to pushback on, or update when the market proves them wrong.

Token Costs Are Rewriting Agency Economics, and Nobody Has Told the Client

WPP fell out of the FTSE 100 in December 2025 after nearly 30 years of membership. Its share price dropped roughly two-thirds over the year. Revenue fell 3.6%. Ogilvy eliminated 700 employees. Omnicom cut 3,000. IPG laid off 3,200 in nine months. Meanwhile, Publicis hit a record 18.2% operating margin with 73% of its business model running through its AI infrastructure CoreAI. The holding company era is not ending slowly. It is collapsing asymmetrically.

Digiday's March 2026 investigation found at least seven distinct agency pricing models with no industry consensus. Pencil uses generation credits. Horizon Media charges a nominal cost-recovery fee. Merge passes costs through metered billing. RPA absorbs AI costs entirely. The variation is not because the industry is evolving. Agencies do not want to have this conversation with clients, because transparency reveals exactly how much labour has been replaced.

The underlying economics are stark. A deliverable that previously required 100 hours of human labour now takes two hours of oversight and $200 in API tokens. AI-native agencies like Supernatural AI (30 people, winning national campaigns) operate at 94–96% margins per asset. Traditional agency operating margins sit at 13–18%. YC's Spring 2026 Request for Startups explicitly calls for agencies that operate with software margins. That call is being answered.

For founders using agencies: Ask three questions in your next contract negotiation: What percentage of deliverables involve AI generation? What is the actual pricing model? Can you show the compute cost of this campaign? Move toward outcome-based pricing. Pay for pipeline, not for hours billed against a tool subscription you are funding.

SEO Is Not Dead, But the Easy 60% Is Gone and Won't Come Back

58% of Google searches now end without any click to an external website. Google AI Overviews appear on 15–35% of all US desktop searches with 2 billion monthly users globally. When an AI Overview appears, position 1 click-through rate drops by 34.5%. HubSpot's organic traffic fell from 13.5 million monthly visitors to under 7 million: the clearest evidence that the old SEO playbook has a structural problem.

But Grow and Convert's observation across their client base complicates the narrative: rankings went up, traffic went down, and conversions stayed flat or grew, what they call "the decoupling." AI Overviews cannibalise top-of-funnel informational queries but leave high-intent commercial queries relatively intact. Pricing pages, comparison content, and proprietary benchmark reports showed growth over the same period that informational content collapsed.

Being cited in AI Overviews is now the equivalent of page one. Seer Interactive's study of 25.1M impressions found that sites cited in AI Overviews receive 35% more organic clicks and 91% more paid clicks than uncited sites. Reddit is cited 5.3M times by Google AI Overviews. 44% of ChatGPT citations come from blogs.

Every Piece of Content You Publish Is Training the Models That May Recommend Your Competitors

Cloudflare's crawl-to-referral data tells an uncomfortable story. A decade ago, Google's ratio was 2:1: two pages crawled for every referral sent. Today: Google is 14:1. OpenAI is 1,700:1. Anthropic is 73,000:1. AI companies are extracting extraordinary value from publicly accessible content and returning almost nothing in traffic. The transaction that powered the internet's content economy for thirty years has broken on one side of the equation.

The content licensing market has responded. AI companies committed an estimated $2.92B in multi-year deals with publishers as of January 2025. Reddit earns $203M annually from AI licensing, roughly 10% of total revenue. News Corp secured $250M over five years from OpenAI. Anthropic paid $1.5B in settlements to authors. The Thomson Reuters vs. Ross Intelligence ruling in March 2025 found that AI training on copyrighted content is not fair use.

Matthew Prince's insight at SXSW was instructive: Reddit got dramatically more in AI licensing than the New York Times, despite the Times being far larger, because unique, human-created content that cannot be sourced elsewhere commands a premium. If you do not have Reddit, you do not have Reddit. Uniqueness, not prestige, is what AI companies pay for.

