AI visibility tool pricing across 36 tools: one check costs a median $0.0205, a 71.1x spread

Study 3 method: what one AI-visibility check costs, and what each tool discloses

Status: approved by Maneesh 2026-09-27 (“approved as proposed”; named brands allowed). The merge commit of this file on main is the pre-registration. No price or disclosure is recorded for the dataset before that merge.

Pillar (D36): growth. Secondary: none. Keywords (strategy/keyword-map.md cluster B): primary ai visibility tool (US 4,500, KD 1), secondary best ai visibility tools (2,200, KD 24). The scout runs before publish to confirm no live page holds the primary. Source seed: detailed.com, AI visibility tools pricing (8 tools). This study replicates that comparison, links it and names it as the seed. It goes further in two ways: it covers every tool in the frame with a public price, not 8, and it puts every price on one unit, the single check.

Question

What does one AI-visibility check cost across every tool with a public pricing page? And what does each tool disclose about how it samples prompts, engines, locations and frequency?

Unit of measure

A check is one prompt run once against one AI engine. For each tool:

cost per check = monthly price at the plan / (prompts tracked x engines x runs per month)

• Plan used: the cheapest public plan that includes AI-visibility tracking. When a tool sells tracking bundled inside a wider SEO suite, the bundle goes in at the cheapest tier that includes tracking, and it is flagged bundled in the data.
• Annual-only prices are divided by 12 and flagged annual.
• Runs per month: read from the plan page or the tool’s documentation. “Daily” counts as 30 and “weekly” as 4.33. If the frequency is not stated anywhere public, the row has no cost per check. It is recorded as frequency undisclosed and is not imputed.
• Currency: the price as published. Anything not in USD is converted at the ECB reference rate on the observation date, and both figures are kept.

Population and frame

• Frame: the union of tools named on public lists of AI-visibility, AI-search or GEO tracking tools, enumerated on the freeze date. Every list’s URL and capture date goes into population.csv. Detailed.com’s list is in the frame. The conflict exclusion in amendment 2 applies to it like any other list, so its tools appear here minus any excluded under that rule. [Corrected on this page under amendment 2; the committed method file is unchanged.]
• Include: the tool tracks brand mentions or citations in at least one AI engine (ChatGPT, Perplexity, Google AI Overviews or AI Mode, Gemini, Copilot, Claude) and has a pricing page.
• Exclude, each with its reason in the row: no public price (“contact sales” only); the product is discontinued or the domain is dead; it is an agency service, not software; the pricing page answers with a bot challenge (read by a human in a real browser or excluded, never bypassed).
• JS-rendered prices are read in a real browser. The rendered page is archived.
• cloro (an engine-API pass-through, not a tracking product) stays out of the median. It is shown as a labelled reference row, next to TIN’s own measured cost of running one prompt through each engine’s official API on the observation date.
• Population freeze: population.csv plus FREEZE.md land on main in their own commit before any price is recorded.

Disclosure metrics (all script-checkable yes, no or not stated, from the tool’s public pages)

1. Engines covered, as a list
2. Prompts per plan
3. Run frequency
4. Whether it says how prompts are run: official API, UI or browser session, or not stated
5. Whether location or country can be set
6. Whether language can be set
7. Whether it states that answers are sampled more than once per run
8. Whether the raw answers or cited URLs are exportable

Outputs

• The median cost per check with its range and n, for all included tools and for the unbundled ones separately.
• Cost per check by engine count.
• A disclosure count per tool (0 to 8), with each item shown. No composite score and no weights.
• The full table: every tool, its plan, price, units, cost per check, disclosure items, the archived snapshot and the observation date.

Rules

• Measured, never judged. No “best” ranking on judgement. The page may sort by cost per check because that is a computed figure.
• Association only. Nothing is written as “cheaper tools are worse” or the like.
• 0 Ahrefs units. Nothing here needs paid data.
• A vendor correcting its price is handled as a dated erratum with an archived page. Nothing buys a position.

Limitations (these go on the page)

• List prices only. Discounts, trials and negotiated deals are not measured.
• A check is not equal in quality across tools, so cost per check does not measure value.
• The frame is only the lists enumerated. A tool on none of them is missing, and the page says so.

