Unbranded Buyer Prompts: Why 65% of AI Discovery Happens Before Anyone Searches Your Brand Name

The pain point: you win branded search, you lose category discovery

Your Search Console shows strong clicks for `[your brand] pricing` and `[your brand] vs competitor`. Leadership reads that as “AI will find us too.” Answer engines build shortlists from different prompts — mostly without your brand in the text.

The Foundation × AirOps Hidden Selection Phase report (Dec 2025–Feb 2026) tested 100 prompts per brand across 50 B2B brands: 65% unbranded discovery (“best CRM for startups”) and 35% branded validation (“is [Vendor] good for…”). On branded prompts, brand-owned content appeared in 77.6% of responses — but on category-level unbranded prompts, only 2.2% of citations pointed to brand-owned domains.

Shareable stat: “2 out of 3 AI buyer prompts never include our name — and on those category prompts, less than 3% of citations go to our site.”

Metric 1 — Mention rate (are you on the shortlist at all?)

Mention rate = share of buyer-intent prompts where the answer engine names your brand in prose. It is distinct from citation rate (link to your domain) — you can be mentioned without being cited (citation gap study).

In multi-brand Foundation × AirOps tracking across 50 B2B brands, mention and citation moved independently week to week — a brand could gain mentions on unbranded prompts while losing citations when a new listicle replaced their help doc as the sourced URL.

Shareable metric pair for dashboards: track mention rate and citation rate on the same fixed prompt library — not aggregate “AI traffic.”

Metric 2 — Category prompt win rate (DerivateX controlled test design)

The DerivateX B2B SaaS AI Citation Study (May 2026) used one buyer-style question per category across 40 categories, 10 fresh sessions each, with web search enabled. 219 distinct tools appeared in answers — but distribution was heavy-tailed: a small set of incumbents dominated unbranded shortlists.

Documented behavior: ChatGPT attached citations to 92.3% of named tools — buyers receive “proof links” even when those links rarely point to the vendor’s site (11.6% vendor-owned citation share in the same study).

Pain point for challengers: you can enter the named shortlist on category prompts while every citation goes to a 2026 listicle that features incumbents — your marketing site never enters the evidence chain.

Shareable link: DerivateX full report PDF

Real-world pattern: the “invisible incumbent” vs “visible challenger”

These patterns recur across public case write-ups and the 2026 samples above — not hypotheticals.

• Visible incumbent: ranks #1–3 on unbranded category prompts *and* owns a citable comparison page (table + FAQ + year in title) — maintains mention and citation
• Invisible incumbent: strong G2 ratings and branded SEO — absent from “best [category]” AI answers because product facts live in PDFs and uncrawlable app shells
• Visible challenger: earns unbranded mentions after niche listicle inclusion — citation lags mention by 4–8 weeks in observed prompt libraries
• Citation-only ghost: blog ranks, brand not named in prose — AI cites the article but recommends a competitor by name (common on Reddit-sourced answers)

What to publish so unbranded prompts can mention you

Pages that win unbranded citations in DerivateX’s retrieval analysis shared structure — not prose quality alone:

Instrument unbranded prompts (minimum viable program)

1. Draft 20 category prompts — zero brand terms (“best…”, “alternatives to…”, “[category] for [use case]”). 2. Run them weekly on the engines your buyers use (at minimum: one general assistant + one search-grounded engine). 3. Log full answer text — not just yes/no. 4. Split metrics — mention vs citation vs competitor co-mentions.

AppScan AI tracks a prompt library with scheduled runs, mention rate, citation rate, and share of voice vs named competitors. Start with the free AI answer check to see verbatim answers on inferred buyer prompts for your domain, then expand to a fixed library in the dashboard.