01

What is changing?

A vendor shortlist can begin before a prospect visits your website or speaks to sales. An assistant may help a buyer frame the problem, identify a category or compare possible suppliers. For marketing leaders, that introduces another place where positioning and proof need to hold up.

Research signal

82%

G2 reported that 82% of B2B buyers had used AI chatbots for software recommendations over the preceding two years, citing its 2026 Buyer Behavior Report in a September 22 announcement.[1]

This is reported usage over a two-year period. It does not establish that 82% start their journey in AI, that search has been replaced, or that AI recommendations caused purchases. The announcement alone does not provide enough methodology to generalize the figure to every enterprise buying segment.

I see this as a reason to investigate our own buyers, rather than a reason to declare the traditional funnel obsolete. Interviews, referral data and actual sales conversations remain essential context.

02

Why it matters to enterprise CMOs

Three implications I would investigate. These are advisory interpretations, not measured outcomes for a particular company.

01

Category language deserves a buyer check.

Your preferred category may not be the phrase a buyer uses. Test a category-led question alongside a problem-led question. If the responses differ, inspect the reasons before rewriting your narrative.

02

Claims need accessible proof.

Being named is only part of the question. Check whether an answer describes your capabilities accurately and whether the evidence is current. A customer example, documented integration or clearly scoped outcome is more useful than an unsupported superlative.

03

Website traffic is an incomplete diagnostic.

A change in visits alone cannot tell you whether AI-assisted discovery helped or hurt consideration. Review referral sources alongside enquiry quality and ask prospects how they arrived at their shortlist. Treat these as complementary signals, not a clean causal attribution model.

03

What I recommend

  1. Choose one buyer decision.

    Define the product, buyer role, segment and market. Start with questions heard in sales conversations. Keep unbranded discovery separate from questions that explicitly name your company.

  2. Capture a small, reviewable baseline.

    Run six priority discovery questions across four assistants: 24 responses. Record the exact prompt, date, model, settings, full answer and displayed sources. Use fresh conversations and document personalization or access differences.

  3. Fix what the evidence supports.

    Prioritize confirmed factual errors, unclear buyer fit or missing proof. Make a scoped change, retain the baseline and retest under comparable conditions. Do not promise that a content edit will produce a particular recommendation.

One measurable diagnostic

Count valid unbranded responses that recommend your company as a candidate, and report that count against the valid responses reviewed. Distinguish a recommendation from a passing mention. Exclude blocked or unusable runs rather than treating them as absence.

For example, 5 recommendations in 24 valid responses is a sample result, not market share, buyer preference or a forecast of pipeline. Repeats help reveal variability; they should not silently inflate the baseline denominator.

Technical access still matters. Google says its AI search features use the established SEO foundations; it does not require special AI files or markup.[2] OpenAI identifies OAI-SearchBot as its search crawler.[3] Check access through both robots rules and hosting controls.

Sources & scope

  1. G2: Buyer Intent and Customer Proof announcement · 22 September 2026. Reports the buyer-usage statistic cited above.
  2. Google Search Central: AI features and your website · Official guidance on technical eligibility and SEO foundations.
  3. OpenAI: Overview of crawlers · Official search-crawler guidance.

Sources reviewed on 23 October 2026. The implications and recommendations are my interpretation. The 5-in-24 example is illustrative; no company audit results are presented here.