Skip to content

The brand ChatGPT names most is in under 2% of answers. How do you calculate your own share?

Ahrefs counted brand mentions in 5.5 million US ChatGPT answers in November 2026. What its method teaches about the base and counting rules of your share.

By Published 14 min read
A figure holds a magnifying glass over the few highlighted cards in a large grid of answer cards, beside a pie chart with one thin slice.

The short answer

Calculate share of answers on your own buyer prompts, not on everything ChatGPT is asked. In Ahrefs' count of 5,516,381 US ChatGPT answers in November 2026, the most-named brand, Walmart, appeared in 1.73%. Fix the base, count each answer once, merge your name variants, check ambiguous names by reading answers, and report a range.

Key takeaways

  • Ahrefs counted brand mentions across 5,516,381 US ChatGPT answers on every topic in November 2026, and the most-named brand, Walmart, appeared in 1.73% of them.
  • A share of answers is only as meaningful as its base: across every topic it measures how often people ask about your category, while across your buyer prompts it measures how often you win.
  • Counting rules move the number as much as the answers do, so a report states whether an answer counts once, which name variants merge and which ambiguous names were checked by hand.
  • Ahrefs read real answers before counting brands that share a name with a word or another company, and the same check belongs in every brand's own tracking.
  • Our view: a share of answers in a report needs four things beside the percentage: the prompt set, the engine, the counting rules and a range from repeated runs on different days.

What did Ahrefs count, and what did it find?

Ahrefs counted how many US ChatGPT answers named each brand, across every topic in its answer index. Ahrefs sells AI visibility tracking, and the data comes from its own index of answers to real, search-backed questions. It took all 5,516,381 US ChatGPT answers in the index in November 2026 and counted the answers that named each company.1 Walmart came first, named in 95,605 answers, or 1.73% of them.1

The rest of the top four sat lower still. Target appeared in 0.94% of the answers, Apple in 0.82% and Amazon in 0.70%, in the same November 2026 count.1 Stores took six of the top 10 places, and carmakers took 15 of the top 100.1 Ahrefs' reading of the answers explains the pattern: stores get named when people ask where to buy, and product brands when people ask what to buy.

The ranking is a useful public dataset, and it is also a lesson in how share of answers works. Every number in it depends on two choices Ahrefs made and stated: which answers form the base, and what counts as a mention. Your own tracking makes the same two choices, often without writing them down.

US ChatGPT answers, all topics, November 2026

The most-named brands, as a share of every answer

1.73%of answers named Walmart, the most-named brandAhrefs, Nov 2026 1
0.94%of answers named Target, in second placeAhrefs, Nov 2026 1
35%of top-100 mentions went to the 23 retailers on the listAhrefs, Nov 2026 1
Source: Ahrefs, the 100 most mentioned brands in ChatGPT, 5,516,381 US ChatGPT answers across all topics.

A disclosure before the method

Sigzen AI sells AI visibility audits and programmes at published prices, so we have an interest in how share of answers gets measured.3 Ahrefs sells in the same market. Everything below works with any tracker that stores the raw answers, and most of it needs only a spreadsheet.

Why does even the top brand sit under 2%?

The base holds every question people ask, and most of those questions have nothing to buy in them. Ahrefs says so in its method: the base includes recipes, health questions and everything else, which is why even the top brand appears in under 2% of answers.1 A low share in that base is not a weak result. It is a small slice of a very wide question set.

Not every answer in a wide base even looks at the web. Ahrefs, in a separate study, found that only 14.4% of 66.6 million free, logged-out ChatGPT responses on GPT-5.6 Luna triggered a web search in August and September 2026.2 A base that mixes small talk, coding help and shopping questions blends answers written from memory with answers written from search. Our piece on ChatGPT's web search rate covers what that does to a prompt set.

So the all-topics share mostly measures demand. Walmart leads because people ask ChatGPT where to buy everyday things very often, and Walmart is the default answer to many of those questions. A payroll software vendor or a dental clinic would show a tiny share in the same base, even if ChatGPT named it in every relevant answer.

The same logic runs the other way. A brand can climb an all-topics ranking because a news story, a product recall or a seasonal event sends more people to ask about it, while its share of the buyer questions that matter stays flat. Reading an all-topics share as a win or a loss confuses what people ask with how the engine answers. Keep the two apart from the first chart.

Three bases for share of answers, and what each one measures

The same answers give three different percentages, depending on what you divide by. Each one answers a different question, and a report that mixes them misleads its readers. Pick the base before you look at the result, then keep it fixed.

  • All answers on every topic. The base Ahrefs used. It measures how much of the whole conversation touches your brand, which mostly reflects how often people ask about your category.
  • Answers to your buyer prompts. A fixed set of questions your buyers ask, each run several times. It measures how often you are named when the question is one you want to win.
  • Mentions among a rival set. Your mentions divided by all mentions of the brands you track. It measures your slice of the names, and rises or falls with your rivals.

