How do you measure how often ChatGPT recommends a business?

Measure it by asking the same set of realistic customer questions repeatedly across several AI assistants and recording whether the business is named in each answer. Divide the answers that name the business by the total number tested. Expressed as a percentage, the result behaves more like market share than a raw mention count.

AiDisco calls the corrected measure Share of AI Answer: how often AI names a business across the questions customers actually ask, counting only questions where nobody mentioned that business first. It is market share translated to the answer layer.

The correction matters because “AI share of voice” is often used loosely. Some measurements include branded questions such as “Is Company A a good choice?” or “What does Company A sell?” Those prompts already place the company in the answer. Counting the resulting mention makes the figure look healthier without proving that the assistant would have selected the company independently.

A basic calculation is:

Share of AI Answer = non-branded answer runs naming the business ÷ all non-branded answer runs tested × 100

Suppose a company tests 50 customer questions across three assistants and repeats every question twice. That creates 300 answer runs. If the company is named in 54, its result is 18%. The percentage says the business entered 18% of the tested answer opportunities; it does not say 18% of customers bought from it.

The questions must remain stable enough for comparisons to be meaningful. The separate guide to which questions to track explains how the set is built. This page measures performance against that set rather than deciding what belongs in it.

Assistants are non-deterministic. The figure will move between runs because the same question can return a different answer. One run is a sample, not a result. Only a trend across repeated runs is meaningful.

How do you tell if competitors are named more often in AI answers?

Test the same unnamed customer questions for every company in the comparison and count each appearance using the same rules. Compare the percentage of answer runs naming the business with the percentages naming its competitors. Do not use separate question sets, different assistants or different testing dates, because those differences can create a false gap.

The cleanest comparison uses one shared matrix:

Question	Assistant	Run	Business named?	Competitor A named?	Competitor B named?
Customer question 1	ChatGPT	1	Yes	Yes	No
Customer question 1	ChatGPT	2	No	Yes	No
Customer question 1	Gemini	1	Yes	No	Yes

Each company can appear in the same answer, so the percentages do not need to add to 100%. AI assistants often recommend shortlists rather than selecting one winner. A business with 24% and a competitor with 37% is not necessarily losing every answer by 13 percentage points; the two may appear together in some runs.

Compare more than the totals. Break the results down by question cluster, assistant and repeated run. A competitor may lead because it dominates one large use case, while the business is competitive elsewhere. That distinction shows whether the gap is broad or concentrated.

A mention count also does not explain the cause. The separate guide to why a website isn’t showing in AI search covers the evidence, authority, clarity and retrieval problems that may sit behind weak appearance rates.

What should a business measure to know if it shows up in AI answers?

Measure the percentage of unnamed customer questions that produce a mention, the competitors appearing in the same answers, differences between assistants, differences between question groups and movement across repeated testing periods. Keep the underlying counts visible. A percentage without the number of questions, assistants and runs behind it can make a small sample look more reliable than it is.

At minimum, record:

• the exact question asked;
• whether the question named any company;
• the assistant and model used;
• the date and run number;
• whether the business appeared;
• which competitors appeared;
• whether any company was named at all; and
• the total number of qualifying answer runs.

Report branded questions separately. They can reveal what an assistant says after the company is named, but they should not increase the discovery measure. Comparison questions should also be separated because both companies were supplied in advance.

The overall percentage answers whether the business enters relevant answers. It should not be stretched into every other question about AI performance. Measurement by buying stage belongs in visibility at the point of buying, where early research questions are separated from questions asked near a decision.

The result also does not diagnose why the company is absent. A low percentage may reflect weak category relevance, unclear positioning, missing third-party evidence, poor crawlability or competitors with stronger supporting sources. Diagnosis should follow measurement rather than being guessed from the score alone.

What does a good rate of being mentioned by ChatGPT look like?

There is no published benchmark that defines a good percentage for every category. Judge the result against three things instead: the competitor set measured under the same conditions, the business’s own previous runs, and the proportion of tested answers in which the assistant names any company at all. These comparisons are more defensible than an invented universal target.

Start with the competitor gap. If the business appears in 22% of answer runs and the leading competitor appears in 24%, the position is different from a market where the leader appears in 68%. The same 22% can represent near-parity in one category and substantial underexposure in another.

Next, compare the result with the company’s own baseline. Movement from 8% to 16% across several comparable testing periods is meaningful even if the business has not yet caught the leader. Movement from 16% to 18% in one run may be ordinary variation.

Finally, consider how often any company is named. Some questions produce general advice without recommending a business. Suppose a company appears in 18 of 100 answer runs, while any company appears in only 40. The formal result remains 18%, but the company captured 18 of the 40 available company-naming occasions. That context is useful, although it should not replace the main percentage.

A good result therefore has no single fixed threshold. It means the company is competitive against realistic rivals, improving against a stable baseline and appearing in a reasonable share of the answers where assistants name companies. The testing conditions and sample size must always be published beside the number.

Why does a company lose customers when AI recommends someone else?

Someone who asks an assistant for options and never sees the company’s name has no reason to look it up. What is lost at that moment is a place on the shortlist, not a tracked click or a confirmed sale. The honest description is exposure lost, because an AI answer is only one input into a customer’s decision.

AI recommendations can shape which names a customer researches next, but they do not prove which answer caused a purchase. The person may already know several companies, search elsewhere, ask colleagues, read reviews or buy later through another channel. A missing mention therefore cannot be converted directly into a precise number of lost customers.

The commercial chain is better expressed as:

No appearance → no AI-generated consideration → lower chance of further research → possible lost opportunity

That chain matters because consideration usually comes before comparison. A company cannot be evaluated if it never enters the set of options. However, “possible lost opportunity” is not the same as attributed revenue loss.

Teams can connect the visibility trend with branded searches, direct traffic, assisted conversions, sales-call mentions and lead quality, but these remain supporting signals unless the customer journey is explicitly tracked. The fuller question of commercial payback belongs in whether AI SEO works.

This distinction protects the measurement from overclaiming. Share of AI Answer measures exposure in relevant answers. It does not claim that every appearance creates a customer or that every absence loses one.

How do you tell if a business is named more often than before?

Repeat the same test under comparable conditions and compare trends rather than isolated percentages. Keep the core question set, assistants, repetition count and scoring rules stable. Record changes by total result and by question group. An increase that persists across several periods is stronger evidence than a one-off rise in a single run.

Use a simple comparison table:

Period	Qualifying answer runs	Business mentions	Result	Change
Baseline	300	36	12%	—
Review 1	300	45	15%	+3 points
Review 2	300	51	17%	+2 points

These figures are illustrative, not benchmarks. The important feature is the consistent test design.

Check whether growth is broad or concentrated. A total increase may come from one assistant, one cluster or one unusually favourable run. Durable improvement is more convincing when it appears across multiple relevant question groups and survives repeated testing.

When the question set changes, report the change. New customer questions may make the later result more representative, but they also break a direct comparison with the baseline. Keep a stable core set for trend reporting and test newly discovered questions in a separate expansion set until enough history exists.

The ongoing service for running these checks belongs under AI visibility monitoring. The reporting interface and presentation of results belong on the AI visibility dashboard. This page defines the measure and the comparison logic rather than the service or software used to operate it.

A trend should always include the sample size, dates, assistants and number of repeated runs. Without those details, “mentions increased” is not independently interpretable.