B2B Buying Journeys When the First Touch Is an LLM

By ScaleGrowth Digital Editorial

The B2B buying journey used to begin with a Google search. For a growing share of enterprise software, fintech, and professional services categories, it now begins with a question typed into ChatGPT, Claude, or Google AI Mode. The buyer asks “what are the leading tools for X” or “which vendor does Y for a 200 person company,” reads the named brands, forms a shortlist in 90 seconds, and never visits the SERP. This is a real, observable shift, and it changes which marketing assets do which job. This piece sets out what changes in the buyer journey, what the new vendor-shortlist mechanics look like, and which structural moves keep a brand in the answer.

The new shortlist mechanics

Three things happen inside an LLM shortlist that did not happen on a SERP. First, the model produces a finite list (typically three to seven vendors) with no “see more results” path. The buyer’s exposure is capped at the named brands. Second, the model produces a one to two sentence description of each vendor, drawn from sources the brand may or may not control. Third, the order is rarely random. Models tend to lead with the brand that has the densest factual support across the indexed sources, not the brand with the highest paid-search bid.

The cumulative effect: a brand that is not named in the shortlist is invisible to a meaningful share of the addressable demand, and a brand named with a thin or misleading one-liner is named badly. Both are pipeline problems. Neither shows up in a standard SEO report. The 8 percent ChatGPT mention rate observed on a major BFSI lender, 15.6 percent on AI Overviews, and 19 percent on AI Mode is the diagnostic shape of this problem. Strong classical authority. Soft generative presence.

How the buyer journey actually looks now

Across the engagements ScaleGrowth Digital has run, a recognisable pattern has emerged for B2B buyers in the 50 to 5,000 employee band.

Stage one. Frame the problem inside an LLM. Buyer types a paragraph describing the situation. Model returns a problem framing and a vendor shortlist. The buyer treats this as orientation, not commitment.

Stage two. Cross-check the shortlist against one or two sources. A specific G2 page, a specific industry blog, a former colleague. The cross-check usually confirms or eliminates one to two names from the model’s list.

Stage three. Land directly on the surviving vendors’ deep pages. Often the pricing page, the integration page, or a comparison page. Not the homepage. The visit shows up in GA4 as direct traffic to a deep URL, with no obvious upstream attribution. See attribution modeling when LLM traffic is untrackable for the measurement implications.

Stage four. Branded search and competitive comparison. Two to four weeks later, the buyer searches the brand by name and the brand plus a competitor. The branded search lift is downstream of the LLM citation that started the process.

Stage five. Demo request or sales touch. The buyer fills the form. Source data: “direct” or “google / organic” (branded). The actual first touch (the LLM mention) appears nowhere in the CRM.

Which assets do which job now

Evidence from a multi-LOB BFSI RFP

A wealth platform across loans, investments, insurance, payments, and financial tools ran an enterprise RFP. The brand had 71,000 organic ranking keywords across 13,600 pages, with 17,200 gap keywords against competitors. A 150 prompt AI visibility test surfaced a 40 percent mention rate on the 50 prompts tested. The brand was named in shortlists roughly two times out of five. Not invisible. Not dominant.

The page-level work that came out of the engagement targeted the stages where the funnel was leaking. A 49,500 monthly search volume head term for “business loan” sat at position 26, against a 180,000 keyword gap to the category leader. The LLM shortlist on equivalent prompts named the brand inconsistently. The corrected output recommended six existing pages for ANALYZE (entity records, schema consistency, content depth) and four new pages for CREATE (a head-term pillar, a comparison page, a pricing-transparency page, a compliance overview). Total: 27,818 lines of JSON-defined content brief, scope-aligned to the LOB subdomain boundaries the brand actually operated within. The structural moves were calibrated to the stages where LLM-mediated buyers were dropping off, not to a rank-position uplift target.

What B2B brands need to commission this quarter

Three commissioned outputs cover the LLM-first B2B journey at the structural level.

Commission one: a category prompt scan. 100 to 300 prompts that the addressable buyer might plausibly type into an LLM. Run them across ChatGPT, Claude, Google AI Mode, and Perplexity. Log mention rate, shortlist position, and the one-liner each model used. Cache the raw JSON. The 300 prompt protocol is the same one used on the BFSI engagement above.

Commission two: an entity audit. Knowledge graph, Wikipedia / Wikidata where applicable, Crunchbase, LinkedIn, G2, Capterra. Inconsistent founder names, year-founded discrepancies, mismatched HQ addresses, and outdated employee counts directly weaken the model’s confidence in the brand. The lender audit found 224 invalid structured-data items and 81 percent of pages without canonicals. Same class of issue, different category.

Commission three: a comparison-page sweep. Three to five competitor-comparison pages, written honestly, schema-marked, and internally linked. These are the assets the LLM most often cites when synthesising vendor shortlists. See the SEO audit service for how this fits the larger sprint.

Practitioner takeaway: five actions for next week

1. Run a 100 prompt scan against your category. Log mention rate, shortlist order, one-liner content per model.
2. Pull deep-page direct traffic. 90 days, segmented by URL depth. Compare to last year same period. The delta is the LLM-influenced surface.
3. Audit your G2 / Capterra / equivalent. The models read these. Outdated information here cascades into a thin shortlist one-liner.
4. Build the three missing comparison pages. Top three competitor matchups. Honest, schema-valid, linked from the pillar.
5. Cache the raw model responses. Before-and-after defensibility depends on the cache. The GEO playbook covers the operational version.

FAQ

Does this apply to brands with under 500 monthly LLM-attributable visits?

Yes, with caveats. The 500 visit floor is a measurement floor, not a strategic one. Even brands with small LLM-attributable traffic have buyers using LLMs as a first touch. The mention rate scan is the more reliable diagnostic than the analytics floor, because the click is missing by definition.

How does this differ from classical SEO for B2B?

Classical SEO optimises for SERP position. LLM-first journey work optimises for inclusion in a shortlist of three to seven brands and for the one-liner attached to the brand. The disciplines overlap (schema, entity clarity, deep page health) but the success metric is different. Position 1 on Google does not guarantee being named in the LLM shortlist. Sometimes it correlates, sometimes it does not.

Is paid search still useful for B2B in this environment?

Yes, with reduced strategic weight. Paid covers branded-defence (so a competitor cannot bid on your brand cheaply) and covers head terms where organic substitution will take more than two quarters. Beyond those two roles, paid in B2B is increasingly catching demand that the LLM-first journey has already qualified.

What’s the lead-time on closing the mention-rate gap?

Three to nine months, depending on category competitiveness and starting baseline. Entity and schema fixes can move mention rate within a quarter. Comparison-page and review-platform work tends to compound over two to three quarters. The cleanest test is the before-and-after on the same prompt cohort.

Get the scan

If your B2B pipeline is growing more “direct” each quarter and you cannot point to the campaigns that earned it, the LLM-mediated journey is probably the missing channel. Commission an AI visibility audit, and the 300 prompt baseline plus shortlist analysis ships in the first sprint.