A 30 Day AI Visibility Plan for an Indian Small Business, and What Two Readings Can Honestly Tell You
A dated four week sequence built against the Indian field as it was measured in October 2026, where the sites holding the answers are small Indian domains and one engine column starts at zero.
Published by AI Knows Us · Updated 17 October 2026
A first month of this work buys two things: a baseline you can defend and one fix you can point at. It does not buy a trend, and a plan implying one is selling you the month instead of the method. Start with the field. One frozen set of 210 Indian buying questions went to each of ChatGPT and Gemini on 16 October 2026, all 420 of those answers were usable, our own domain turned up among the sources twice, both times on ChatGPT, and the Gemini column stayed empty. That is the Indian starting line. We sell the measurement and the publishing work below, which is why our own number is at the top of the page.
Week one: write the questions an Indian buyer actually types
The month is only comparable to itself if the question list stops changing, so week one produces a file you freeze with a version and a date. Twenty to forty questions suits one business, taken from sales calls rather than website copy.
The Indian part is the qualifiers, because they change which sources an engine reaches for more than the category word does. A buyer writes in the city, or the neighbourhood. They write a rupee budget, usually per month. They ask whether GST is included or whether there is a tax invoice. Some ask in Hindi, Tamil or Marathi. Track the versions your buyers really use. One rule is not negotiable and costs nothing: your company name goes in none of the questions, because a question naming you hands the engine the answer and the reading then describes your wording rather than your position.
Week two: take the baseline, and expect the two columns to disagree
Ask every frozen question on the engines you chose, in a fresh session each time, with memory and personalisation off, and record one row per question per engine: date, engine, whether you were named, where in the answer, and every source it cited. Keep that source column: it is what writes your plan for you.
Then read the engines apart, because ours did not behave alike. Across both of our runs Gemini cited our domain 0 times in 921 readings. On 15 October 2026 ChatGPT's most cited domain turned up in 176 of its 306 readings, while Gemini's best supported domain reached 19 of its 291. On one of them you are trying to displace a short list that is already in place; on the other there is nothing in place to displace. So read a first round zero there as where that surface begins, not as something your website did. Report the two as separate fractions with separate dates, because an averaged number hides which of the two you are absent from.
Week two, the same week: fetch your own pages the way a crawler does
Set aside half an hour inside the baseline week for the cheapest check here. Pull your own important pages with no browser and no scripts running, then read what you actually get. Look for five things in the text itself: the price basis, the GST treatment, the address, the phone number and the service area. Anything assembled in the browser is not in the document an assistant reads. We found this on our own site rather than a client's: our pricing page served a crawler rupee figures and no other currency on 16 October 2026, and it was repaired the same day. The version that matters for a geo priced site is in one site, two currencies, and the pricing page an assistant actually reads.
Week three: write one page, and know who you are competing with
Choose the one buyer question where you go unnamed most often and the buyer is nearest to paying, and write a page that answers only that. The answer goes in the first paragraph, with a figure somebody can check and the date you checked it. One page of that shape beats twenty restating your category.
You will not be competing with the international names you were braced for. On 15 and 16 October 2026 we read ten named cities, six buyer questions each on both engines, 120 readings in all, and our domain appeared in none. Three of the ten were Indian and each had its own winner.
| city | readings | most cited local sources |
|---|---|---|
| Mumbai | 12 | searchmetric.in 4, agencies.semrush.com 3, goodfirms.co 3 |
| Delhi | 12 | goodfirms.co 3, nicodigital.com 3, digiustad.com 3 |
| Bangalore | 12 | onecity.co.in 4, aiseo.solutions 4, niyamkavach.com 3 |
Three cities, three sets of winners, every one a small Indian site. The wider Indian question set of 15 October 2026 said the same thing, with upgrowth.in leading on 52 readings and themediaverse.in behind it on 48. Those pages win by being specifically about the Indian version of the question. There is no figure for a city nobody measured, so if you trade in Pune or Ahmedabad, carry the market figure and label it as one.
The 20,143 citations in that 16 October run split 82.0% to somebody's own commercial website and 0.7% to a review directory. So the month is a publishing job rather than a listing job: claim your profiles, keep them accurate, and spend the rest of the time on the page. What the Indian directory layer is genuinely for, including the Bureau of Indian Standards review standard, is in our guide to reviews and directories for Indian businesses.
Week four: run the same file again, and read the result fairly
Re-run the frozen questions on the same engines in the same conditions, and write the second reading beside the first as fractions with both dates. Then do the arithmetic that decides whether it counts as movement. At 30 questions on two engines you hold 60 readings, so one reading is worth more than one and a half percentage points, and a change smaller than that does not exist in your data.
Thirty days hands you a start point and one check against it, which is a smaller thing than a direction of travel. Most first months come back close enough that the honest sentence is that nothing has been shown to move yet, and for a month that is the correct answer rather than a failure.
We hold no before and after intervention study of our own, with a treated set and an untouched comparison set, so there is no percentage point gain on this page. The design, and the four rates it has to report, are published in full in the repeated run outcome study design. For the shape a full quarter takes, a worked ninety day marketing programme filled in is the longer version of this calendar.
What the month costs
Done by hand the bill is time: a day to freeze the list, two to three days for the baseline, a day for the page, half a day to re-read. Our own free tier prints its own limits: an instant check of 5 buyer questions on 2 AI apps, and a free plan of 10 buyer questions on 2 AI apps run once with no monthly refill, read from our own site on 16 October 2026. Ten readings answer whether you were named, and who took the slot. When next month needs a rupee figure instead of a diary, the one verified rupee price in the category sits in Indian pricing for AI visibility monitoring tools in October 2026.
Questions Indian owners ask about a first month of AI visibility work
Can I do this myself, or do I need to hire somebody?
Do the first month yourself. Sitting with the answers is how you learn which page took your slot and what it said to deserve it. A spreadsheet, a clean browser profile and about five working days cover it.
How much should a small business spend in the first month?
On tools, as little as possible until you know which question you are losing. Even a Rs 499 subscription is worth deferring until the frozen list exists, because a subscription pointed at the wrong questions measures the wrong thing on a schedule.
What should I publish first?
One page answering the single buyer question you are absent from most often, with the answer in the first paragraph and a number somebody can check. Not the home page, because you cannot tell which page to rewrite until the baseline says which question you are losing.
Why is our Gemini column still zero after a month?
Because there is no entrenched list on that surface to join. Keep it in its own column and judge it on the second and third rounds.
What the answer looks like today, and who wrote it
An owner in Bengaluru asking an assistant who to use locally gets a fluent reply with three or four names in it, assembled from a handful of pages. On the Indian readings we took, those pages belonged to small Indian firms. Your analytics has no row for that conversation, and the shortlist got written while you were not looking. The month above is how you find out, and AI Knows Us exists because somebody has to ask the question and keep the answer. The source table and question set behind these Indian figures are in our benchmark of 597 readings across ChatGPT and Gemini in three markets, and the Indian cut of the newer run is in 420 Indian readings on 16 October 2026. Our product reads five surfaces; these two runs read two of them, so nothing above describes Perplexity, Claude or Google AI Overviews.