How the Gist 100 is made

When someone asks an AI assistant what to buy, the answer names a handful of brands and leaves the rest out. The Gist 100 measures who gets named. For each brand it reports one number: the share of AI answers about its category that mention it.

That is a measure of visibility, not of quality. The brand at the top of a ranking is the one the assistants bring up most often when a buyer asks an open question. It may not be the best product, the biggest seller or the one with the happiest customers. Being named often usually reflects how much has been written about a brand, where, and how consistently.

The questions

Each category has eight buying questions, written the way a buyer would type them:

• the plain question: what is the best one right now
• a request for recommendations
• the cheap option: affordable, budget, free or low-cost
• what is worth paying more for
• two questions for a particular buyer or need, such as a startup, a mid-size company, a beginner or sensitive skin
• a shortlist: which three to consider first
• one question about a quality buyers care about, such as which is easiest to adopt or which lasts longest

Three rules hold for every question. No question names a brand, so nothing is suggested to the assistant. Every assistant is asked exactly the same words. And the set is fixed: the same questions are used each time the list is measured, so one edition can be compared with the next.

This edition covers 256 questions in 32 categories, 16 in business software and 16 in consumer goods. Every ranking page prints the eight questions it used.

The assistants and the answers

Each question is put to four assistants: ChatGPT, Perplexity, Claude and Google AI Overviews. Each one is asked the same question ten times, with web search switched on. Answers change from one asking to the next, and ten of them show the pattern where a single answer shows a moment.

The questions run through Gist’s own measurement platform. Every answer is collected fresh, with no chat history and no personalization. For Google, the answer is the AI Overview shown above the search results for that question.

That makes 10,240 lookups. Not all of them produce an answer to read. Google shows an AI Overview for some searches and not for others, and some searches did not come back in time. In this edition, 41 Google searches returned no AI Overview and 660 did not return a result in time. Those are left out, so Google’s figures are shares of the AI Overviews Gist read, not of every search it made.

Answers counted in this edition

The answers were collected on October 5 to October 6, 2026.

Counting names

Every answer is read by software that looks for brand names. It works from a list of names for each category, built by hand and checked against what the answers wrote: each brand, its common spellings, and the product names that stand in for it. An answer that recommends HelloSign is counted for Dropbox Sign, the product’s current name. An answer that recommends Divvy is counted for BILL, which owns it. The list holds about 2,500 names.

A few rules keep the count honest.

• Once per answer. A brand counts once in an answer, however many times the answer repeats it.
• Exact matches. A name has to appear as it is written. A misspelled brand is missed.
• A mention is a mention. The count does not judge whether an answer praised a brand or warned against it. Most mentions in answers to these questions are recommendations, but the count does not check.
• Only the things being chosen. Stores, marketplaces, review sites and publications are recognized and left out of the rankings. So is a product named only as something another product connects to. An accounting package mentioned as an integration in a payroll answer is not counted as payroll software.

The score

A brand’s score is the share of answers in its category that name it. It is worked out for each assistant first, and the four shares are then averaged, so each assistant counts equally however many answers it returned. A score of 80 means the brand is named in about four of every five answers.

Each ranking page shows the ten most-named brands in its category and leaves off any brand named in fewer than one answer in ten. Where two brands have the same score, the one named first in more answers is placed higher.

Scores a few points apart should be read as a tie. Ten answers per question is enough to see a pattern, not to split hairs.

Two more numbers appear on the pages. Named first is the share of answers in which a brand is the first one mentioned. By assistant is the brand’s share in each assistant’s answers taken alone, and it often tells a different story from the average. A brand can be in nearly every answer from one assistant and none from another.

From 32 rankings to one list of 100

The Gist 100 puts every category together. Each brand appears once, in the category where it scores highest, and brands are ranked by that score. Because a score is a share of a brand’s own category, a CRM platform and a running shoe can sit side by side. The list ranks how firmly each brand holds its own category. It does not compare the two with each other.

In this edition the hundredth brand on the list scores 53.

The websites behind the answers

Assistants show their sources. Gist records every link an answer cites and counts, for each website, the share of answers that cite it at least once. The shares are averaged across the four assistants in the same way as the scores. That is the basis of the section called “Where the answers come from” on each ranking, and of the list of websites AI cites most.

Two cautions. A citation shows what an assistant pointed to, not what persuaded it. And in this edition Perplexity returned the same list of sources for all ten answers to a question in 251 of 256 questions, though it worded each answer differently. Its citation figures behave more like one reading per question than ten.

The words around the numbers

Every ranking comes with a written read of what the answers show, and the analysis section carries longer pieces and case studies built on the same counts. One rule covers all of it: every number in the text is taken from the counts, and the text does not claim to know why an assistant prefers one brand to another. Counts show patterns. They do not show causes.

The company profiles on the ranking pages are a separate thing. They say who a company is, and they come from the company’s own pages and from Wikipedia, not from the AI answers.

A brand’s logo is the icon that brand publishes for its own website, and it is shown only to identify the brand. Where a page carries a photograph or a product picture, the brand supplied it in its public press kit, and the line under the picture says whose it is. The artwork on the articles is made of those logos and nothing else.

What the list cannot tell you

• It is a snapshot. Assistants change their models and their sources often, and a ranking can move between editions.
• It is one set of questions. Ask about a different need, a different budget or a different country and the names change. One word such as “affordable” reorders most categories.
• It is not your customer’s screen. Answers were collected fresh. A logged-in person may see an answer shaped by their history.
• Google rests on fewer answers. The missing lookups make Google’s figures less steady than the other three.
• It counts names. It does not measure sentiment, sales, market share or customer satisfaction.

Independence

The Gist 100 is a measurement and is never for sale. A brand cannot pay to appear, to move up or to be left out. Lists made with a partner are labeled as partner lists. They are picks, not measurements, and paid partnerships say so on the page.

The list uses its own fixed questions. It never draws on the scans Gist runs for customers, and being a Gist customer has no effect on a brand’s place.

Gist sells software in one category it could have ranked: tools that track how brands appear in AI answers. That category is left out on purpose.

The Gist 100 is not authorized, sponsored or endorsed by any company it ranks. Brand names, logos and pictures belong to their owners.

The data

The numbers behind every ranking can be downloaded as a spreadsheet: one row for each brand in each category, with its score, its score in each assistant, how many answers named it and how often it was named first. The citation counts are a second file.