The Last Mile of Trust Is Now the Bottleneck: Correct Is Not the Same as Credible

AI can generate correct information at near-zero marginal cost. That is the promise and the problem. Correct and credible are not synonyms, and in B2B, where someone is about to sign a multi-year contract: the distance between those two words is the entire sales cycle. Forrester's 2025 data shows that while 89% of B2B buyers use generative AI at every purchase stage, 19% feel less confident because of inaccurate or unreliable AI information, and 40% cite conflicting information as their top complaint.

Forrester found that when purchases include AI-generated features or AI-assisted research, the buying group doubles in size: more stakeholders are added to verify what the AI surfaced. The average group is now 13 internal and 9 external participants. More people means more scrutiny. More scrutiny means more trust signals required, not fewer.

This is the paradox of the moment: AI makes information cheaper and buyers more sceptical of it simultaneously. The result is a premium on trust signals AI cannot generate: video testimonials from named customers, analyst recognition, case studies with specific verifiable metrics. Not "50% improvement" but "reduced CAC from $2,400 to $1,100 in six months."

Content Cycles Have Compressed to Days: The Bottleneck Is No Longer Production

AI has compressed content creation from weeks to hours. That should be liberating. For most B2B content teams, it has created a new kind of anxiety: when anyone can produce anything quickly, the question of what to produce becomes urgent and unanswerable at the pace the tools allow. CMI's 2025 survey found 81% of B2B marketers use generative AI tools, but 54% take an ad hoc approach, meaning they are producing more without a clearer strategy for why.

When AI generates most content, the individual piece becomes commodity. What retains value is the insight embedded within it: the data point nobody else has, the framing that makes a familiar problem legible in a new way, the opinion your ICP agrees with but has not yet articulated publicly. The content marketers thriving in this environment have stopped thinking of themselves as producers and started thinking of themselves as editors. Their job is not to generate content. It is to select what matters from the noise.

Advertising Dollars Are Migrating from Platforms to People

The structural argument against programmatic B2B advertising has never been stronger. a16z's framework captures it precisely: "From 1997 to 2024, the core internet business model was distraction: monetising partial human attention through advertisements. LLMs and agents do not get distracted." If agents increasingly mediate discovery and purchasing in B2B, the $291B global online advertising market faces an existential question about what it is buying.

The migration is already observable. Sponsored newsletters from industry experts, podcast partnerships, and community event sponsorships are delivering B2B pipeline at lower cost per conversion than LinkedIn Ads for a growing number of companies. Richard van der Blom's analysis of 1.8M posts found a 50% decline in views for 95% of LinkedIn creators. The accounts maintaining reach were founder voices with genuine engagement, not brand pages running promoted posts. A $5,000/month newsletter sponsorship with the right domain expert often delivers better qualified pipeline than $50,000/month in LinkedIn Ads.

Your Content Library Must Become a Living System: Static Artifacts Are Decaying in Real Time

A blog post published 18 months ago with a statistic from a 2022 report is now a liability. When a buyer asks ChatGPT about your category and your outdated content is cited, the wrong number follows you into a conversation you are not in. The half-life of static content has dropped to months, and for certain types, pricing, comparisons, technical specs. It is now days.

Mintlify monitors connected codebases and proposes documentation updates the moment code ships. GitBook's Computed Content takes an OpenAPI spec and auto-creates documentation that refreshes every six hours. Both platforms support llms.txt, a standard that generates flat Markdown output optimised for AI consumption, making your documentation preferentially citable by LLMs over competitors using static HTML.

Postman's acquisition of Fern (API documentation) and liblab (SDK generation), combined with MCP integration, represents the most complete implementation: your product documentation becomes a real-time discovery channel for both human developers and AI agents simultaneously. Your docs are not a cost centre. They are the first thing an AI agent reads when evaluating your product.

The MQL Is Dead. Forrester, Who Invented It, Has Officially Said So

The marketing qualified lead was invented by SiriusDecisions (now part of Forrester) over twenty years ago. In 2025, Forrester published a blog series called "Saying Goodbye to MQLs: We Promise It's Not Clickbait." Kerry Cunningham, who was at SiriusDecisions when the concept launched, was direct: "Back when we created the MQL, understanding the engagement of a single person was pushing the limits of technology. Twenty years later, technology has evolved immensely, and the buying process has too."

