From Reference to Inference: How B2B Buyers Skipped Google and Went Straight to AI
11 Feb, 2026
•
9mins read
In this week’s episode of Voices of Search, we spoke with Tim Sanders, Chief Innovation Officer at G2, about a seismic shift in B2B buying behavior. With 79% of global B2B buyers saying AI search has changed how they conduct research, the game has fundamentally changed. G2 serves over 100 million software buyers annually and has a front-row seat to this transformation—and the data is startling.
By August 2025, 50% of B2B buyers started their research on chatbots instead of Google. That number was 29% just four months earlier. Tim revealed how buyers moved from “reference to inference,” why commercial citations have different rules than background citations, and what enterprise brands must do to win verification layer work in an AI-first discovery world.
Key Takeaways From this Episode:
The Death of the Librarian Model
Tim’s career spans from Alta Vista and Yahoo to watching Google’s PageRank revolution. It’s a shift that varies greatly in both speed and scope.
“The software buyer outside of procurement has moved from reference to inference,” Tim explained. “Reference meant you manually went to Google, entered keywords, got blue links, and began to build a spreadsheet. That process took between five and 12 hours.”
Now buyers do inference: “Share the goal, share the pain, let AI do all the synthesis, and let it do inference to give you the short list.”
In April 2025, 29% of B2B buyers started on chatbots instead of Google. By August, that number hit 50%. The tipping point happened within just four months.
Why AI Wins: Buying Isn’t Your Job
“Unless you’re in procurement, buying stuff isn’t your job. Your job is producing value and operational excellence. It’s a tax on our time,” Tim said.
When scope increases—from headcount freezes or reductions—tools that boost productivity get adopted fast. Buyers now use “best of” prompts: “Give me the best three CRM solutions for a medium independent hospital with mobile endpoints.” They’re one-shotting the short list in minutes.
But they don’t stop there. “The user doesn’t trust enough to stop the session and contact salespeople. But they still do verification work and click through.”
This creates verification layer traffic—and why G2 expects a million human visitors from LLMs in 2025.
Commercial Citations vs. Background Citations
Not all citations are equal, and Tim draws a sharp distinction that most marketers miss:
“Background citation is everything but the ‘buy these three CRM’ part. You don’t have to verify it because there’s no risk,” Tim explained. “Click-through for background citation is probably one quarter of 1%.”
Commercial citations—the purchase recommendations—have radically different engagement: “That click-through rate is somewhere between 3 and 15% depending on price tag, risk factor, and the user’s trust in the model.”
G2’s success stems from prompt-matching. According to Radix, 60% of G2 citations come from “best of” guides—listicles, grids, and awards.
“Our ‘best of guides’ win because they’re based on thousands of customer reviews. They’re all verified. We require screenshots proving you own the software. We have your LinkedIn profile,” Tim said.
Models are also tuned to avoid regrettable purchases. “They want to make sure you’re satisfied. A lot of times, we get the traffic vendors used to get in organic search, but traffic from language models is much higher conviction.”
Calendar Age vs. Cognitive Age
Tim also introduced a framework that changes buyer targeting: calendar age versus cognitive age in buying.
“When buyers come to you, they’re earlier from a calendar standpoint but much further along cognitively,” Tim explained.
What took 13 days now happens in three. “3 days in, they’re talking to the vendor, but they’re 90% through the bundle. They’ve got a short list, a rubric for deciding. They’re just tiebreaking.”
This paradox changes conversion strategy. Despite being further along, buyers are fresher. “They haven’t experienced fatigue. They’ll respond to novel offers that may not be verification work. That’s why they sign up for webinars and newsletters—they’re fresher with more bandwidth.”
Time kills curiosity. Traditional buyers arrive exhausted. AI-assisted buyers arrive energized, despite being closer to decisions.
The Regrettable Purchase Framework
Understanding AI recommendations requires understanding their core objective: avoiding regrettable purchases.
Tim drew parallels to Facebook. “Zuckerberg got insights that regrettable minutes are the number one signal of platform deterioration. Reduce regrettable minutes, get more loyalty. Same at OpenAI and Gemini.”
For OpenAI’s future, it’s clear: “2027-2028, their top two revenue sources will be ads and shopping. Their multiple has to be like Meta meets Amazon.”
Shopping focus means teams tuning specifically for purchases. “The dominant methodology is avoiding regrettable purchases.”
