Skip to main content

Foundations

How does a business get recommended by AI?

A business gets recommended by AI when an assistant can find and resolve it, answer the buyer's question with current explicit facts, support those facts with credible sources, and offer a clear next step. Improve those conditions across the nine EAC signals, then re-scan the same prompt family to confirm that the assistant's answer—not merely the page—changed.

The outcome ladder

Recommendation is one outcome in a chain. The states are useful for diagnosis, but they are not a guaranteed or universal sequence for every assistant or query.

OutcomeWhat the assistant doesWhat to improve next
MentionedNames the business in an answer or local result.AI findability, profile consistency, and crawler access
CitedUses a source that supports a claim about the business.Answerability, citation confidence, and structured data
RecommendedIncludes the business as a suitable option for the buyer's prompt.Authority support, freshness, and prompt-specific facts
ActionableProvides or reaches a next step such as call, book, quote, order, or directions.Action path and a working contact or booking route

The public EAC Standard defines these outcomes and the nine signal families. A mention is not a recommendation, and a recommendation is not useful if the buyer cannot take the next step.

Diagnose → fix → re-scan

1. Diagnose the answer, not just the page

Build a small prompt family from real buyer intent: service, location, price or estimate, eligibility, open-now, comparison, and booking questions. For every check, record the exact prompt, assistant or surface, timestamp and location context where relevant, business outcome, cited source, and the suspected signal gap. The EAC Standard is the evidence checklist; the live audit supplies the current observation.

2. Fix the highest-confidence gap

Map the observation to one signal and one concrete change. Put the answer in visible, self-contained HTML before persuasive copy, keep name/address/phone/services/hours aligned across public profiles, and mirror the same facts in appropriate structured data. Use the smallest relevant playbook:

Do not treat a green technical check as proof of recommendation. The public method explains the signals and verification logic, not proprietary weights or a guaranteed ranking recipe.

3. Re-scan the same prompt family

Re-run the same questions after the change and compare the answer, cited source, outcome state, and action path. A fix is complete only when the assistant can use the improved evidence; if the answer did not change, diagnose the next limiting signal instead of assuming the page update worked.

Measure the current answer

Test the same buyer questions before and after a fix. Look for a supported recommendation and a usable next step, not just more mentions.

Run scan

Compact industry example: med spa

The med spa industry guide uses “best med spa for Botox near Scottsdale” as a high-value prompt and identifies treatment clarity, provider credentials, consultation or pricing policy, and booking action paths as priority facts.

Illustrative answer-ready block — not a live business finding: “Example Med Spa provides Botox, dermal fillers, and laser hair removal in Scottsdale. Its licensed providers and consultation policy are listed on the treatment page, with a direct consultation link for booking.”

That block gives an assistant a service, location, trust context, and next step in one attributable passage. To verify it, ask the same Scottsdale Botox question again and check whether the business is mentioned, cited, recommended for the stated intent, and connected to a working consultation path; replace every illustrative fact with the business's verified facts before publishing.

Concise Q&A

Does adding schema guarantee a recommendation?

No. Schema can help organize facts, but visible content, crawlability, corroboration, freshness, assistant behavior, and query fit still matter. The structured-data signal explains the role and limits.

Should I optimize for one assistant?

No. Test the assistants and local surfaces that matter to the buyer, using a repeatable prompt family. Record which surface produced each outcome because recommendation behavior and source selection can differ.

How do I know a fix worked?

Compare the same prompt family after the change. Look for a faithful answer, an attributable source, a suitable recommendation, and a working action path; a page edit without an answer change is not verified.

Provenance and evidence limits

This page synthesizes the public EAC Standard, Answerability, Publish citable facts, and med spa guide. The outcome definitions and workflow are AgentFound's public methodology. The med-spa passage is explicitly illustrative, not a live scan or causal claim; the standard does not publish proprietary signal weights, thresholds, or guarantees.