The structural problem is mathematical. MQLs track one person doing one action. B2B buying groups now average 8–13 stakeholders. When one person downloads a whitepaper, the MQL system records a lead. When 12 people from the same company are researching your category simultaneously (on Reddit, through peer referrals, on G2, in private Slack communities) the MQL system records nothing. Teams that shifted to account-level buying group qualification saw forecast accuracy jump above 90% and sales cycles shrink by 25%.

The Real Moat Is Speed of Insight: AI Has Made Speed of Output Irrelevant

When production is free, the constraint moves upstream to judgment. When anyone can generate a blog post in four minutes, the question of which post to generate, and what position to take, becomes the only differentiator. AI has not eliminated competitive advantage in content. It has moved it from execution to insight.

The pattern of companies building durable brand authority in B2B is consistent: they noticed something important before consensus formed, they articulated it clearly while it was still uncomfortable, and they distributed it at the moment their ICP was beginning to feel the problem. The founder who published about the death of MQLs in 2022 seemed contrarian. By 2025, Forrester agreed. BrowserStack built 87,668 keyword rankings through technical insight deployed with consistency. Zerodha built 16 million users through Nithin Kamath's willingness to say things about investing that incumbents were afraid to say.

Speed of insight requires a specific practice: dedicated time each week to scanning signals before they become mainstream. Customer support ticket themes. Earnings calls from category leaders. VC essays and portfolio announcements. Subreddit discussions where practitioners are articulating frustrations that no vendor has yet addressed. The founders who do this weekly, not quarterly, are the ones who appear prescient. They are just paying attention earlier.

The Creator-to-Distributor Inversion: Platforms Own Distribution, People Own Trust

For two decades the content model was simple: you created, platforms distributed. That relationship has inverted. LinkedIn, Reddit, and ChatGPT now hold distribution. They decide which content surfaces, which voices get amplified, which sources get cited in AI Overviews. But what has not changed: the creator holds the trust relationship with the audience. A platform can amplify your words. It cannot inherit your credibility.

Nithin Kamath's LinkedIn posts reach hundreds of thousands through the LinkedIn algorithm. The reason they convert is Nithin, not LinkedIn. The platform is a vehicle. The trust is his. The inversion creates a specific strategic implication: your content strategy should be platform-agnostic, but your trust-building must be deeply personal.

The question to ask is not "what does the LinkedIn algorithm reward this quarter?" It is "what does my ICP trust, and why?" The answer to the second question changes slowly. The answer to the first changes every six months. Build your strategy around the slow thing.

Community Is the Only Unscrapable Moat: AI Cannot Access What It Cannot Crawl

AI can replicate your blog post. It can generate a version of your whitepaper that makes the same arguments in different words. What it cannot do is join your customer advisory board. It cannot access the thread in your private Slack where three customers are comparing your product with two competitors. It cannot be the reason someone in a peer group recommends you to a colleague they trust.

The biggest successes in Indian consumer tech share a community-first pattern. Pocket FM at $200M revenue, Kuku FM, Dream Sports at $4B in gaming transactions: all built content as the wedge and community as the moat. The same architecture is emerging in B2B globally. Pavilion turned a premium revenue leadership community into a $10M business with high net revenue retention. SaaSBOOMi, deliberately small and curated, has built more successful Indian SaaS founders than any accelerator programme in the country.

For Indian B2B founders selling internationally, community does double duty: it builds the trust that geography sometimes requires you to earn explicitly, and it creates a reference network your competitors cannot access. A recommendation from someone in a buyer's peer community is worth more than any analyst badge you can earn. The bar for "community" is not a 5,000-member Slack workspace. A curated 30-person customer advisory group meeting quarterly generates more defensible business value than a large but shallow audience.

If This Is True for You, Do This

Use the conditions below to identify where your company sits. Not everything applies to everyone. The matrix tells you where to act first, where to monitor, and where to build.