Right now, 45% of your B2B buyers are using AI tools during their purchase research (Gartner via DigitalCommerce360, Mar 2026). Some of them are asking ChatGPT, Claude, or Perplexity about you. Some of them are getting confidently wrong answers back.
Whether to do something about that this quarter is the question. We sell AI brand visibility infrastructure. So we're biased toward "yes." But there are real cases where the answer is "not yet," and signing up the wrong founder costs us more than skipping them. This is the version of the conversation we have on intro calls.
Three cases where the answer is no. Three signals that mean yes. Skip to whichever you need.
For the upstream framing, see the GEO field guide. For the audit protocol we use to stress-test brand accuracy on three frontier models, see the five-prompt wince test.
Why is a GEO vendor publishing this?
We built Ooky because we kept watching B2B brands get misattributed by ChatGPT, Claude, and Perplexity in ways that cost them deals. That's the bias to keep in mind. It's also the bias that funds this post, so we have an interest in being honest: there are common B2B shapes where AI brand visibility is not yet your best-payback spend, and pretending otherwise wastes everyone's time.
We'd rather show you the decision and lose you for a year than land a customer who churns in a quarter. The research below is third-party (Gartner, Vectara, Seer Interactive), the three don't-buy cases are the same ones we walk founders through on intro calls, and the three yes signals are what we look for ourselves before we tell someone to start.
What does the research say about AI in B2B buying?
Three numbers anchor the decision. The first sizes the surface. The second sizes the accuracy problem. The third sizes the upside of getting it right.
1. Surface: 45% of B2B buyers used AI during a recent purchase
Gartner's 2025 sales practice survey found 45% of B2B buyers used AI tools at some point in a recent purchase journey, with n=646 (Gartner via DigitalCommerce360, Mar 2026). Nearly half of B2B buying motions now touch an LLM somewhere in the research cycle, which is the surface you're either showing up on or not.
2. Accuracy: frontier models exceed 10% hallucination on grounded summarization
Vectara's Hallucination Leaderboard refresh found every frontier reasoning model exceeded 10% hallucination on grounded summarization, with Gemini-3-Pro at 13.6% (Vectara Hallucination Leaderboard, Nov 2025). Long-tail entities (most B2B brands under 200 employees) tend to fare worse than the leaderboard average, per the broader hallucination literature. The question is not whether your brand gets misattributed. It's how often and how badly.
3. Upside: AI-referred traffic converted 3 to 9x organic on one Seer client
Seer Interactive measured a single client (n=1,370 AI conversions, Oct 2024 to Apr 2025): ChatGPT-sourced traffic converted at 15.9%, Perplexity at 10.5%, Claude at 5%, Gemini at 3%, versus Google organic at 1.76% (Seer Interactive, Jun 2025). One client, vertical undisclosed, so the absolute numbers shouldn't be treated as a benchmark. The order of magnitude (a 3 to 9x lift) is what's worth carrying forward.
Three honest data points. None of them tell you whether to act. That's the next two sections.
Three cases where you can wait
You can wait on this if you're in any of the three cases below. The research above only works if you have an inbound motion. Take that away and the upside disappears.
- Pre-product or zero-inbound. No discovery as a pipeline source means no AI-discovery either. Spend on demand-gen first. The five-prompt wince test still applies if you want to see what AI says about you today. It's free either way: 20 minutes manually, or zero time on Ooky's free tier, which puts it on a schedule and trends the visibility score over time. A paid layer on top of zero inbound is still premature.
- Pure-outbound, named-account motion. SDR-driven, tight ICP, prospects do not search for you because you find them. AI brand visibility matters less than a clean dialer and a sharper sequence. The brand-accuracy axis still applies (your prospects will check ChatGPT mid-evaluation), but the referral-capture upside is structurally zero.
- PLG with strong direct or branded traffic. Buyers already know your URL, type it, and the AI conversation happens after they've found you. Brand-accuracy still matters for the comparison-shopping window, referral-capture less so. Revisit when your free-to-paid funnel includes meaningful AI-sourced traffic.
One caveat across all three: brand mention accuracy still applies. Even if the referral-capture half of the value is zero today, your competitors will eventually optimize, and the gap between what AI says about you and what AI says about them widens over time. The honest answer there is "the wince test still matters, just don't pay for the rest yet."
Three signals it's worth it
Three signals to check, in order. Hit all three and the answer is yes, this quarter. Hit two and it depends on which two. Hit one or none and see the three cases above.
1. Inbound is more than 20% of closed-won
Pull your last 90 days of closed-won deals and tag the source. If discovery (inbound search, content, word of mouth) is north of 20%, AI brand visibility compounds on a channel you already use. Below 20%, you're optimizing a channel that doesn't yet move your number.
2. Your ACV is above $5K
Higher ACV means each misattributed buyer costs more, and the math gets faster to pay back. Below $5K, you need either very high volume or a particularly research-heavy ICP to make the time investment worthwhile. We use $5K as a heuristic threshold from intro calls, not a published benchmark — adjust to your unit economics.
3. Your buyers research for weeks before contacting you
Long evaluation cycles mean buyers spend more time on AI tools during research, and the 45% adoption figure from Gartner concentrates in those long cycles. If your motion is impulse-led or rep-led from cold contact, AI is less load-bearing. If buyers compare you to three competitors over a quarter, AI is doing a lot of the comparison.
Three for three is a strong yes. Two for three with inbound and research-heavy is also a yes, because the ACV bar is the easiest to clear over time. Two for three without inbound is closer to a no, because the referral-capture value still depends on that channel existing in the first place.
This pattern hits hardest for local-discovery businesses where buyers search AI for "best X near me." Private medical clinics, dental practices, legal firms, accounting practices. All three signals tend to fire together. Research-heavy buyers. ACV that clears the bar without effort. Inbound that's most of the pipeline. If you're one of these, AI brand visibility is one of the highest-return moves you can put on next quarter's list.
If you want to verify the assumptions before deciding, the five-prompt wince test shows you exactly what AI says about your brand today. It's free either way: 20 minutes manually, or zero time on Ooky's free tier. If you'd rather understand the retrieval mechanic that determines those answers, see what do_not_infer does. For the broader category framing, see the GEO field guide.
Sources
- DigitalCommerce360, citing Gartner. "B2B buyers, sales rep-free purchasing, AI reshapes sales." March 17, 2026 (survey fielded August-September 2025). Retrieved 2026-06-08. https://www.digitalcommerce360.com/2026/03/17/gartner-b2b-buyers-rep-free-purchasing-ai-reshapes-sales/
- Vectara. "Introducing the next generation of Vectara's Hallucination Leaderboard." November 19, 2025. Retrieved 2026-06-08. https://www.vectara.com/blog/introducing-the-next-generation-of-vectaras-hallucination-leaderboard
- Seer Interactive. "Case Study: 6 Learnings, 1 site - How Traffic from ChatGPT Converts." June 3, 2025. Single client, vertical undisclosed, n=1,370 AI conversions, October 1, 2024 to April 30, 2025. Retrieved 2026-06-08. https://www.seerinteractive.com/insights/case-study-6-learnings-about-how-traffic-from-chatgpt-converts