AI-Visibility Audit — Texas de Brazil, Greenville

A live, public audit of how a real Greenville restaurant shows up to ChatGPT, Claude, Perplexity, Gemini, and Copilot — and the eight specific fixes that would close the gap. Same methodology AvoTech runs for paying clients.

Why this audit, on this page

The five generative search engines — ChatGPT, Claude, Perplexity, Gemini, and Microsoft Copilot — answer "best Brazilian steakhouse in Greenville" by stitching together structured data, extractable text, and third-party citations. A page that ranks well for traditional Google may still be invisible to those engines if the structured signals are missing.

Texas de Brazil's Greenville location page is a useful subject because the parent brand is well-known (the engines have it cached in training data) but the local-page signals are mostly inherited from the corporate template. That mismatch is where the wins live.

Headline scorecard (12 dimensions)

Verdict: 7 wins, 3 hard misses, 2 fix-mes. The location page does the basics right but leaves the AI-search wins on the table.

The eight specific fixes

1. Add FAQPage schema with the questions diners actually ask

The corporate site has parking guidance, dress code notes, and reservation policy somewhere — but none of it is in FAQPage JSON-LD. Adding a single block that answers the five questions a typical diner Googles ("Is there parking at Haywood?", "Dress code?", "Vegetarian options?", "Kid-friendly?", "Reservation policy?") gives ChatGPT and Perplexity exact text to cite.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {"@type": "Question", "name": "Is parking available?",
     "acceptedAnswer": {"@type": "Answer", "text": "Free parking ..."}}
  ]
}

2. Add Menu / hasMenu schema (or link the existing PDF)

If a guest asks ChatGPT "what's on the menu at Texas de Brazil Greenville?", the engine has nothing structured to cite. A hasMenu reference on the Restaurant schema, even just a link to the existing menu PDF, is the minimum-viable fix. A full Menu + MenuSection + MenuItem structure earns rich answers.

3. Replace generic og:image with a Greenville-specific photo

Current og:image is /wp-content/uploads/2021/04/homepage-mobile.jpg.png — the corporate homepage shot. Swap to an exterior or interior photo of the Greenville location specifically. Engines that pull og:image for visual answers (Perplexity, Bing chat) will get a Greenville shot, not Dallas.

4. Put the questions people ask on the page

Answer the questions people actually ask, in plain HTML on the location page: the address, the hours, parking, reservations and the price of the full churrasco. Answer engines quote visible questions and answers; markup alone is not enough.

5. Publish /llms.txt and /ai.txt

Both return 404. /llms.txt is becoming convention for guiding AI crawlers ("here's the high-signal stuff: menu, hours, history, our 30-year story"). /ai.txt is the consent-and-attribution layer. Adding both is a 10-minute upload that signals AI-readiness to crawlers and to anyone who checks.

6. Add hreflang for Spanish + Portuguese

A Brazilian steakhouse in a US market with growing Hispanic and Brazilian-expat populations is leaving signal on the table. Even a Portuguese landing page for the Greenville location with proper hreflang tags improves Perplexity / Gemini matches for non-English queries.

7. Add a Q&A "voice-search friendly" intro paragraph

Generative engines reward pages that answer in the page body, not just in schema. A 2-3 sentence opening that explicitly says "Texas de Brazil in Greenville is a Brazilian steakhouse on Haywood Road, serving rodízio churrasco from 4 PM…" gives the engines a clean extraction target. The current page hides this in nav and CTAs.

8. Add YouTube video schema linkage

The og:video tag is missing. If the location ever publishes a YouTube tour or chef interview, marking it via VideoObject + linking from the Restaurant schema gives the engines a multimedia surface. Free win once the asset exists.

What's intentionally NOT recommended

This audit deliberately doesn't touch:

• Site speed / Core Web Vitals — the corporate WordPress is fine; engine ranking is the binding constraint, not pixel-paint.
• Backlink-building — out of scope for a location-page audit.
• Google Business Profile — separate workstream; partially covered by the existing schema but worth a follow-up.
• "Move to a SPA" — would actively hurt extraction.

What this case study demonstrates

An AI-visibility audit isn't a list of generic SEO advice. It's a specific, citation-aware diagnosis of how five named engines extract from a specific URL — and a costed list of fixes a small team can ship in a week. This whole writeup took 30 minutes of public-data inspection. A paid audit goes deeper: live engine queries against the brand, competitor schema benchmarking, and a prioritized 6-week implementation plan.