Where Do Most Decision-Makers Look for a Supplier in 2026?

51 percent of B2B decision-makers today more often start their supplier research in an AI chatbot than in Google - a year earlier that share was 29 percent. The drop in Google queries is a signal that the starting point of the purchase process is migrating to a layer where there is no your CRM, no your salesperson, and no your Google Analytics.

I measured this on myself. In August I ran 20 queries through ChatGPT, Perplexity and Gemini - in Polish and in English, phrased the way a customer looking for a company in my category would ask. In the first run, ChatGPT mentioned AGAPE in 5 of 20 answers, Perplexity in 3, and Gemini in 2, giving 10 mentions out of 60 answers. In the remaining cases, the models recommended companies I know from the market, as well as several I had never heard of before. I'm running a retest on September 30, because AI visibility is a metric you have to measure in a cycle.

Monday, 9:00 a.m. You open your CRM and see three fewer inquiries than usual, and in Search Console the curve has been dropping for the third month in a row. You call two regular clients and hear that they were just comparing offers - except no one called you, because your client chose a supplier in a chat window a week ago. Zero trace in the CRM and zero chance of a counteroffer.

I know this sentence from conversations better than from reports: "I'm out at meetings from referrals myself, so I don't have time to manage the company." A B2B SME owner sees the shift in B2B buying habits in the numbers: fewer contact forms, fewer phone calls, fewer leads from organic search results. On top of that comes a figure that cuts off any debate about exceptions: Haney Strategy, commenting on G2's data, points out that 94 percent of buying groups have a finalized supplier shortlist before anyone on the sales team even picks up the phone, and the supplier who gets the first call wins about 80 percent of deals. The battle for the contract is decided before the business owner even finds out the client was looking for a supplier at all.

Where research has moved: data that is reshaping the funnel

The drop in Google queries now has the face of concrete measurements. Search Engine Land reports that desktop searches per user in the US fell year over year by nearly 20 percent, while in Europe that decline is 2-3 percent. A Bocconi University study shows the mechanism behind this shift: after users gain access to ChatGPT Search, traditional queries initially drop by 9.4 percent, and after 20 weeks by 17 percent. Rabbit Rank adds the other side of this phenomenon - traffic to reference and niche pages drops by 15-30 percent, because users get part of their answer without clicking through to results. Zero-click search is now the standard for the first contact with a category, and Google is responding with its own features: AI Overviews and AI Mode have billions of active users a month.

In B2B, the migration is faster. G2's "The Answer Economy" report from March 2026, based on a survey of 1,076 B2B decision-makers, shows that 51 percent of B2B software buyers more often start their research with an AI chatbot than with Google, up from 29 percent a year earlier. Chatbot use at any stage of the process rose in that time from 60 to 71 percent, and 53 percent of buyers rate AI-assisted research as more productive than traditional search - a year earlier only 36 percent felt that advantage.

The impact of these recommendations on deals is bigger than most business owners assume. 69 percent of buyers changed their original supplier choice under the influence of a chatbot recommendation, and 85 percent think more highly of a supplier mentioned in an AI answer. Four out of five people say a chatbot sped up their purchasing decision, 41 percent use it to compare suppliers' strengths and weaknesses, and another 41 percent regularly use Deep Research mode to evaluate suppliers.

Forrester, in its "The State of Business Buying, 2026" report based on a survey of nearly 18,000 buyers, states that 94 percent of them use generative AI in the purchasing process. That's a different metric from the 94 percent of buying groups with a closed shortlist - one speaks to tools, the other to the moment of decision, and it's worth keeping them separate. The tension shows in the details: 36 percent of buyers feel more confident thanks to genAI, and 20 percent feel less confident because they encountered imprecise or incorrect information. That uncertainty pushes verification of answers back into the network of trust: reviews, communities and analysts, and in Forrester's research, nearly twice as many buyers point to generative AI and conversational search engines as their most significant source of information - ahead of supplier websites and contact with a salesperson.

The decision-making group is growing along with this shift. The average B2B purchasing decision today involves 13 internal stakeholders and 9 external participants, and WebFX adds a number that explains why sales finds out about a tender too late: 83 percent of buyers finalize their purchasing requirements before they even talk to a salesperson.

Responsive shows that a quarter of buyers already treat generative AI as their main supplier-research tool, and two-thirds use it as much as or more than search engines. In the tech industry, adoption is even faster: 80 percent of buyers use genAI at least as often as search engines, and 56 percent name chatbots as their main channel for discovering new suppliers - in other industries, 28 percent do so. Sopro, in a study cited by 1DOT, adds a generational dimension: 66 percent of top-level UK decision-makers use ChatGPT, Claude, Copilot or Perplexity for supplier research; in the 25-34 age group it's 85 percent, and in the 55+ group only 23 percent. Your next contracts will be signed by people for whom a chatbot is the first place they check the market.

The mechanics of the Answer Economy: from click to answer

The old B2B funnel looked like this: need, a phrase typed into Google, clicks, a whitepaper, asking friends, reviews, shortlist, phone calls, offers - every stage left a trace in analytics and in the CRM. The new funnel compresses the first steps into a single chat window, where the buyer describes the industry, company size, budget constraints and legal requirements, and the model returns a synthesized recommendation in the form of a list of three to five suppliers with justification.

Here's what real prompts, as cited by 1DOT, look like:

• "Which construction consulting firms in the Midlands specialize in pre-development planning?"
• "We need a new ERP system for a manufacturing company with 200 employees - what should we look for?"
• "Which marketing automation agencies in Poland serve manufacturing companies with 100 to 300 employees?"