Each base has a fair use. The first suits market sizing and public rankings. The second is the one a brand can act on. The third is useful in competitive reviews, as long as the rival set is frozen. Our guide to picking a rival set explains how to choose one that means something.

Three bases, three questions

Every topic

  • Base: all answers in an index
  • Measures: demand for your category
  • Use: market sizing, public rankings

Your buyer prompts

  • Base: a fixed prompt set, repeated runs
  • Measures: how often you win the question
  • Use: targets, plans, before and after

Rival set

  • Base: mentions of tracked brands
  • Measures: your slice of the names
  • Use: competitive reviews
Illustrative: Sigzen AI's comparison of the three ways to calculate share of answers.

Which base should a brand report?

Report your buyer-prompt share as the headline, and keep the other two as context. A buyer-prompt share answers the question leadership actually asks: when people ask the questions that lead to a purchase, how often do we appear? It moves when your pages, your listings or the engines change, and it can carry a target.

The all-topics share rarely belongs in a brand report. For most companies it rounds to zero, and when it moves, the cause is usually a change in what people ask. If you sell in a category that does appear in public rankings, quote the ranking as market context with its date and base, and keep it off the trend line.

The rival-set share has a trap. It can rise because a rival dropped out of the answers, not because you gained. Show it beside the buyer-prompt share, so a reader can see whether your own appearances changed.

Our view: one base per chart, named in the chart title. "Share of answers, 150 buyer prompts, ChatGPT, three runs" is a label a finance team can audit.3 "AI visibility 12%" is not.

Counting rules that move the number

Once the base is fixed, the counting rules decide what lands in the numerator. Ahrefs published four cleaning rules beside its ranking, and they make a practical checklist for anyone counting mentions.1 Each rule can move a brand's share, so each belongs in your method note.

The first rule is one count per answer. Ahrefs counted the answers that named a brand at least once, so an answer naming Walmart three times counts once.1 Counting every repeat instead rewards long answers and lists that repeat a name, and it makes engines that write longer answers look kinder to you.

The second is merging name variants. Ahrefs counted The Home Depot and Home Depot together, and Walmart with walmart.com.1 Your list should cover the trading name, the legal name, the domain, common abbreviations and the old name if you renamed.

The third is keeping sub-brands apart when they carry their own name. Ahrefs did not count Disney+ inside Disney.1 Decide in advance whether a product line counts as the company, and write the decision down.

Counting rules to fix before the first run

  1. One count per answer

    An answer that names you three times counts once.

  2. Merge your name variants

    Trading name, legal name, domain, abbreviations, former names.

  3. Decide on sub-brands

    Product lines with their own name count apart, or together, but always the same way.

  4. Read ambiguous names

    Check a sample of answers by hand before you count a name that is also a word or another company.

  5. Freeze the rules

    Change them only with a dated note and a recount of the earlier weeks.

Illustrative: Sigzen AI's counting checklist, adapted from the cleaning rules in Ahrefs' method.

How do you tell your brand from a word or a namesake?

Read a sample of the answers before you count, and keep the sample's result in the method note. Ahrefs left GE, United and Ram out of its ranking because too many mentions meant something else.1 For "Delta", it read a sample, found that 86% of the answers it checked meant the airline, and counted those answers for Delta Air Lines.1

That is the right approach for any brand with a common name. A string match on your name will count answers about the word, a place or a company with the same name. If you share a name with another firm, those answers can inflate your share in one country and hide a real problem in another. Our guide to separating your company from a namesake covers the fixes once you find the overlap.

Make the check repeatable. Draw a random sample of matched answers each month, mark each one as yours, someone else's or unclear, and record the share that is yours. If that share falls, tighten the match, for example by requiring your domain or product name nearby, and recount.

Our view: if you cannot say what share of your matched answers were really about you, the share of answers is not ready to report. One hour of reading each month costs far less than a quarter of decisions made on a polluted number.

A wide field of small answer cards with a ring drawn around one cluster, a few highlighted cards inside it and most outside.
The share depends on where you draw the circle: every answer, or the answers to your buyers' questions.

How many answers make a share you can trust?

Enough runs that a single odd answer cannot move the headline, spread over different days. Engines give different answers to the same prompt from run to run, so one pass through a prompt set is a sample, not a census. MaxAEO, which sells AI visibility tracking, has written about how repeated runs of the same prompt behave as clusters rather than independent answers, which narrows what a large run count can tell you.4

That point matters for the base. Ahrefs' ranking rests on millions of answers, so its percentages are steady even at small values. A brand's buyer-prompt set might hold a few hundred answers. At that size the range around a share is wide, and two adjacent months can differ by chance alone.

The practical answer is to repeat each prompt on separate days and report the share with an interval. Our own audit runs 150 buyer prompts across five engines, each prompt on three separate days, and reports share of answer with confidence intervals against five competitors.3 Any tracker that stores raw runs can do the same arithmetic.

