We didn't learn this on someone else's restaurant. We built it on our own.

The restaurant

Wok & Karahi — "Home of the Crispy Beef" — is one of the very few halal kitchens serving Chinese and Indian/Pakistani under one roof, at 3422 FM 2920 Rd Unit #120, Spring, TX. It's owned by Abbas Zoeb, an ex-Google software engineer, who runs it as AZ's proving ground. It's highly rated in its category and was named to You Had Me At Halal's Top 10 Halal Restaurants of 2024. In other words: a real, busy, well-loved independent — not a lab.

The problems we found on our own restaurant

When we pointed AZ's own methods at our place, we found the same leaks that quietly drain most independents:

• The website was invisible to AI. The old site returned HTTP 403 to bots — so ChatGPT, Gemini, Perplexity and Claude literally couldn't read the menu, hours, or address. In the "ask an AI where to eat" era, that's being absent from the fastest-growing front door.
• AI thought we were two different restaurants. Our business was showing up under two separate names/identities to AI assistants — splitting our reputation and confusing every answer about who and where we are.
• Margin leaking to the 30% apps. Third-party delivery apps take ~30% plus fees and own the customer. Direct orders ran about 8.8% of the mix versus 20%+ on the 30%-fee apps — a wide-open recovery opportunity.
• Money left on the table at the counter. Guests who order from both cuisines spend a +63% higher ticket — but nothing was steering them there.

What we built — and what actually changed

1. A site AI assistants can finally read

We rebuilt the site in code as an AEO-native site (wokandkarahitexas.com): full Restaurant / Menu / FAQ schema, a menu statically rendered so no-JS crawlers see every dish, local landing pages, llms.txt, and a clean sitemap. The 403 problem is gone — the menu is now machine-readable, so when someone asks an AI where to eat, we're in the answer. (Verifiable live today.)

2. One business, one name — the entity fix

We consolidated the restaurant into a single, consistent identity across the site and its structured data, so AI assistants stopped treating us as two businesses and started describing one restaurant, correctly. This is the same "unified identity" discipline we bring to every client — get the machines to agree on who you are before you fight for how you rank.

3. The direct-order push — already running, 1+ year

Every order that leaves the kitchen carries a thank-you note, customized by order type. Third-party-app customers get a card that reads "Paying up to 50% extra?? Order on our website, save big" — because the apps mark our prices up and pile on fees, while direct means delivery within 5 miles for $7.99, no extra fees, cheaper prices, and more deals. A quiet, low-tech machine that converts the app's own customers into direct ones. The opportunity is the gap between that ~8.8% direct today and the 20%+ sitting on the 30%-fee apps.

4. Steering guests to both cuisines

Because a guest who orders both Chinese and Indian/Pakistani spends a +63% higher ticket, our menu, site, and phone agent all nudge toward the pairing — turning our single biggest differentiator into a higher average check.

5. A review engine that only lets the good ones through — 1+ year

Two tracks: a "Scan me" QR card is brought to the table only when a guest is clearly happy (and reminded of the experience, so they actually scan); an unhappy guest is handed the owner's business card to raise it directly with the owner — before a negative review is ever posted. The result: Google reviews roughly doubled in a year (~400 → 872), now 4.6★, with new ones running five-star.

6. Tablet consolidation — 1+ year live

Third-party delivery orders now flow straight into the Clover POS and print to the kitchen automatically. Before, staff had to accept each order on a separate delivery tablet and re-type it into the POS to make a ticket — which, in a rush, meant lost, late, and wrong orders. That entire manual step is gone.

7. An AI phone agent that actually works

Answers every call in a locked, human-sounding voice (English + Urdu/Hindi), takes orders straight into the POS, sizes catering, upsells like a friend, remembers regulars, and handles complaints with real empathy. Built on 28 integrated tools, gated so it can never place a bad order, and put through a 15-scenario QA gate. It's real-write capable and built — you can hear it live on the restaurant's own line, (281) 362-5354.

The results

Timing note: the rebuilt site went live 2026-06-28 and the phone agent shortly before, so a clean "before/after revenue" window on those two doesn't exist yet. The tactics that have run a year+ (reviews, thank-you notes, tablet consolidation) already have real results. Here's the honest scoreboard:

• 4.6★ across 872 Google reviews — roughly double the ~400 a year ago, running five-star since the review system went in. (Verifiable live.)
• 403-to-bots → fully AI-readable site. (Verifiable now.)
• Two identities → one consolidated business entity for AI assistants.
• Tablet consolidation eliminated manual re-entry and the lost/late/wrong orders that came with it — 1+ year live.
• +63% higher ticket when guests order both cuisines — now actively steered.
• ~8.8% direct vs 20%+ on the 30%-fee apps — the margin-recovery opportunity, in the open.

And the numbers only the owner can put a hard figure on, once the newer tactics have run long enough:

• Third-party commission recovered: [OWNER: $ / month, from the Clover fee model]
• Direct-vs-3P order-mix shift after the push: [OWNER: baseline ~8.8% → current %]
• After-hours / missed-call demand captured by the phone agent: [OWNER: # calls captured, est. $ once tracked]

Frequently asked questions