Study 3, amendment 1: the standard workload

Status: filed 2026-09-28. method.md is frozen on main, so this rule lands as a new amendment. It was agreed with the repo-side PM on 2026-09-27 and belongs to the pre-registration. It merges on its own (split from study 4’s amendment on 2026-09-28), before population.csv is frozen and before any price is recorded.

Where this amendment and method.md differ, this amendment wins.

The workload

One fixed job, the same for every tool:

• 100 prompts
• on ChatGPT, Perplexity and Google AI Overviews (all three)
• run daily (30 runs a month)

That is 9,000 checks a month (100 x 3 x 30).

How each tool is costed

• Workload cost = the monthly price of the cheapest public plan, plus any publicly priced add-on, that covers the whole workload: at least 100 prompts, all three engines, daily runs. Annual-only prices are divided by 12 and flagged annual, as in the method.
• A tool whose public plans cannot run the workload (an engine missing, fewer than 100 prompts at any public tier, or no daily frequency) is recorded as cannot run workload with the reason. It is never costed by extrapolation.
• A tool whose run frequency is not stated publicly is recorded as frequency undisclosed, as in the method, and gets no workload cost.
• “Contact sales” above a public tier counts as no public price for the workload.

Output

For every included tool, the workload cost is reported alongside the existing cheapest-plan cost per check, never in place of it. The page adds:

• the workload cost per tool, with the plan used;
• the median workload cost, with range and n, over the tools that can run it;
• the count of tools that cannot run the workload, with reasons.

What does not change

The check unit, frame, inclusion and exclusion rules, disclosure metrics, the existing outputs, the cloro reference row, 0 Ahrefs units, and the limitations in method.md all stand.

Study 3, amendment 2: tools TIN uses are excluded, counted and not named

Status: filed 2026-10-16 by the repo-side PM on Maneesh’s decision of the same day (option A). method.md and amendment-1.md are frozen on main, so this rule lands as a new amendment. It merges before population.csv and FREEZE.md, and before any price is recorded, so it is part of the pre-registration.

Where this amendment and the earlier files differ, this amendment wins.

The rule

1. Conflict exclusion. Any tool that TIN uses in its own operations on the freeze date is excluded from the study, to avoid a conflict of interest.
2. Applied before the freeze. These tools are removed from the frame before population.csv is written. They do not appear in population.csv, the dataset, the charts or the page, by name or by any detail that identifies them (domain, pricing, logo). This overrides the method’s rule that every excluded row carries its reason in the data.
3. Counted, not named. FREEZE.md and the published method state the count in one sentence: “Tools TIN uses in its own operations are excluded, to avoid a conflict of interest. n excluded.” There is no other wording, no names and no description of the use.
4. Who decides which tools. The repo-side PM gives MAV the count and the identities privately before the freeze; MAV never writes the names into any file in the repository. The identities stay out of the repo on purpose.

Why

A study that leaves out a well-known tool with no stated reason reads as an oversight or a hidden arrangement. A rule that names nobody keeps the page honest without promoting anyone.

What does not change

The check unit, the frame, every other inclusion and exclusion rule, the disclosure metrics, the standard workload in amendment 1, the cloro reference row, 0 Ahrefs units, and the limitations in method.md.

Study 3 population freeze, 2026-10-16

This commit freezes the population. No tool is added or removed after it. It lands on main on its own, before any price or disclosure item is recorded for the dataset (method.md, “Population freeze”).

Tools TIN uses in its own operations are excluded, to avoid a conflict of interest. 2 excluded.

Frame

The union of tools named on public lists of AI-visibility, AI-search and GEO tracking tools, enumerated on 2026-10-16 in a real browser. Every list, with its URL and capture date, is in frame.csv: the detailed.com seed plus nine roundups returned for the study’s keywords and for “GEO tools” and “AI search visibility tracking tools”. Two further lists answered with a bot challenge and were not enumerated; they are recorded there, never bypassed. A tool named on none of these lists is outside the frame, and the page says so.

Screen

Each tool was screened on its vendor’s own public pages (no account, no sign-up, no bot-challenge bypass) for the inclusion rule: it tracks brand mentions or citations in at least one AI engine, and it has a public price. No price was recorded in the screen.

population.csv has one row per screened tool: status, the reason for every exclusion, the lists that named it (sources), its pricing page, and the evidence quote it was screened on. Tools sold inside a wider SEO suite are included and will be flagged bundled in the dataset, per the method. The cloro reference row is not part of the population (method.md) and is added at collection.