GTM AI Index
Subscribe

The measurement, delivered monthly

The index is free to read and its full record is free to download. A subscription is the finished edition on the day it compiles: the standing computed, the movement qualified against the noise floor, your product’s rows in their own file, and the whole edition in one database. It is the work of running this instrument every month, done once and shared. An expanded subscription adds a monthly run of the flagship models on your category, for you and whoever else subscribes to it.

Three ways in

Each is independent. Two of them are bought here.
1

Claim your page

Free

Every product named by a model has a page, and so does every product in the vendor table that was not named. A claim gives a verified contact, the right to propose corrections to the vendor table, a claimed mark on the page, and an email whenever the standing moves by more than the noise floor. It gives no change to labels, shares or verdicts, no preview, and no say over which quotes appear.

Find your product

The claim form is on the product page.

2Standard Edition Pack

The finished edition, delivered monthly

The same data as the free record, joined and early. Every edition on the day it compiles, fourteen days before the free record.

$9.95a month, or $99 a year
The models
Claude Haiku 4.5
Anthropic
GPT-5.4 mini
OpenAI
Gemini 3.5 Flash
Google
Perplexity Sonar
Perplexity
Grok 4.1 Fast
xAI
Mistral Small
Mistral
DeepSeek V4 Flash
DeepSeek
Llama 4 Maverick
Meta
Qwen 3.7 Flash
Alibaba
Kimi K2
Moonshot
GLM 4.7 FlashX
Z.AI
MiniMax M2.5
MiniMax

73 categories, each asked six ways on behalf of a small, a mid-market and an enterprise B2B company, from the September 2026 Edition.

The category report73 PDFs, every edition

One written report per category: the standing in each segment, what the models agreed on, the sources they cited, and a summary in prose. Every category on this pack.

Read a sample: CRM, September 2026 PDF
Or one edition, no subscription.Single edition pack · $9
Subscribe to the Standard Edition Pack

Fourteen days, no questions asked, on every purchase here: the refund policy. A subscription keeps the price it was bought at. The files are CC BY 4.0. Monthly and annual are chosen at the checkout.

3Expanded Edition Pack

The flagship models on your category, every month

Asked the same six framings, on behalf of the same buyer segments as the public index, on the one category. Delivered in the pack files and over MCP.

$49.95a month, or $499 a year, per category
Everything in the Standard Edition Pack

Every edition of the public index on the day it compiles, all the files below, the MCP server. Nobody needs both subscriptions.

Plus six flagship models, on one category
Claude Opus 5
Anthropic
Claude Opus 4.8
Anthropic
GPT-6 Astra
OpenAI
GPT-5.6 Sol
OpenAI
Gemini 3.1 Pro
Google
Perplexity Sonar Pro
Perplexity
It never enters the public standing. One subscription is one category; another category is another subscription.
The category reportYour category, eighteen models

One report over the whole tier, standard and flagship models pooled, with the two tiers set side by side inside it. The standard reports for every other category come with the pack.

Read a sample: CRM, September 2026, eighteen models PDF
Subscribe to the Expanded Edition Pack

The button stays disabled until a category is chosen. Fourteen days, no questions asked: the refund policy. A subscription keeps the price it was bought at. The files are CC BY 4.0.

What arrivesIn both packs, in the same files. Downloaded from the site with the license key on the receipt; a subscription key opens the newest edition. The release note is emailed the day an edition compiles.
standing.csv
Share, rank, negative rate, labels and verdict for every category and product.
movement.csv
Every product's change since the previous edition, flagged against the noise floor.
labels.csv
Every judge label with its verbatim evidence quote.
vendors/<product>.csv
Your rows in their own file.
responses.csv · prompts.csv · models.csv
The run behind the standing.
index.sqlite
Everything in one database, with a leaders view.
README.md
The data dictionary.
MCP server
The index served to Claude, Cursor or any MCP client, with the license key.
Why a team needs this

When a buyer asks an AI assistant which product to use before they build a shortlist, the first answer is often the first pitch they hear.

The index asks each category six ways, on behalf of three sizes of buyer, across twelve models, every month, and a judge labels every product named with the sentence it was named in. That is a measurement a team can act on, not an anecdote from one chat window.

50%

of first choices changed when the same model was asked the same question twice, with nothing changed. One ask is a coin flip.

The index asks seventy-two per category, per buyer size: six framings across twelve models, every month.

Same model, same question, two runs.

Eight things a subscriber can do

Each one is a question, a view the files produce, and the reason it is worth having. Nothing here needs a chat window re-run by hand.
01Battlecards

What does the model tell a buyer about us, and about the three products we lose to?

ModelYoursRival ARival BRival C
One cell per product per model: the label, and the quote behind it.

