WEB3wallet apps / dApp interfaces / token-gated membershipExplore Web3 →
All articlesAI

51% of B2B buyers now start with an AI chatbot: what it means for vendors

More B2B buyers now start their research in an AI chatbot, long before they talk to sales. What that means for vendors and how to stay in the answers.

Pazl teamProduct lead
Published Updated 8 min read
51% of B2B buyers now start with an AI chatbot: what it means for vendors
In this article

A procurement manager at a mid-size manufacturer needs a new inventory system. Ten years ago, she’d Google “inventory management software,” click three review sites, and fill out a form on a vendor page. Today, she opens ChatGPT, describes her warehouse setup in two sentences, and asks it to compare options. By the time a sales rep hears from her, she’s already ruled out half the market, and the vendors she ruled out never knew she existed.

The number behind this shift, that 51% of B2B buyers now start their research with an AI chatbot rather than a search engine or a sales conversation, has been circulating across B2B marketing publications for a while now. Whether the exact figure holds up to scrutiny matters less than what it’s pointing at: buyer research has moved to a channel most vendors haven’t built anything for. If your product information lives in a gated PDF, a sales deck, or a page a bot can’t parse, you’re invisible to a growing share of your funnel before your team ever picks up the phone.

Three different “chatbots,” one confused buyer journey

“AI chatbot” gets used loosely, and the confusion causes real strategy mistakes. Buyers are actually moving through three distinct tools:

General AI assistants, ChatGPT, Google Gemini, Anthropic’s Claude, Perplexity. A buyer asks an open question (“what’s the best ERP for a 50-person distribution company”) and gets a synthesized answer pulled from whatever public content the model can access or has been trained on.

AI search overviews, Google’s AI Overviews now sit above the traditional ten blue links for a large share of commercial queries, summarizing several sources into one answer block. A buyer might never click through to your site at all; they read the summary and move to the next query.

Vendor chat widgets, the chatbot sitting in the corner of your own website, whether it’s a scripted bot from a platform like Intercom or Drift, or something closer to a full AI agent that can hold a real conversation.

Each of these needs a different fix. General assistants and AI search overviews pull from whatever’s publicly indexable and well-structured. A vendor’s own chat widget is the one channel you fully control, and it’s usually the one doing the least.

Your best content is probably invisible to the tools that matter now

Most B2B sites still gate their best material: pricing behind a “request a quote” form, technical specs in a PDF datasheet, comparisons buried in a sales deck a rep sends after a discovery call. That structure made sense when the goal was capturing an email address for the sales funnel. It’s a dead end for an AI assistant trying to answer “how does this compare to Competitor X” or “what does this actually cost.”

A few fixes matter more than a full content overhaul:

  • Put pricing logic in plain text on the page, even if the exact number depends on a quote. “Starting at $X, based on user count and integrations” is something a language model can read and repeat. A form with no numbers on it is not.
  • Use FAQPage and Product schema markup so search engines and AI crawlers can extract structured facts instead of guessing from paragraph text.
  • Write comparison content honestly, including where a competitor genuinely does something better. Models trained to synthesize answers tend to favor sources that read as balanced rather than purely promotional.
  • Stop hiding specs behind lead forms. A buyer using an AI assistant to shortlist vendors will simply skip anything they can’t extract, and the assistant won’t fill out your form on their behalf.

None of this replaces a sales process. It just means the sales process now starts later, after a buyer has already formed an opinion using sources you may not control.

What a good on-site AI agent actually does

The chat widget on your own site is the one place you can build something better than a generic answer engine, because it can know your actual catalog, actual pricing, and actual availability instead of guessing from public web text.

We built something adjacent to this for an events agency that had hit a wall on growth: the more weddings and corporate events they booked, the more their staff drowned in repetitive chat threads, availability questions, vendor logistics, follow-ups on quotes. Hiring more coordinators was expensive, and the agency didn’t want service quality to slip during peak season. We built an AI agent, internally called Vera, that handles text, voice, photo, and video messages from clients, keeps context across the whole conversation, creates tasks in Trello automatically, and only pulls in a human coordinator when a conversation actually needs judgment: a pricing exception, a complaint, a decision the agent isn’t authorized to make. Clients get an answer at 11 p.m. on a Saturday instead of a promise of a callback on Monday. The team handles more events without adding headcount, and coordinators spend their time on negotiation and service instead of typing the same answer for the fortieth time.

The B2B parallel is direct. A buyer researching your product at midnight doesn’t want a form that promises a callback in two business days. They want a real answer about pricing tiers, integration support, or contract terms, right then, from something that can actually check your CRM instead of reciting a canned script. The gap between a chatbot that answers five FAQ questions and an agent that reads a buyer’s actual situation and gives a specific answer is exactly the gap between a bot buyers abandon after one message and one they trust enough to keep talking to.

