When ChatGPT asks you which supplier is best — and you're not there

Data from the Forrester B2B Buyer Journey 2026 survey are clear: 72% of B2B buyers today use ChatGPT when evaluating suppliers. Additionally, 44% turn to Perplexity when shortening their shortlist, and according to Bain & Company, the average B2B buyer performs 17 AI queries per week. This is no longer an experiment — it's a standard part of the purchasing process.

And here comes the problem. According to the Crackle PR Q2 2026 AI Citation Benchmark, 51% of B2B technology brands have zero citations across ChatGPT, Perplexity, and Gemini. In other words: the customer asks AI who is best in a given category, and half of the companies simply don't exist in the answer.

For Czech B2B companies — from software houses to manufacturing enterprises seeking foreign partners — this means only one thing: if AI doesn't know you, you're not in the game. And it's not just about English-speaking markets. Both ChatGPT and Gemini understand Czech very well, and Czech buyers are increasingly using them for supplier research — whether it's for an ERP system, a logistics partner, or a manufacturer of industrial components.

Seventeen touches before you pick up the phone

Gartner states in its B2B Buying Journey 2026 report that before a B2B buyer contacts a salesperson, they go through an average of 17 touchpoints. Forrester adds that 71% of them read an article from a reputable media outlet before requesting a demo. And AI search is becoming the channel that serves these articles to buyers.

The Edelman Trust Barometer 2026 then adds a crucial number: B2B buyers trust an AI-recommended supplier 3.4 times more than one they encounter in paid advertising. Trust is shifting from banners to language model responses.

GEO: A new discipline without which you won't be visible

A new field has emerged — Generative Engine Optimization (GEO). While SEO optimizes websites for search engines like Google, GEO optimizes content, authoritative signals, and data structure so that language models themselves notice you. The difference is fundamental: AI models don't cite pages just based on keywords, but based on domain authority, frequency of media citations, and source trustworthiness.

Crackle PR found that 71% of citations in ChatGPT come from earned media (i.e., articles in the media, not from one's own website), with only 29% from owned content. Press releases are failing — only 12% of Google AI Overviews responses link to them. The correlation between domain authority and citation frequency is r = 0.62, meaning one quality article in a respected media outlet can carry more weight than a dozen posts on one's own blog.

For Czech companies, this means a specific task: an AI visibility audit. Ask ChatGPT, Perplexity, and Gemini exactly as your customer would — "best logistics provider in the Czech Republic", "industrial automation supplier Central Europe", "quality custom software developers Prague". If you don't appear, you have a measurable visibility problem that directly impacts your pipeline.

Czech specifics further complicate the situation. Czech-language content has significantly less representation in the training data of large models than English — therefore, AI models more often cite English sources even when queried in Czech. For domestic companies, a bilingual GEO strategy (in Czech and English) is practically a necessity if they want to be visible to both domestic and international customers.

Anthropic and Blackstone bet 1.5 billion on implementation, not models

While media attention focuses on benchmark races between GPT-5.6, Claude Fable 5, and Gemini 3.5, Anthropic made an inconspicuous but strategically crucial move. Together with investment giants Blackstone, Goldman Sachs, and Hellman & Friedman, it launched a joint venture Ode valued at $1.5 billion (approximately 34 billion CZK).

Ode doesn't build better AI models. Instead, it sends engineers directly to client companies, where they build and configure AI workflows on-site, connected to real company data and existing systems. This is a "forward-deployed engineering" model, made famous by Palantir in the defense sector — and which Anthropic is now applying to the commercial sphere.

Why is this groundbreaking? Because it openly admits what no one wanted to say aloud until now: the problem with enterprise AI is not in the models, but in the implementation. A July 2026 study by MIT FutureTech and Carnegie Mellon shows that only 11% of S&P 500 companies have deeply integrated AI. The rest are stuck in the pilot phase.

For European, and thus Czech, companies, this carries an interesting lesson. If even the biggest players in the market admit that implementation is the real bottleneck, there's no point in endlessly comparing model benchmarks. A much more important question is: do you have someone (or a team) in your company who can not only select AI but also deploy it into live operation?

Token hangover: Why Uber exhausted its annual AI budget already in April

Nikesh Arora, CEO of Palo Alto Networks, put it bluntly: token prices must drop by 90% within two years for enterprise AI to be cost-effective at a production scale. This is not an exaggerated warning — Uber this year exhausted its entire annual AI budget already in April. Microsoft had to temporarily suspend registrations for GitHub Copilot because the infrastructure couldn't keep up. Corporate bills for AI agents are growing faster than most CFOs planned.

To illustrate: at current API call prices, a company with 500 employees deploying AI assistants across departments can easily spend hundreds of thousands of Czech crowns per month just on token fees. And that's without accounting for integration, training, and maintenance costs.

Czech companies are in a specific situation in this regard. While American corporations have budgets that can absorb even inefficient pilots, a medium-sized Czech company cannot afford an expensive AI experiment twice. It is all the more important to plan costs in advance — model the token price trajectory, set a budget cap, and most importantly, measure actual return on investment, not just the number of generated lines of code.

What should Czech companies do about this? Three concrete steps

1. Conduct a GEO audit this week. Open ChatGPT, Perplexity, and Gemini and ask them seven questions your ideal customer would use. Record where (and if at all) you appear. If the result is fewer than three mentions, you have a measurable deficit that impacts your business.

2. Invest in earned media, not another blog post. Data from Crackle PR show that AI models primarily cite articles from respected media (TechCrunch, Forbes, VentureBeat), not press releases or corporate blogs. One quality PR article carries more weight than a dozen of your own posts. For Czech companies, this means targeting media like CzechCrunch, Hospodářské noviny, or industry titles with good domain authority.

3. Don't jump for another AI model — solve implementation. The lesson from Anthropic and Ode is clear: the models are already good enough. The problem is that companies don't know how to deploy them. Before spending more money on licenses, verify that you have people (internal or external) who can get AI up and running operationally. Otherwise, you'll face Uber's fate: you'll exhaust your budget before you see results.