Fifty-one percent of B2B software buyers now start vendor research with an AI chatbot rather than Google, according to a G2 survey of 1,076 decision-makers across North America, EMEA, and APAC. Consideration is forming inside a conversation your attribution model can't see, your SDRs can't interrupt, and your MAP won't log. By the time an intent spike or MQL shows up, the shortlist may already be locked.
Circle & Square founder and CEO Neeta Goplani made this the centerpiece of her B2BMX Insights Q&A, published on Demand Gen Report. Her argument is blunt: AI visibility is not a pipeline strategy. Showing up in chatbot answers is a precondition, not a result. And if you're measuring the wrong prompts at the wrong stage, you end up with what Goplani calls "a false sense of relief," which is worse than having no data at all.
The Visibility Gap Ops Teams Should Actually Worry About
Traditional demand gen measurement starts when a buyer does something trackable: visits a page, downloads a PDF, fills out a form. Goplani's point is that AI-assisted discovery happens upstream of all those signals. A buyer asks a chatbot to explain a problem, compare approaches, and name vendors worth evaluating. That entire sequence can happen without generating a single first-party event in your systems.
G2's data reinforces the scale. Eighty-five percent of respondents said they think more highly of a vendor when an AI chatbot includes it in an answer, and one in three purchased from a vendor they hadn't previously heard of. The more conservative interpretation (per MarketSplash's secondary analysis) pegs the "discovered a previously unconsidered vendor" figure at 33%, not the headline 69% that combines multiple response categories. Either way, the directional signal is the same: consideration is shifting before any trackable demand gen event fires.
Measuring the Wrong Prompts Creates False Confidence
Goplani describes a client that looked healthy on paper. The brand appeared when someone asked for a specific vendor recommendation. But when her team examined the earlier buyer conversations ("What is this problem? Why is this happening? How should I solve it?"), the client was barely present. Competitors were everywhere.
This is the operational risk: AEO and AI-visibility tools can return positive scores if you point them at the wrong questions. You check whether you show up for "best [category] software" and feel good. Meanwhile, the buyer's actual decision-shaping query was something like "why does [problem] keep happening in [context]," and your brand is absent from that answer entirely. Circle & Square's approach starts with what Goplani calls "Buyer Zones," mapped from forums, Reddit, review sites, search behavior, sales conversations, and customer feedback. The goal is to identify which questions actually influence decisions, then measure visibility against those, not the ones that feel good to track.
Evidence Quality, Not Just Presence
Being mentioned is table stakes. A Search Engine Journal analysis examining 1,855 citations across 160 AI answers found that 73.7% of responses had weak or missing evidence for important claims, even when the vendor appeared in nearly every answer. Visibility got them on the list; evidence quality determined whether they stayed.
Digital Journal reported that 75% of source pages cited in AI vendor recommendations were published by vendors selling in the category, and 55% ranked or included their own products. AI answers are, in many cases, recycling vendor-authored content back to buyers as if it were independent guidance. If your comparison pages lack third-party validation or verifiable claims, a criteria-based follow-up prompt may filter you out.
What This Means for Your Measurement Stack
Goplani's framing has a practical implication for ops teams: you need to separate three distinct things that often get conflated. Being mentioned in an AI answer (visibility). Being recommended with supporting evidence (credibility). Creating pipeline that connects to revenue (impact). Most AEO dashboards measure the first. Some attempt the second. Almost none connect to the third.
The hypothesis worth testing: if you optimize content for the problem-framing questions buyers ask before they ever search for a vendor name, does that shift the composition of inbound pipeline toward earlier-stage, less competitive opportunities? You can't instrument a chatbot session you don't own, but you can run a holdout on content changes and watch for shifts in source-of-first-contact and time-to-opportunity over a 90-day window.
The trade-off is real: investing in pre-consideration content means pulling budget or effort from bottom-of-funnel campaigns that produce trackable, attributable pipeline today. Goplani's argument is that by the time those signals arrive, the shortlist is already formed. The pipeline you're measuring may be the pipeline you were always going to win (or lose), and the actual leverage point is upstream, in a place your current stack can't see. Measuring the visible part of the funnel and calling it complete is its own kind of risk, the kind that shows up as a slow, unexplained decline in win rate while your visibility scores stay green.
