Anonymized North American B2B PR / communications consultancy (SMB). This page documents an Assess engagement on locked buyer prompts—not a named client marketing claim. Improve → Review retest numbers stay Pending until a real retest ships. Method boundaries: FAQ.
Updated 2026-08 · Label: anonymized / illustrative Assess pattern
Situation
A North American B2B PR consultancy (anonymized) sells retainer and project work to mid-market brands. Buyers increasingly ask AI assistants which firm to hire for earned media and message discipline. The public site stated services in agency-generic language, but did not make entity facts, include/exclude boundaries, or “who we are for” scannable—so locked buying-intent prompts about hiring a PR / communications partner showed weak or inconsistent brand presence versus better-structured peers.
What we did
Under Assess (first layer of AEO/GEO Engagement), MentionBench delivered a readiness package the client could act on without buying a monitoring SaaS seat or an unlimited content retainer. Cross-check: How we work.
- Site readiness read: existence / structure / cross-platform consistency signals on the public pages that matter for buyer prompts
- Competitor mention–style gap table on a small locked English prompt panel (buying and “who to hire” intent)—one-time contrast, not an always-on dashboard
- Prioritized fix list ordered by leverage (what to change first, why, and what “done” looks like)
- Froze the locked prompt panel + baseline mention counts as the formal Before for a later Improve → Review window
We did not expose internal tooling, sell a product seat, or promise rank / share-of-voice / permanent answer ownership.
Verifiable result
After Assess (shipped): three client-facing artifacts—readiness conclusion, competitor gap table, prioritized fix list—plus a frozen locked-prompt baseline recorded as X/N on the agreed panel. The buyer can state what to change first and how acceptance will look.
After Improve → Review (pending): the same locked panel will be retested and published as “Prompt set: N. Before (post–Assess): mentioned on X/N. After (post–Improve): Y/N.” Until that retest completes, AI shortlist lift stays Pending—no invented percentages, no vanity traffic substitutes.
Evidence chain: this page · Case Studies · FAQ · self-case pattern for comparison: MentionBench Self.
What this proves
Assess is a proof loop entry: lock prompts → ship an interpretable readiness package → freeze Before → only then claim After. A second vertical (PR consultancy SMB) shows the same skeleton as the self case without inventing client outcomes.