None of them contain a company name. There's a problem, context, and the expectation that the model will do the work that used to be done by a person on the buying team. When 51 percent of B2B decision-makers start their research with a chatbot - most often ChatGPT, which accounts for around 62-63 percent of B2B software supplier discovery - your first conversation with the customer happens without you.

Models build these answers on specific sources. ChatGPT, in about 87 percent of cases, answers in line with the top 10 results in Bing's index, while Perplexity favors fresh content and cites sources heavily. G2 describes this shift as a move from a "reference" model to an "inference" model: previously the engine pointed to sources and the user did the synthesis, today the model does the synthesis and the user reads a finished answer. Your spot on the list depends on what the model has to read about you.

Buyers don't stop working once they close the chat window. The result ends up in Word, Google Docs or Outlook, where Copilot or Gemini prepares a comparison and a brief for the buying team, along with a set of questions for suppliers. The model's recommendation becomes a document circulating inside the client's company, and your company is present in it or absent from it before anyone sends out an RFP.

For many buyers, this is the first and only contact with the supplier market. 95 percent of ChatGPT users still use Google, so AI adds a new front alongside search - and it's on this front that the shortlist decision gets made. Brand in ChatGPT and visibility in Google are two fronts of the same battle, and your presence on both is decided by the evidence models can read.

Four layers models build an answer from

• Assistants: ChatGPT, Perplexity, Gemini, Copilot and Google's AI Overviews - this is where the first contact with a category happens.
• Review platforms: G2, Clutch, TrustRadius and Google reviews - this is where the model verifies reputation.
• Communities: industry groups, LinkedIn, private threads - this is where customers share experiences the model can cite.
• Documentation and backlinks: product pages, help centers, articles and link profile - this is where the model finds material for synthesis.

Visibility in AI is built across all four layers at once, because the model draws on several sources before writing an answer.

Trust signals: what models read as evidence

G2 asked buyers what most strongly reinforces trust in AI recommendations, and the answer is clear: citing data from a review site. For half of heavy AI users, this is the single most important trust signal, and buyers name reviews themselves as the second most important influence on the shortlist, right after the chatbot.

This evidentiary layer translates into deals. 33 percent of buyers chose a supplier they had not previously known about solely because a chatbot recommended it. For smaller companies, that's the best news in the whole report, because the game for a spot on the shortlist is open, and getting on it requires clear trust signals that can be built from real customers and results.

Trust signals to audit

• Google reviews from recent months, because data freshness works in your favor.
• Profiles and rankings on review sites: G2, Clutch, TrustRadius and industry roundups.
• Case studies with numbers, published in open channels the model can read and cite.
• Consistent category language: the same service name on your website, in profiles and in materials.
• Citations on review sites that the model points to as the source of its answer.

Visibility in AI is built from the same elements that build trust with people: concrete reviews, results, and a consistent description of what you do.

The cost of doing nothing: what you lose when models don't know you

Not being present in model answers means losing control over the first phase of the funnel. Google queries decline because part of the traffic migrates to the AI layer, where there's no direct trace in Google Analytics - and 69 percent of supplier-choice changes happen beyond the sales team's reach, which means competitors present in the recommendations gain an advantage without the business owner even knowing.

This advantage compounds every quarter. In purchases involving genAI elements, the decision-making group doubles in size and the process lengthens by about 30 percent, giving a supplier already on the shortlist more chances to strengthen its position at every stage of verification. 83 percent of B2B purchases today involve genAI elements, so this scenario applies to most of the tenders that will land on your desk in the coming quarters.

The TCO of a lost lead, in four line items

• Lost revenue from a lead that never made the shortlist - a deal you'll never see in your CRM and never count in your win rate.
• The cost of keeping the old channel running while traffic to reference pages drops by 15-30 percent - you keep paying for SEO, content and ads at declining effectiveness, and the cost per completed task rises every quarter.
• The cost of rebuilding visibility: reviews, profiles, case studies, a Schema Graph and consistent category language - one position at a time, month after month.
• Regulatory and reputational risk: a burned domain operates under the deliverability regime (spam rate, sending reputation), while penalties fall under the personal-data-protection regime, where consent and opt-in are what matters - two separate line items in the risk calculation you run in cold blood, without scare tactics and without guarantees.

Every quarter without visibility in AI is a quarter in which someone else answers your customers' questions and builds relationships you'll only find out about a year from now.

Three steps you can check today

This mechanism can be broken down into parts and addressed point by point, and the first step will take you fifteen minutes. 95 percent of ChatGPT users still use Google, so your SEO and content remain the foundation - that work keeps paying off and needs a new front in the form of reviews, industry rankings and case studies that models treat as credible sources. Adapting to AI search extends what you're already doing: the same foundation, a new front.

The 20-query test

• Open ChatGPT and enter a prompt describing your category and the customer profile you're looking for.
• Check whether your company appears on the list, and note who shows up instead of you.
• Repeat the same test in Perplexity, where the sources and the list look different.
• Run the same test in Gemini and compare the three results.

The result will show whether you're visible in the place where 51 percent of decision-makers start their purchase process, and along the way you'll get a list of the companies you're losing to before anyone even calls. In my case, the first run of this test produced 10 mentions out of 60 answers, and I'm running a retest on September 30, because after August I rolled out changes to my signals and want to see whether they're working. At AGAPE we run the same test before every piece of visibility work, because without a baseline result you can't measure the effect.

Today's checklist

• Enter your prompt into ChatGPT, Perplexity and Gemini.
• Note who showed up on the list instead of your company.
• Check your Google reviews and your profiles on review sites.
• Pick one case study with a number and publish it in an open channel.
• Set a retest for the same day, 30 days from now.

How many of the three models mentioned your company - and did it even make the list at all? Write TEST in a DM and I'll send you the spreadsheet of 20 queries I use for this check - complete with a spot for your result. Check before a competitor does.