Reading an all-topics ranking without misusing it

Use a public ranking for what it measures: who dominates the general conversation on one engine, in one country, in one month. Ahrefs' list covers US ChatGPT answers in November 2026, so it says nothing about Gemini, Google AI Overviews or another country.1 Check those three facts before anyone pastes the ranking into a slide.

The industry split is the most useful part for planning. Retailers held 23 of the 100 places and 35% of the top-100 mentions, and carmakers came second with 14%, mostly from questions about reliability, resale value and which model to buy.1 If your category is one ChatGPT already gets asked about at that volume, the buyer questions there are worth mapping first.

Avoid two misreadings. A brand's absence from a top-100 list says nothing about how it performs on its own buyer questions. And a high place does not mean ChatGPT recommends the brand, only that it names it. Our method for coding how answers describe a brand separates naming from recommending.

Illustrative: one set of answers, three shares

Buyer-prompt share: named in 150 of 600 answers25%
Rival-set share: 150 of 900 tracked mentions17%
All-topics share: the same answers in a whole index~0%
Illustrative: Contoso Growth, a fictional vendor; invented round numbers showing how the base changes the share.

What should you ask a vendor about its share of answers?

Ask for the base, the counting rules and the range before you look at the percentage. Any tracker, agency or audit that quotes a share of answers made the same choices Ahrefs made, and a good one can name them in a minute. If the answers are vague, the number cannot be compared with anything else you hold.

Five questions cover most of it. Which prompts form the base, who wrote them and when did they last change? Which engines and countries ran them, and on which dates? Does an answer count once, and which name variants are merged? How were ambiguous names checked? How many runs sit behind each prompt, and what range comes with the share?

Ask for the raw answers too. A share you cannot trace back to the answers that produced it is a claim, not a measurement. Most trackers can export the answers with the prompt, engine and date, and a sample of those rows lets you repeat the count yourself.

Our view: a vendor that publishes its method, as Ahrefs did for its ranking, deserves more trust than one that publishes only a score. The same standard applies to our own reports, which carry the method note on the first page.

When the base has to change

Change the prompt set on a schedule, not whenever a number disappoints. Buyer questions shift, products launch and markets open, so a prompt set that never changes drifts away from what buyers ask. A quarterly refresh keeps it current without breaking the trend line every month.

Split the set into a fixed core and a rotating part. The core stays the same for at least a year and carries the trend line and the target. The rotating part follows new products, seasons and questions from sales calls, and is reported on its own line.

When the core must change, recount the earlier periods on the new base wherever the stored answers allow it, and mark the chart with the date. The same rule applies to counting rules: a new name variant or a tighter match means a recount, not a silent jump in the line.

What your report should show from this month

Put the method beside the number, every time. A share of answers without its base, engine and counting rules cannot be compared with last quarter, with a rival's claim or with a vendor's ranking. Four lines of method under each chart are enough.

  1. Base. The prompt set, its size, its countries and when it was last changed.
  2. Engine and access. Which engine, which tier or login state where you know it, and the run dates.
  3. Counting rules. One count per answer, the variant list, the sub-brand decision and the result of the monthly sample read.
  4. Range. The interval from repeated runs on separate days, so readers can tell a real move from noise.

Keep public rankings on a separate page, labelled as market context, with the publisher's base and month. They help a leadership team see where AI answers already shape buying in their sector. They do not replace a number your team can move.

If you want the buyer-prompt base built for you, our AI visibility service starts with a prompt set from your buyers' questions, the counting rules written down and a share with a range for each engine.

Our view: Ahrefs' ranking is most useful to brands for its method section. The base and the cleaning rules it published are the questions to ask of any share of answers, including your own tracker's and ours.

Sources

  1. Ahrefs, The 100 most mentioned brands in ChatGPT: 5,516,381 US ChatGPT answers across all topics in Ahrefs' answer index, November 2026 (Nov 2026)
  2. Ahrefs, ChatGPT query fan-out study: 66.6M free, logged-out GPT-5.6 Luna responses, Aug–Sep 2026, all countries; factual subset 20,000 US prompts; citation ranks US, Mar–Sep 2026 (Oct 2026)
  3. Sigzen AI, published prices (Nov 2026)
  4. MaxAEO, AI visibility sample size and clustered runs (Jul 2026)

Questions readers ask

  • Share of answers is the percentage of AI answers in a defined set that name your brand at least once. The definition only works with its base: all answers an engine gives on every topic, the answers to your own buyer prompts, or the mentions within a tracked rival set. Each base gives a different percentage, so a report names its base beside every figure.

Keep reading

Free. Enter your domain and Sigzen AI checks how five engines answer your buyers' questions, against up to three competitors.

No account needed. Two minutes to submit; we email the full five-engine report when the check has run.