The COAST Framework
What COAST stands for.
COAST stands for Comprehension, Occurrence, Accuracy, Shortlist Position, and Traceability — five lenses on whether AI understands you, finds you, gets facts right, shortlists you, and can point to supporting public evidence.
The COAST Framework measures what ChatGPT, Gemini, Perplexity, and Google AI Overviews (core), with other relevant surfaces added when agreed say when buyers ask who to hire — then improves what those answers depend on and retests to prove what changed. In a recent engagement, a healthcare practice with 26 years in business and 200+ Google reviews went from being overlooked by AI to winning new clients who said AI sent them.
Asking “What is [Business]?” tests recognition. Asking “Who are the best [category] providers in [location]?” tests discovery. A business can be easy to describe by name and still get skipped when buyers ask by category.
C
C — Comprehension
Does the system correctly understand the entity, services, audience, geography, and differentiators?
Example: when a buyer asks for a family law firm in Bradenton, does AI correctly place you in that category and market — or confuse you with a different practice type or city?
O
O — Occurrence
How often does the business appear across repeated, commercially relevant prompt families and platforms?
Example: across ChatGPT, Gemini, Perplexity, and Google AI Overviews, how often do you appear when buyers ask category questions — once, often, or almost never?
A
A — Accuracy
Are material facts correct, current, specific, and internally consistent?
Example: does the answer get your services, location, and key facts right — or invent an outdated specialty, wrong area, or details you no longer offer?
S
S — Shortlist Position
Is the business absent, merely named, included in the recommendation set, or prioritized with a rationale?
Example: when AI names options, are you the prioritized pick with a reason, a late mention, or missing while competitors take the shortlist?
T
T — Traceability
What visible sources and public evidence support the answer, and how stable are those source patterns?
Example: which public sources is the answer leaning on — your site, listings, reviews — and do those sources actually support recommending you?
Score scales, dimension weights, and prompt packs stay proprietary. You receive inspectable findings — prompts, platforms, dates, and answer excerpts — not a black-box formula.
Measure the answers buyers hear, improve the public evidence those systems read, then re-ask comparable questions to prove what changed. Exact packs, thresholds, and fix order are delivered in client work—not as a public syllabus.