What each AI visibility metric means
A number without a definition is decoration. These are the four we publish and exactly what each one counts.
| Metric | Definition |
|---|---|
| Citation rate | The share of the category prompt set where at least one leading Indian brand is named in the generated answer. It counts brand presence, not sentiment and not position. |
| AI Overviews presence | The share of the same prompts where Google returned a generated overview above the organic results, checked on the same day from an Indian IP address. |
| Displacement risk | A judgement, not a measurement. It combines AI Overviews presence with how much of the category's traffic historically came from informational queries. Low, medium or high. |
| Unclaimed slots | The inverse of citation rate. It is the share of prompts where no leading Indian brand appears, which is the space a brand can move into. |
Twenty buying prompts per category, run weekly across ChatGPT, Gemini, Perplexity and Google AI Overviews from Indian IP addresses.
AI visibility index method: The AI visibility measurement method, weekly
- 01Write the prompt set
Twenty buying prompts per category, written the way an Indian buyer types into a chat window rather than the way a keyword tool phrases it. Long, specific, tied to a use case, a budget or a material. The set is frozen so the trend stays comparable across editions.
- 02Run four engines, same day
ChatGPT, Gemini, Perplexity and Google AI Overviews, run from Indian IP addresses on the same day each week. Four engines matter because the answers diverge: a brand can be named in two and absent in the other two.
- 03Log every mention
For each prompt we record whether a leading Indian brand is named, where in the answer it appears, and which source the engine used to say it. The source is the part most people skip and the part that tells you what to do next.
- 04Roll up, never publish brand level
Category averages are published. Brand level results go to that brand alone, through the free scan. We do not publish a league table of who is losing.
- 05Date the edition and leave it alone
Each edition is dated and archived. Numbers are recomputed for the next edition rather than quietly edited into the last one.
Published before you ask, not after the third call
Method, pricing, limits and tools. All of it live on this site right now.
What this dataset does not claim
Published because a measurement product that hides its limits is not a measurement product.
- This is a pilot dataset, not a census. It describes the prompt sets we run, not every query in the category.
- Model answers vary by run, account, region and index refresh. Single checks are noise. We report rolling averages and so should anyone quoting this.
- Category prompt sets are written by us. A different set would produce different numbers, which is why the prompt shapes are published below.
- Brand level results are never published. They go to that brand alone, through the free scan.
- Numbers are recomputed each edition. Editions are dated and previous editions are not silently rewritten.
Published before you ask
Pricing, method, limits, scorecard cadence and seven working tools, all live on this site right now rather than appearing on call three.
You can qualify yourself out in ten minutes without speaking to anyone. That is the point.
Scan window and sample
Each edition publishes its own scan window, prompt count and engine list here as it closes, so a number can always be traced to a period.
| Edition | Scan window | Categories | Prompts per category |
|---|---|---|---|
| Edition 01 | Pilot, window published on close | 12 | 20 |
Editions are archived on the editions page. Our keyword demand dataset for India is published in full at what India searches for.
Where the prompt sets sit
Each category page publishes the prompt shapes its set is built from, alongside the numbers.
Citing the India AI Visibility Index
The index is free to cite. Attribute it as the India AI Visibility Index by GrowWithBA, name the edition, and link to the category page rather than to a screenshot, because the numbers are recomputed each edition.
If you are checking our work, the honest test is to write your own twenty prompts in a category, run them across the same four engines for four consecutive weeks, and compare the rolling average. A single run will not match, and it should not.
Twenty buying prompts, four engines, you against three named competitors. Report in one working day. No call needed.