AI visibility

Our pilot AI visibility audit scored 35 out of 100. Here's what it taught us.

Before we sold a single AI visibility audit, we ran a pilot. One Delhi services business, the questions its customers actually ask, put to ChatGPT, Gemini and Google's AI. The result was worse than expected, and it shaped everything we do now.

Harsh Khanna, Founder · 4 min read

The pilot

We asked three assistants (ChatGPT, Gemini and Google's AI Overviews) 20 questions a buyer might ask when looking for a business like this one. Things like a digital marketing agency for a restaurant in Delhi, who builds websites for small businesses, or who does AI visibility audits in India. Some were in English, some in Hinglish, the way people really type.

Every answer was captured with a screenshot, in a fresh chat with no history and no prior prompting about the business. Then the whole thing was scored out of 100.

35/100

Pilot score

0 of 54

Discovery answers that named the business

6 of 6

Answers that knew it when asked by name

60

Answers captured

Pilot audit. The same questions and settings are used for every re-test.

What we found

AI knew the business, but never recommended it. Asked about it by name, all three assistants described it correctly: where it is, when it started, what it does. But on every question a new customer would ask, all three recommended someone else. The gap was not recognition. It was discovery.

Shortlists came from Google profiles with reviews. For local questions, Google's AI showed its picks as Business Profile cards, each with between 9 and 257 reviews. The pilot business had no reviews and sat under the wrong category. Being physically in the right neighbourhood was not enough; the assistants named whoever had a profile that said so, with customers vouching for it.

One article can decide the whole answer. Asked for GEO consultants in India, Gemini built its entire answer from a single 'top agencies' article, written by an agency that placed itself third. AI answers right now lean heavily on whoever has published a clear page on the exact question.

Published prices get quoted. Asked who does AI visibility audits in India, ChatGPT replied with a price table built entirely from prices firms had published on their own sites. A business that says 'depends on scope' gives AI nothing to repeat.

Answers move. The same question asked twice, an hour apart, returned different shortlists. One test is a snapshot, which is why every audit we run is re-tested on a fixed schedule with the same questions.

How the pilot became our method

The pilot told us something most businesses don't know yet: a good website is no longer the finish line. It is one signal among several, and AI weighs the others just as heavily. So the pilot became the way we work:

  1. Test before touching anything. Real customer questions, per industry, in English and Hinglish, across all three assistants, captured as evidence.
  2. Fix the profile first. Right name, right category, hours, photos and a steady flow of genuine reviews. It is the cheapest fix and the one AI leans on most for local questions.
  3. One page per question. A page that answers 'AI visibility for restaurants' beats a homepage that mentions everything.
  4. Consistency everywhere. The same name, address and phone across every directory AI trusts, with structured data that ties the business, its people and its services together.
  5. Re-test on the same questions. If the score doesn't move, the fix didn't work, and we say so.

What we will not do, for anyone: fake reviews, paid 'best agency' lists dressed up as independent, or prompting the assistants before a test. They inflate a score without making a business any easier for a real customer to find.

Where the score goes from here

AI answers do not update the moment a business changes. Google's AI Overviews work from Google's index, so a corrected profile or a new page shows up once it is recrawled. ChatGPT and Gemini blend live web results with what their models already learned, which is why third-party mentions and reviews take longer to land but last longer once they do.

That shapes the curve. Profile and on-site fixes move the local and brand questions within weeks. Reviews, listings and mentions compound over two to three months and move the harder discovery questions, the 'best agency in Delhi' kind, where the real customers are.

35

Pilot score

Day 30

First re-test

Day 90

Full re-test

90+

The target

Re-tests use the same 20 questions and settings. Results are published here when they're in.

If you want to know where your own business stands today, the quickest start is our free AI visibility check: three of your customers' questions, three assistants, results in 48 to 72 hours.

Questions

Straight answers.

Want to see what AI says about your business?

Get the free AI check →