The Direct Index
Report · Pretoria · scanned 2026-11-14

Kelly's Indian Meals

kellysindianmeals.co.za is hard for agents to use.

Listed by Overture Maps.

MenuCallWebsite

What agents are told

Assistants

What assistants answer

Each assistant gets three questions about this business: the official website, where to book and where to order. Each cell shows where the answer sent people.

The assistants haven't been asked about this business by name yet. They're asked about a sample of each restaurant index and about every claimed business; what they answer is shown here, beside the score, never in it.

Named in none of the 66 answers to 22 everyday questions about Pretoria.

Named instead

In the answers that didn't name Kelly's Indian Meals, the places named most often, and in how many answers.

See who was named →

Owners

The website's grade

An AI assistant can open Kelly's Indian Meals's website, though its robots.txt turns some of the assistants' crawlers away. It can read what Kelly's Indian Meals offers. To book or order, it is sent to Uber Eats, a marketplace.

44.2%

24.3 of 55 points available

1/5

agent-readiness level, Basic Web Presence

4

things to fix

Scored on what the scan measures for every site, the same way for everyone. The action test and what the assistants answer are shown beside the score, never in it (specification 1.1, 18 November 2026).

Who can read the site, by kind of agent
Kind of agentWhat the scan sawIn the score
Search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot)blocked by robots.txt: PerplexityBot10 points, the two rows together
Fetchers that read a page for a person (ChatGPT-User, Claude-User, Perplexity-User)let in by robots.txt10 points, the two rows together
Training crawlersblocked by robots.txt: GPTBot, Google-Extended, anthropic-ai, CCBot, Bytespider, cohere-ainot scored
Requests under crawler names from Cloudflare's networkgot the pagenot scored
The scanner's own signed requestgot the page2 points: the text without JavaScript

See what an AI agent reads on this site →

Where it stands

How Kelly's Indian Meals compares

Where it stands

214 of 377 in Pretoria

Ahead of 43% of the other sites scanned in this index.

What the top 10% of the index doesTop 10%This site
Take bookings or orders on their own domain100%no
Publish the menu or price list as HTML92%yes
Let the assistants in (robots.txt)97%no
Structured data with address, phone and hours18%no
Have an llms.txt36%no

Compared with its peers

Average scoreA or BThis site
Sites using Uber Eats 2,750 sites62%33%44.2%
Businesses like yours

Businesses like Kelly's Indian Meals in Pretoria

GroupSitesAvg scoreA or BThis site
Businesses of the same kind in Pretoria37753%27%44.2%
Sites using Uber Eats in Pretoria662%33%44.2%

After the top fixes

With the three fixes above (No structured data for the business; Ordering only through Uber Eats; robots.txt blocks PerplexityBot), the score would be about 95%. Against today's scores in Pretoria, that would rank about 9th of 377 (it is 214th today). The others' scores move too, so this is an estimate.

Findings

What to fix first

  1. No structured data for the business +14 points no_structured

    Add schema.org JSON-LD for the business (Restaurant, LocalBusiness or the closest type, with name, address, telephone and openingHoursSpecification, plus servesCuisine and acceptsReservations). Every assistant reads this block first, and the fix kit below has it filled in. How to fix it

    Passes when the home page has schema.org JSON-LD that names the business and its type. Recheck it

  2. Ordering only through Uber Eats +12 points ordering_marketplace

    Every transaction an agent makes goes through the marketplace, which takes its commission. Add a direct way on your own site. Even a simple form will do. How to fix it About this vendor

    Passes when the home page links an ordering page on the site's own domain, or ordering software. Recheck it

  3. robots.txt blocks PerplexityBot +1.7 points robots_blocks

    These are the assistants' search crawlers and the fetchers that read a page when someone asks. Remove the Disallow lines for them, or add explicit Allow rules. Training crawlers can stay out: that costs nothing here. Training crawlers stay out (GPTBot, Google-Extended, anthropic-ai, CCBot, Bytespider, cohere-ai): the owner's choice, and it costs nothing in the score. Google uses Google-Extended for Gemini's answers too (grounding in Gemini Apps), so blocking it also keeps the site out of those. How to fix it

    Passes when robots.txt lets OAI-SearchBot, Claude-SearchBot, PerplexityBot, ChatGPT-User, Claude-User and Perplexity-User fetch the home page. Recheck it

  4. No llms.txt no points, but worth doing no_llms_txt

    Optional for now and cheap to add. It's a short text file that describes the business for language models. How to fix it

    Passes when /llms.txt answers with a short text file (not scored). Recheck it

Send the fixes to your web person

One email with the fixes and the link. Nothing is stored.

Already right

  • AI crawlers get the page GPTBot, ClaudeBot and PerplexityBot all got a normal response.
  • Menu or price list is an HTML page Agents can read the items and prices.
  • Uber Eats works with agents through its partners Uber's US terms of use (12 October 2026) bar scripts that burden the services and any use to build software or train AI; its Claude and ChatGPT integrations only browse menus, with checkout in the Uber Eats app. Terms read on 2026-10-15: https://www.uber.com/legal/en/document/?name=general-terms-of-use&country=united-states&lang=en. An assistant that isn't a partner may not be able to finish there. A direct path on your own domain works for all of them.

Builder: unknown · contact page /about: nothing new; menu page /documents: nothing new · What to paste

Owners

Own this business?

Own this business? Claim it, correct what agents are told, recheck a fix, or hand the list to your assistant: For the owner

For developers and agents

What agents get, and what the site publishes

The direct record agents get for this business, the rails an agent could transact through, and the agent standards the site publishes. Show the details