A matrix of product by model, each cell the label the model gave, with the verbatim quote under it. Objection handling written from what the buyer's assistant already said, refreshed monthly with nobody re-running the chats.

From labels.csv joined to responses.csv
02Retrieval

When the model recommended in our category, what did it read first?

Review siteComparison postYour siteForum threadVendor blog
Cited domains behind the answers, ranked.

Cited domains ranked per category, and your own domain's share of citations edition over edition. It shows which third-party pages carry the recommendation into the answer. That is a fact about retrieval, not a lever on the model.

From responses.csv, cited URLs where the model searched
03Buyer size

Are we the small-business pick and absent at enterprise, or the reverse?

SMBMid-marketEnterprise
Share by segment, three standings, never pooled.

The same category, computed three times, once per buyer segment. Positioning gets evidence from the buyer side of the screen: which size of company the assistant thinks the product is for.

From standing.csv, per segment
04Framings

Do we come up when the buyer asks plainly, only when budget is tight, or as the product to avoid?

DirParaCompBudgScaleNeg
One bar per framing, for one product.

Six bars per product, one per framing. The framing a product wins is the buying situation to sell into, and the negative framing is the objection list, sourced. The paraphrase pair says whether the answer survives a rewording.

From labels.csv by framing
05Movement

Did our standing move this month, and is the move bigger than noise?

Noise floor
Rank per edition, with the measured band drawn on it.

A rank timeline with the noise band drawn on it, and a movers table for the category. A single month's shift is usually inside the noise. The index says which ones are not, with the date and the model version attached.

From movement.csv, history.csv, models.csv
06In the deck

What have the assistants said in our favor, in words we can show a prospect?

Claude Opus 5 · directFirst choice
GPT-6 Astra · comparativeFirst choice
Gemini 3.1 Pro · budgetAlternative
Quotes stacked by model, each with the edition it came from.

Every label for one product, with the verbatim evidence quote, the model, the framing and the edition beside it. A slide can show what an assistant said rather than paraphrase it, and the number on the site can be traced to the same row.

From vendors/<product>.csv
07The whole index

Where is no product the consensus, and where are we named without being ranked?

ConsensusClear leaderContestedA product of yours appears
Marketing
Sales
Customer
Revenue operations
GTM data and infrastructure
Partner and channel
One square per category, 73 in all. The outlined squares stand for a subscriber's products; the verdicts are drawn by the build.

The verdict map: 73 categories by go-to-market function, colored by verdict. Category strategy from the assistant's point of view: where the field is settled, where it is not, and where a product is already in the conversation without leading it.

From standing.csv across all 73 categories, with the verdict per category
08By assistant

Which assistants recommend us, and do the flagship models say the same as the models on a free plan?

Twelve modelsThree products
Where they agree, and which assistant is the outlier.

An agreement matrix, model by product, and the standard run against the expanded one for your category. The answer a buyer hears depends on which assistant they opened, so a strong standing in three and none in the fourth is a specific, named gap.

From vendor_by_model.csv, plus the expanded run

Into the tools a team already runs

Nothing above needs a new place to work.

A warehouse

index.sqlite loads as one file and the CSVs load as tables. One query reproduces any number on the site, so a dashboard built on it can be checked against the public page.

An assistant

The MCP server puts the edition inside Claude, Cursor or any MCP client: ask what the models said against a product this month and get the labels and quotes with the edition they came from.

A battlecard system

Per-product files carry every label and quote for one product, which is the shape a card or enablement system ingests.

A monthly digest

The release note and the movement email arrive on the day an edition compiles. The files behind them arrive with the same key.

Why it has to be this dataset

Five properties. Each is a thing the files carry, not a claim about the market.

Many asks, not one. Seventy-two answers per category per segment: six framings across twelve models, every month.

A judge, not a reading. Every product named is labeled by the same model with the same rubric, and the label carries the sentence it came from.

An error bar. The calibration repeat measures how much the models move with nothing changed. Movement is only called when it clears that floor.

The record is open. Every prompt, answer, label and the rubric are published under CC BY 4.0. The pack is the same record, joined, early, and in the shape these uses need.

It is monthly. Model versions change and what they retrieve changes. The version string on every answer says which one answered.

Pricing

Flat. Annual is ten times the monthly price, two months free. A subscription bought at one price keeps that price for as long as it stays active.
WhatWhat it isMonthlyAnnual
SubscriptionEvery edition on the day it compiles: the standing and movement computed, your rows, one database, the MCP server.$9.95$99
Expanded, per categoryEverything in the subscription, plus the six flagship models on one category every month, in the same files and over MCP.$49.95$499
One editionA single edition's pack, kept for good. No subscription.$9once

Ask a question

Both packs are bought above. Anything else, say what you have in mind: a category the index should cover, a cut of the data, a question about the method.