Fixed-price engagements for this kind of build typically start around $3,300 (≈£2,600) for a scoped, working version, running up depending on how many systems it needs to talk to: your CRM (HubSpot is the common default for B2B go-to-market teams, with NetSuite showing up more as companies move into the mid-market and need tighter financial data ties, and QuickBooks still handling the books for a large share of smaller B2B sellers). Price and scope get fixed in the contract before work starts, which matters here specifically because “AI agent” as a category has a wide range of what people mean by it.

The privacy and disclosure details nobody budgets for

An AI chat agent on a B2B site is, functionally, a system that collects personal information: names, work emails, sometimes phone numbers, and a transcript of everything a prospect typed. In the US, there’s no single federal privacy law covering this, but California’s CCPA/CPRA framework does, and if you’re selling to buyers anywhere in California, your chat widget’s data practices are in scope regardless of where your company is based. That means your privacy notice needs to actually describe what the chat tool collects and where it goes, your site needs to honor Global Privacy Control signals from browsers that send them, and you need a working process for someone to request deletion of their chat history if they ask. The California Privacy Protection Agency and the state attorney general’s office are the enforcement bodies here, and a growing number of other states have passed similar laws with their own thresholds, which is why most companies selling nationally end up building one privacy workflow that satisfies the strictest state rather than tracking each one separately.

A second, less obvious issue: if your chat agent cites customer reviews, case studies, or comparison claims (“most customers see faster implementation than with Competitor X”), that falls under the Federal Trade Commission’s rules on endorsements and substantiation. An AI agent repeating an unverifiable claim is treated the same as a sales rep saying it out loud. If you can’t back a number with a real source, don’t let the bot say it.

None of this requires a legal department to get a basic B2B chat agent live. It does mean the build should include a consent notice, a documented data map of what the agent stores and for how long, and a deletion process someone can actually execute, not just a promise in the privacy policy that nobody built the mechanism for.

What this actually costs to fix

Ranges below are broad market indicators, not quotes. Actual cost depends heavily on how much of your content and tech stack already exists versus needs to be rebuilt.

Fix What’s typically included Rough market range
Content restructuring for AI extraction
FAQ pages, schema markup, rewritten pricing and comparison pages
$2,000–$8,000 (≈£1,500–£6,300)
Scripted chat widget upgrade
Better routing, CRM logging, basic qualification questions
$3,000–$10,000 (≈£2,400–£7,900)
Full conversational AI agent, CRM-connected
Multi-format input, task creation, human handoff logic
$10,000–$40,000+ (≈£7,900–£31,500+), scope-dependent
Privacy and disclosure workflow
Consent notice, data map, deletion process, GPC handling
$1,500–$6,000 (≈£1,200–£4,700)

Most agencies quoting fixed-price work in this space bill against milestones, with a deposit collected via ACH or a card charge through a processor like Stripe and the remainder due at delivery or sign-off. That structure protects both sides: you’re not paying in full for something that isn’t built, and the vendor isn’t carrying the full cost of the build on faith.

Frequently asked questions

Do we need to rebuild our whole website to be “AI-ready”? No. The highest-value fixes are usually pricing transparency, schema markup, and un-gating comparison content, none of which require a redesign. A full site rebuild only makes sense if your current structure is genuinely broken for other reasons too.

How long does it take to get a working AI chat agent live on our site? A scoped, CRM-connected agent handling a defined set of buyer questions typically takes a few weeks from signed scope to launch, depending on how many systems it needs to integrate with. Anything promising a working, tested agent in days is either very narrow in scope or skipping testing.

What do we need to provide before work can start? Access to your CRM (HubSpot, NetSuite, or whatever you run), your actual pricing logic even if it’s tiered or quote-based, and someone on your team who can answer product questions the agent’s builder won’t know off the top of their head. Projects stall most often when this access or information takes weeks to arrive after the contract is signed.

Is a chatbot on our site enough, or do we also need to fix how AI assistants like ChatGPT see us? Both matter, but they’re separate problems. A chat widget only helps buyers already on your site. Getting picked up correctly by ChatGPT or Google’s AI Overviews depends on your public content being structured and specific enough for those systems to extract and cite accurately.

What’s the biggest risk in building this fast? Skipping the privacy workflow. A chat agent that logs names, emails, and conversation transcripts without a deletion process or a consent notice creates exposure that’s cheap to avoid up front and expensive to fix after a complaint or an audit.

If your buyers are already asking an AI assistant about you before they ask you, the fix starts with what that assistant can actually see and what your own chat can actually do. Email us at hello@pazl.ai and we’ll tell you honestly whether your site needs a content fix, a chat build, or both.

Written by

Pazl team · Product lead

Pazl designs and builds websites, mobile apps, AI automation and Web3 products.

About Pazl →