Commercial citations operate under entirely different neural networks than background citations. “Many purchase recommendations don’t link to vendors unless they’re in shopping deals with OpenAI. They link to proof points—review sites, Gartner reports, bloggers—to avoid regrettable purchases.”
Keywords Are Not Prompts
One of Tim’s strongest warnings: keywords don’t equal prompts, and assuming they do creates massive blind spots.
“The use of keywords will no longer reflect how people prompt AI,” Tim said. He shared an example about voice keyboard software.
A competitor called their product “AI voice dictation” and bought all related keywords. But actual prompt data showed everyone in the ICP—IT people driving enterprise adoption—called it “voice keyboard.”
“Tremendous misalignment. All the voice IT cool kids never use Google. You won’t see it in AdWords keyword research. You’ll miss it.”
Early adopters moved to ChatGPT and “find their own way of talking about things in the prompt world that may not show up in keywords.”
Worse, keyword optimization creates false positives. “You’ll think you’re doing better than you are because you’re going to win that which you engineer for.”
The solution? Get actual prompt data from panel providers like Scrunch.
Expected Value: The New Marketing Discipline
Tim advocates for applying expected value (EV) calculations to every AI visibility decision—a framework borrowed from poker and forecasting expert Nate Silver’s book “On the Edge.”
“We need to have an edge in every decision by using expected value calculations to take the emotions or the obsolescence out of a decision,” Tim explained.
Where EV Applies
Common scenarios requiring EV analysis:
The good news? “ChatGPT and Gemini are great at doing low hallucination EV work if you give it the right context.”
The Changing Value of Reviews
Reviews remain critical, but what makes them valuable has shifted. Tim highlighted three evolving factors:
Models don’t value five-star ratings with “Buy this book. It’s great” blurbs. They value narrative depth that helps predict user satisfaction.
Why Video and Transcripts Win
When asked what media form would be most valuable for AI visibility, Tim didn’t hesitate: “Video with transcripts.”
The reasoning goes beyond simple accessibility. “They’re answer-shaped. They’re natural language. They’re human. They are perceived as real, and they match the language training for most models.”
The Forcing Function Effect
Video creates what Tim calls a “forcing function” for better content. “When we’re creating landing pages, we have a messaging approach: problem, solutions, feature, benefits, proof. But when we’re writing an FAQ, we have compelling questions and answers. Think about our conversation today. It’s very much Q&A.”
This natural Q&A format aligns with how models are trained. “It creates that very answer-shaped content the models prefer in verification work. They’re actually trained to understand that better.”
Most importantly, video prevents corporate messaging pollution. “You can’t make corporate messaging out of one of these conversations. We’re just talking to each other. You can’t push me back to say the thing like your marketing team wants to always say the thing.”
Models detect and downweight vendor messaging designed to manipulate purchases. Authentic conversation can’t be corporatized, which is exactly why models value it.
Tim also critically notes that “you have to produce the transcript. Make sure the transcript’s visible to the bot, or you’ve missed it completely.”
Answer-Shaped Content vs. Message-Shaped Content
One of Tim’s most actionable insights centers on content structure. “Refactor content to be answer-shaped instead of message-shaped and to incorporate more natural language.”
Traditional B2B content follows messaging frameworks: problem, solution, features, benefits, proof. This structure signals vendor bias to AI models.
Answer-shaped content responds to questions with natural language explanations. It sounds like humans talking, not marketing copy. FAQs, interviews, podcasts, and video conversations all naturally create answer-shaped content.
“It fits better within their allergy to corporate vendor messaging that might create a regrettable purchase,” Tim said. Models are explicitly tuned to detect and discount sales language.
The Markdown Advantage
Tim identified one underrated signal for AI optimization: markdown summaries at the top of pages.
“Markdown language key takeaways from a page that are placed at the very top of the page are going to create the most crawling, reference, and training inclusion of your web pages than anything,” Tim explained.
This applies especially to sales pages, product pages, and pricing pages. “A small markdown file with the key four takeaways will dramatically improve the chances that your content is considered during test time.”
Pricing also deserves special attention. “It’s so complicated for business buyers. Sometimes the opaqueness of pricing gives leverage to the seller, but you need to run EV because you have to now account for the fact that it creates confusion for the model.”
Don’t Sleep on Co-Pilot
While most B2B marketers focus on ChatGPT and Gemini, Tim warns against ignoring Microsoft Co-Pilot.
“5% Co-Pilot—don’t sleep on Co-Pilot. Enterprise buyers get guard-railed. All they have is Co-Pilot Chat, and they’re enterprise buyers.”
