When AI omits your brand, a competitor may appear

In short: a missing brand is not a neutral blank in an AI answer. In our set of 3,002 answers from real audits, 37% of answers that did not contain the audited brand still named at least one competitor. That is an observed pattern, not evidence of a model’s intent or a promise about any market.

The 42%, 37% and 4.6 figures come from the July 2026 study, not from a new collection. The average position of around 2 comes from the June 2026 barometer.

What the 3,002 answers show

We reviewed 3,002 answers generated in real audits. In each answer, we checked for the audited brand and for competitors in the generated text. The practical question is simple: when the brand is not named, which other actors are visible?

The first measured result covers the full set: 42% of answers name at least one competitor. The second result isolates answers where the audited brand is absent: 37% of that subset still includes a competitor. Neither number is a market average or a rule for every company.

When competitors are detected, the mean count is 4.6. The observed situation is therefore not always a two-brand contest. Some answers list several options or alternatives. The measurement records named entities, not the force of a recommendation or a reader’s eventual choice.

Why absence deserves attention

Rank matters only after a brand enters the answer. The June 2026 barometer measures an average position of around 2 for cited brands. That is a finding from that barometer, not a placement guarantee for a specific brand. The blind spot is the answer where the brand does not appear at all.

An AI can answer a comparison, recommendation or explanatory question without naming every possible actor. A competitor’s presence therefore does not prove preferential treatment, commercial intent or a deliberate exclusion. For an absent brand, it does show that the same exchange included other options.

The useful diagnosis is an observed presence gap: the brand is not detected while one or more competitors are. That gap can then be examined by prompt, model, language and measurement date.

How to read the three figures

42% is the share of all analysed answers that contain at least one competitor. It does not mean that 42% of people select a competitor, nor that every such answer is a recommendation.

37% applies only to answers without the audited brand. More than one third of those answers contain a competitor. Keeping that condition matters: it cannot be compared directly with a general visibility rate unless the scope is identical.

4.6 is the average number of competitors within answers that name at least one. It is not the number of competitors in a market, nor the average size of a panel. It indicates that observed competitive answers can be crowded.

Measuring share of voice without overclaiming

A useful measurement keeps the source answer, the model, the prompt and the detected entities. This makes it possible to separate an isolated absence from a repeated pattern. A global percentage alone can hide large differences between prompts and engines.

It is equally important to separate observation from explanation. Missing documentation, an often-reused third-party source, an overly broad query or model variation may be hypotheses worth testing. The data here does not establish any of those causes on its own.

Our AI visibility barometer provides the broader reading context. For the comparison with other actors, see our guide to competitive share of voice.

The Observatory and the AGS methodology set out the measurement frame. Our study of disagreement across models adds the model-by-model view.

A practical reading example

Imagine the same question is sent to several models. One names the brand alongside two competitors; another omits it while naming four. That result does not support a claim that the second model prefers the four. It tells you to inspect the wording, available evidence and whether the pattern recurs.

Another case is a brand that appears only in prompts already containing its name. That can differ sharply from a spontaneous mention in a sector question. Both observations matter, but they answer different questions and should not be combined into one score without explanation.

A comparison becomes more defensible when it uses the same prompt set, period and entity standard for the brand and competitors. Counting competitors with a looser rule would distort the gap.

What can be improved without making promises

Editorial work begins with the prompts where a presence gap is observed. It may reveal an unclear definition, an undocumented use case, a missing explanatory page or important facts scattered across the site. The objective is to make information understandable and verifiable, not to force an answer.

Core brand information should remain consistent across the pages and sources readers can check. Identity details, dated evidence, product descriptions and direct answers reduce ambiguity. They do not guarantee visibility, inclusion or citation.

Competitive analysis also needs honesty. If several competitors recur in the same answers, examine the questions they cover and the evidence they make available. Repeating their language without adding substance does not close the underlying gap.

Why models must stay separate

Models do not mechanically produce the same answer. A brand can appear in one system and be absent in another, or be described with different levels of precision. That variation is why model-by-model records are more useful than an undifferentiated total.

Every audit is a snapshot of a prompt set, models and a date. It should not be treated as permanent proof about an entire market. It also does not predict what a model will answer tomorrow.

Tracking becomes more useful when conditions remain comparable. You can then observe change, while avoiding automatic claims that a single action caused it when other variables also changed.

Limits of this study

The brands in this set chose to be audited. The sample is not representative of every company, sector or country. The percentages should not be projected onto a wider population.

Entity detection depends on extraction quality and name resolution. A brand may appear under a variant, similar names can be ambiguous, and a competitor can be mentioned without being recommended. We count detected presence.

Models and outputs also change. These figures describe the analysed scope; they do not guarantee how an AI will respond in a later situation or what any reader will retain.

The question worth keeping

Do not ask only where the brand ranks when it is cited. Ask which relevant answers omit it and which actors are present in those same answers.

A rigorous answer preserves context and keeps measurement separate from hypotheses. That is what turns share of voice into a diagnostic tool rather than an isolated headline number.

What a careful next review looks like

Start with a small, stable set of questions that represent real decisions a prospective customer might make. Record the exact wording rather than relying on a remembered paraphrase. For every answer, retain the date, the model, the brand names detected and the source text. This creates an auditable baseline. It also prevents a vivid single answer from being mistaken for a recurring market pattern.

Then group answers by intent. A question asking for a shortlist, a question asking for a definition and a question that already names the brand have different expectations. Review the absence rate inside each group before drawing an overall conclusion. If a gap appears only in one intent, the response is not necessarily more pages; it may be clearer evidence or a more precise explanation for that intent.

Finally, revisit the same question set under comparable conditions. A later difference can be recorded as a change in the observed answers. It should not be described as a ranking gain, a citation gain or proof of causality without additional evidence. This restraint keeps the measurement useful for decisions.

Questions and answers

What happens when an AI doesn't cite my brand?

In more than a third of cases (37%), a competitor appears in your place. Across all analysed answers, 42% cite at least one competitor, with 4.6 rivals named on average when there are any.

Should I worry about rank or first about appearing?

First about appearing. Once cited, a brand's average position is good (around 2). The real risk is absence: the AI then confidently recommends a competitor to your customer.

How many competitors do AIs cite on average?

In answers that name competitors, there are 4.6 on average — rarely just one. AI visibility is therefore a dense competitive field.

How do I measure the competitors recommended in my place?

A share-of-voice audit detects, answer by answer and model by model, the presence of your brand and that of your competitors, then computes the gap. That is what we do across 50+ models.

About the author

Davy Abderrahman

Founder of

As a specialist in visibility in AI answers (AEO/GEO/LLMO), I help agencies and consultants measure and analyze their clients’ presence on ChatGPT, Claude, Gemini, Perplexity and other AI answer engines.

AEO GEO LLMO AI visibility AI audits
How do your brands appear in AI answers?

Agencies, consultants and in-house teams: compare the answers generated for your own scenarios, languages and selected models. Our infrastructure, your scenarios, your expertise.

Was this article helpful?

- (0 votes)