More B2B buyers now start with an AI chatbot: what it means for vendors

A procurement manager at a manufacturer needs a new inventory system. Ten years ago, she'd search for "inventory management software," click through a couple of reviews, and fill out a form on a vendor page. Today, she opens ChatGPT or Alice, describes her warehouse 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.

Where the number comes from and what it actually means

The shift is often summed up as "51% of B2B buyers now start with an AI chatbot." We traced it to the source: G2's The Answer Economy report, published in April 2026. In March 2026, G2 surveyed 1,076 B2B software buyers and decision-makers, and 51% of them now start their research with an AI chatbot more often than with Google. In G2's 2025 report, that figure was 29%. At the same time, 61% use AI search alongside Google rather than instead of it.

Two caveats matter. This is a survey of software buyers, not of all B2B purchasing. It was run by a review platform, and it has no separate data for Kazakhstan; we found no local study on the topic. But the direction is clear even without an exact figure: vendor research has moved to a channel most companies 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, Gemini, Claude, Perplexity, Alice AI. A buyer asks an open question ("which inventory system suits a 50-person distributor") and gets a synthesized answer pulled from whatever public content the model can access or was trained on.

AI answers in search. In Google these are the AI Overviews above the regular results; in Yandex, the Alice AI answers in Search, which are built from pages in the results and link to their sources. A buyer might never click through to your site at all: they read the summary and move to the next query. In Kazakhstan both engines matter: according to StatCounter, Google had 68.25% of searches in the country in October 2026 and Yandex 29.61%.

Vendor chat on the website and in messengers. The chat window in the corner of your site, a bot in WhatsApp or Telegram: a scripted bot, or a full AI agent that can hold a real conversation.

Each of these needs a different fix. General assistants and AI answers in search pull from whatever is publicly indexable and well-structured. Your own chat on the site and in messengers is the one channel you fully control, and it's usually the one doing the least. We cover how to get into AI answers in detail in how to get your company cited by ChatGPT and Perplexity.

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, comparisons buried in a deck a rep sends after the first call. That structure made sense when the goal was capturing a contact for the sales funnel. It's a dead end for an AI assistant trying to answer "how does this compare to the competitor" 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 a stated amount, 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 (schema.org) 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 that synthesize answers from sources tend to favor material that reads 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.
• Keep the Russian and Kazakh versions of your pages and your directory listing current. Buyers in Kazakhstan ask in both languages, and a 2GIS listing is one more public place where your address, phone and hours should match the website.

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

What a good on-site AI agent actually does

The chat 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.

An example. An events agency hits a wall on growth: the more weddings and corporate events it books, the more its coordinators drown in repetitive threads about availability, vendors and quote follow-ups. Hiring more coordinators is expensive, and service quality can't slip in peak season. An AI agent in that situation handles text, voice, photo and video messages from clients, keeps context across the whole conversation, creates tasks in the task tracker on its own, and only pulls in a coordinator when a conversation 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, 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, integrations 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 situation and gives a specific answer is exactly the gap between a bot buyers abandon after one message and one they keep talking to.

What it costs with us: an AI agent on one channel working from your knowledge base starts from 480 000 ₸; an AI sales agent with CRM and telephony is 800 000–2 150 000 ₸, depending on how many systems it needs to talk to: amoCRM or Bitrix24 for deals, 1C for prices and stock, WhatsApp and the website as channels. Price and scope are fixed in the contract before work starts, which matters here specifically, because "AI agent" means very different things to different vendors.

The data and disclosure details nobody budgets for

An AI chat agent on a website is, functionally, a system that collects personal data: names, work emails, sometimes phone numbers, and a transcript of everything a prospect typed. In Kazakhstan this is governed by the Law of the Republic of Kazakhstan "On Personal Data and Their Protection":

• Consent. Data may be collected with the person's consent (Article 7, paragraph 1), and it must be possible to prove the consent was obtained (Article 8, paragraph 1). So you need a clear consent text before the conversation starts and a record of it.
• Storage in Kazakhstan. Personal data is stored in a database located in the country (Article 12, paragraph 2). The agent's conversations and requests live on a server in Kazakhstan.
• A model abroad. If requests containing data go to a language model abroad, that is a cross-border transfer, and it has to be named in the consent.
• Withdrawal. A person can withdraw consent, and the operator has fifteen working days to act on it (Article 8, paragraph 7). You need a working way to delete a chat history on request, not just a promise in the privacy policy.

The details and wording are in our guide to personal data on a website.

A second, less obvious issue. The Law "On Artificial Intelligence" No. 230-VIII requires users to be informed when services are provided using AI (Article 21, paragraph 1), and makes owners of AI systems responsible for their output complying with the law (Article 21, paragraph 5). In plain terms: if your chat agent cites reviews, case studies or comparison claims ("our customers go live faster than with the competitor"), you answer for it as if a sales rep had said it. 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 chat agent live. It does mean the build should include a consent text, a description of what the agent stores and for how long, and a deletion process someone can actually execute. Check the details of your case with a lawyer.

What this actually costs to fix

The cost depends heavily on how much of your content and tech stack already exists versus needs to be rebuilt. So instead of market ranges, here is what each piece of work includes and what drives its price.

Our reference points: an AI agent from 480 000 ₸, a sales agent with CRM and telephony at 800 000–2 150 000 ₸. Content restructuring is estimated after we review the site. The price is fixed in the contract, with a six-month warranty. Projects like this are usually paid in stages: the client isn't paying in full for something that isn't built, and the developer isn't carrying the whole build on faith.