The smallest AI visibility budget that changes anything, and the one page a finance team will sign

The minimum budget for AI visibility is zero, and that is the honest answer rather than a soft one. The first useful reading costs nothing: a scan on your own domain, or fifteen buyer questions asked by hand in the engines' own free interfaces, three engines, three runs each, labelled yourself. Two of the next three moves are engineering hours you already pay for. Paid monitoring earns a budget line at one specific moment, which is when you have to report a number that somebody will push back on. For us that line starts at Rs 11,799 a month in India and 299 US dollars in the United States, which is Growth, the cheapest of our plans that reads AI apps at all, for 100 buyer questions on five engines every week, read from our own pricing page on 20 October 2026. We sell this software, so treat our figures as a seller's figures and check them on the page.

Order the spend so the free moves come first

A finance team almost never refuses an AI visibility budget on price. It refuses on sequence, because the first version of the request usually puts the subscription before the two fixes that make a subscription worth reading.

• Let the assistants fetch you. A crawler that is blocked cannot cite you whatever else you buy, and this is a robots file and a CDN rule rather than a purchase. Blocking AI crawlers costs you recommendations sets out which agents to check for and what a block looks like from the outside.
• Put the facts in the served page. Prices, currency, tax treatment, address and service area belong in the HTML the server sends, not in something a browser assembles afterwards. On 16 October 2026 our own pricing page failed this: fetched as GPTBot it carried rupee figures and no dollar, pound or euro figure at all, because the conversion ran in the browser. It was repaired the same day and the region table is now printed on every request. Engineering hours, usually once.
• Run fifteen questions by hand. Fifteen buyer questions, your own brand name in none of them, three engines, three runs each. That is 135 recorded answers for the cost of one afternoon, and it tells you whether you have a problem at all.
• Then buy the reading. A subscription stops being optional when the number leaves your own desk, because a frozen question set, a fixed refresh and an exportable raw answer are what survive a sceptical reader.

What a free floor actually looks like across the category

Several suppliers publish a way to start at no cost, and the limits are what matter rather than the word free. Read on their own pages on 20 October 2026: Profound publishes a free trial of fifty prompts run daily for seven days across ChatGPT, Gemini and Google AI Overviews, in one language and one region, with no history, no export and no API. AthenaHQ publishes a free tier with a stated credit allowance across the engines it names, and unlimited users. Rankscale publishes a seven day trial of its entry plan. Otterly, Semrush and Ahrefs each carry a start free or try free route beside their published plans. Ours is a scan on your own domain that needs no account and can be run again after thirty days, plus a signed in free plan that grants a one off allowance of credits.

Read those limits as a sampling design rather than a discount. Seven days of daily prompts answers a different question from thirty days of weekly ones, and a trial with no export leaves you without the evidence a finance team will ask for.

The one page that gets the budget approved

Finance signs numbers it can audit. Six rows do that work, and none of them is a chart.

Row	What goes in it
The question set	How many buyer questions, written in buyers' words, with your brand name in none of them, frozen and dated
The denominator	Readings taken, not just the hits. A rate with no denominator under it is a claim
The engines, separately	One row per engine. A blended score across five surfaces hides the only one your buyers use
Cited, as its own row	How often one of your pages was among the material an answer was built from
Named, as a different row	How often your company appears in the answer text. This is the harder number and it moves separately
The review date	A fixed date to run the same set again, chosen before anybody sees a result

The fourth and fifth rows are the ones that win the argument, and they are also the ones most proposals merge. Being read as a source and being recommended by name are two different findings that fail for different reasons, and a single blended figure covering both is the thing a sharp finance director will take apart first.

Our own rows, so you can see what a starting point looks like

Here is our own paper, which is the least flattering version available and the reason it is useful. Across 1,260 readings on 16 October 2026 there were 20,143 citations spread over 4,947 domains. Our own domain was cited in 6 of those readings, 0.5 per cent, with zero in the United States and zero on Gemini. Separately, and counted as a different row: our name appeared in the answer text zero times in those 1,260 readings, while rival products were named 1,314 times, and of 250 distinct questions 208 had a rival named and us absent. On our own pricing page we publish the same split for a client case study, cited in 78 per cent of Indian readings and named in about 1 per cent, alongside 88.4 per cent on ChatGPT against 23.2 per cent on Gemini. The published readings behind those figures carry their export date.

Two things follow for a budget request. A category where the leading measured outcome is a cited share of four fifths and a named share near one per cent is a category where the work is long and the honest proposal says so. And a starting figure of zero is normal, which makes the first review point an improvement question rather than a verdict.

What to write in the row marked expected result

Write the timing you can evidence and leave the rest open. We publish two measured windows: a new page is first used as a source in 9 to 16 days, and a buyer question flips to naming you in a further 4 to 8 weeks. Our own pricing page labels those as projections over ninety days from an ordinary starting position rather than promises, and states plainly that we have not measured how long a campaign takes to move a cited share across a whole category, which nobody has. So the row reads: first citation of a new page inside three weeks, a measurable re-read of the same frozen question set at twelve weeks, and the decision to continue taken on that re-read. A finance team will sign a review date far more readily than a percentage.

Questions finance teams ask about AI search spend

What is the minimum budget to make a difference in AI visibility?

Nothing, for the first reading and the first two fixes. Crawler access and putting your prices and facts in the served HTML are the two changes with the largest effect per rupee or dollar in this whole category, and both are engineering time. Money starts to be the constraint at the point where you need the same questions re-read on a schedule by somebody other than you.

How do I justify AI search spend to a finance team?

Hand over the six rows above with your own numbers in them and a fixed review date, and keep cited and named as separate lines. The argument that works is not a return multiple, it is that the answers buyers are reading about your category exist, are being given today, and are recorded nowhere in your analytics.

Can I move budget off search to pay for this?

Move it off the parts the measurement argues against rather than the technical and publishing work, which both channels use. In the 16 October 2026 run, 82.0 per cent of 20,143 citations sat on a commercial site somebody owns and 0.7 per cent on a review directory, so directory campaigns are the first line to question and published depth is the last. What each one costs and what each return is counted in works the two budgets through line by line.

What does the cheapest useful paid plan include?

Ours is Growth at Rs 11,799 or 299 US dollars a month on 20 October 2026: 100 buyer questions watched on ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, checked weekly, 1,200 scheduled answers a month (one credit is one buyer question on one AI app, and a Claude check uses 3), and 20 pages written and published. The cheaper SEO Start plan at Rs 4,999 or 149 US dollars writes pages and runs no AI app checks at all, which is the distinction to get right before comparing it with anything.

How do I know the spend is working at month three?

Re-read the identical frozen question set on the date you fixed at the start, report each engine on its own row, and compare ranges rather than points. Where two ranges overlap, the right sentence is that the number has not moved yet, and saying that in month three is what makes month eight believable.

The cheapest thing in this category is the question list

Fifteen questions written in your buyers' own words, with your company's name in none of them, is the only artefact here that costs nothing and decides everything else: which engines matter to you, which suppliers can price your workload, and whether the number you report in twelve weeks means anything. Write that list before you price a single plan. When buyers ask an assistant who to use in your category today, the answer is already being given, and the list is how you find out what it says.