Tim calls Co-Pilot “the Canadian version of ChatGPT” because “Canada always gets it 18 months later.” But the inference quality is catching up. “Even though Co-Pilot is not as rich in inference as GPT 5.2, it’s close. A lot of the things you figure out for ChatGPT will help you in Co-Pilot.”
For enterprise-focused brands, Co-Pilot deserves analyst-level attention. “If you’re an enterprise and you sell to enterprise people, you need to treat Co-Pilot like an analyst. It may not be Gartner. It may not be Forrester. It’s IDC. Treat it like an analyst. Respect it. Study it. Understand how it lives and breathes.”
The Fundamentals Still Matter
Despite all the change, Tim emphasized continuity over disruption. “SEO has never been more important because real-time research still relies on ranking to some extent outside of deep research. The fundamentals of SEO are transferring to success in AEO to a large degree.”
G2 treats AEO as a new swim lane, not a replacement. “We’re not abandoning SEO. We’re adding AEO.”
The core principle remains constant: “If you market from a place of empathy—that what you’re trying to do is help the right buyer find the right product and never regret the purchase, and you make that your North Star—you’re aligning with something that will probably be as enduring as Larry Page’s PageRank system.”
The Latency Warning
But Tim offers one critical warning: gaming tactics won’t last long. “You won’t be banned from all the large language models like you can be banned from Google. The bad news is the latency between finding a hole in the system and it getting closed is shortening a lot.”
With Google, content farms sometimes had two years of success before updates killed them. “It ain’t going to be no two years. It’s one model to the next because regrettable purchases are super easy to verify.”
His advice? “If you work at a company that offers a crappy service, find a new job.” Genuinely good products that help buyers will win. Everything else is borrowed time.
Empathy Wins in an Inference World
B2B buying has fundamentally transformed. Buyers share goals and pain with AI, get synthesized recommendations, then verify through trusted sources like G2. Commercial citations matter more than background citations. Calendar age no longer equals cognitive age. Keywords don’t equal prompts.
The opportunity is massive for brands that understand verification layer work. Traffic from AI is higher quality, higher conviction, and more likely to convert despite arriving earlier in the calendar timeline.
As Tim said: “People like you do this.” The fundamentals of empathy-driven marketing haven’t changed. But the mechanics of discovery, verification, and decision-making have been completely rewritten.
The brands that thrive will optimize for inference, not reference. They’ll create answer-shaped content, not message-shaped marketing. And they’ll understand AI models are tuned first and foremost to avoid regrettable purchases.
Start with empathy. Build for verification. And remember: in a world where hallucination rates matter, the best strategy is simply telling the truth.
_Voices of Search is a daily SEO and content marketing podcast hosted by_ _Jordan Keone_ _and_ _Tyson Stockton_ _. The show delivers actionable strategies and data-driven insights to help marketers navigate the ever-evolving world of search engine optimization and content marketing. New episodes air weekly, covering everything from technical SEO to AI discovery, featuring industry leaders and practitioners sharing real-world frameworks and proven tactics._
_Subscribe to Voices of Search on Apple Podcasts, Spotify, or your favorite podcast platform. Follow_ _Previsible on LinkedIn_ _for updates and subscribe to the_ _VOS YouTube channel_ _for video episodes and clips. You can also visit the official VOS site to explore the_ _full episode archive_ _and submit your SEO questions for future episodes._
Jordan Koene is the co-founder and CEO of Previsible. With a deep expertise in search engine optimization, Jordan has been instrumental in driving digital marketing strategies for various companies. His career highlights include roles in high-profile organizations like eBay and leading Searchmetrics as CEO.
Navigate the future of search with confidence
Let's chat to see if there's a good fit
Table of Contents
Navigate the future of search with confidence
Let's chat to see if there's a good fit
More from Previsible
The Executive’s Perspective on SEO
 Tyson Stockton
•
1 Sep, 2022
 Tyson Stockton
•
26 Jul, 2022
Previsible 2023: Year in Review
 Jordan Koene
•
31 Dec, 2023
Three Pillars of Scalable Enterprise SEO
 Ana Fernandez
•
14 Dec, 2024
SEO Jobs Newsletter
Join our mailing list to receive notifications of pre-vetted SEO job openings and be the first to hear about new education offerings.
"" indicates required fields
Δ
Name
This field is for validation purposes and should be left unchanged.