AI listening. What AI says about you.

More and more searches go through an AI assistant. We question the main models about your brand and your category, the way your customers would, and read what they recommend, cite or leave out: your AI visibility, and your reputation in their answers.

Get contacted by a consultant Or book a meeting

A call with a consultant, on your brand.

  • A consultant contacts you within 30 minutes, on weekdays from 9am to 7pm
  • Your question framed, with the right sources and tools
  • A costed proposal after the call

Noted. A consultant contacts you within 30 minutes, on weekdays from 9am to 7pm.

A consultant, not a sales team.

Trusted by 50+ organisations · 160+ projects since 2022

WHAT IT PICKS UP

What AI listening lets you hear.

The brands recommended in your category

The sources the models cite

Mistakes or outdated facts about you

USE CASES

The questions it answers

Each question leads to the use case where this listening makes the difference.

AI LISTENING, IN DETAIL

What to know before you listen to what AI says about you.

Written by Antoine Khaitrine and Adrien Krebs, co-founders of Licter · updated 19 October 2026

What is AI listening?

AI listening means questioning AI assistants the way your customers would, then reading what they answer about your brand and your category: the brands they recommend, the facts they quote, the sources they link to, and what they leave out.

It is a form of listening, like social or search listening: instead of a conversation between people, you read the summary a model draws from it. For many searches, that summary is now the first thing a customer reads.

It is also called AI visibility, or visibility in generative AI: the place a brand holds in assistants' answers, measured question by question.

Which AI assistants should you listen to?

The ones your customers use for the question that matters to you. In practice: ChatGPT, the AI-generated answers in Google (AI Overviews and AI Mode, built on Gemini), Gemini itself, Claude, Perplexity and Grok. Each has its own sources and its own way of answering, which is why a brand can be cited by one and ignored by another.

Several of these assistants now search the web before answering and cite their sources: ChatGPT with its search, Claude since 2025, Perplexity by design, and Google for its AI Overviews. What they say therefore also depends on what the web says about you today, not only on their training data.

What should you measure?

Five things, question by question. Presence: is the brand cited, and in how many answers? That is its share of answers. Rank: does it appear first or at the end of a list? Tone: is what is said favourable, neutral or critical? Accuracy: are the facts right and current? Sources: which pages does the assistant link to back its answer?

Mistakes are often the most useful finding. For example, a 2022 price quoted as current, or a discontinued range presented as available: the fix goes through the sources the model consults, not through the model itself.

The limits of AI listening

Answers are not stable: they change from one session to another, from one model version to the next, and sometimes with the user's history. A single measure says little; it is asking the same questions again, over time, that gives a reliable trend.

AI listening does not say how many people ask these questions either. For volume, it has to be crossed with search listening, which reads what people search for on Google, YouTube and Amazon.

AI listening and GEO: what is the difference?

AI listening measures and reads. GEO, for Generative Engine Optimization, acts: it tries to improve a brand's place in those answers by working on the sources the models use. The first is the starting point and the yardstick of the second: without a regular reading, you know neither where you start nor whether what you change works.

Which questions should you ask AI?

Your customers' questions, not your marketing team's. There are three families: choice questions, for example "which health insurance for a family?", where the assistant recommends brands; questions about the brand itself, its price, its products, its reputation; and comparisons with a named competitor.

Choice questions matter most: that is where a brand wins or loses a recommendation without anyone having searched for it by name. The same question is asked several times, and to each model, because answers vary from one time to the next.

Where do the answers come from?

From two places. First, the model's training data, frozen at a date, which is why an assistant can describe a brand as it was two years ago. Then, for assistants that search online, the pages they consult at the moment of answering. At Google, a page must be indexed and able to appear with a snippet to serve as a link in AI Overviews.

These assistants' crawlers are set separately: at OpenAI, the one that feeds ChatGPT search and the one that collects for training are independent. A site that blocks the first does not appear in ChatGPT search answers. Checking these settings is part of the listening.

How Licter runs an AI listening study

The framing sets the questions your customers ask and the models to question. The same questions are then put to each model, several times, and an analyst reads the answers: what they recommend, cite, leave out or get wrong. The readout ends with a plan: the sources to correct or to feed.

After a first measure, we start again every month or quarter with the same questions, so the results can be compared. Licter is an independent consultancy: we publish no assistant and do not sell visibility in their answers.

Sources Google · Aperçus IA et Mode IA, Google Search Central · Fonctionnalités d'IA, Anthropic · Claude can now search the web, Perplexity · Perplexity crawlers, OpenAI · Overview of OpenAI crawlers, arXiv · GEO: Generative Engine Optimization (2023)

METHOD

How it runs

  1. Day 0

    Framing

    The questions your customers ask, and the models to question.

  2. Day 2

    Prompting

    The same questions, put to each model.

  3. Day 5

    Reading

    What they recommend, cite, leave out or get wrong.

  4. Day 7

    Plan

    The sources to correct or to feed.

Indicative timings for a first read.

LIMITS AND COMPLEMENTS

Its limits, and what completes it.

Its limit

AI listening

How many people ask these questions: that is search listening.

AI listening What it cannot say on its own

In the offer

Social Insights

Studies on demand, read by experts.

Social Insights See the page

In the offer

Nox

AI-assisted monitoring.

Nox See the page

Frequently asked questions

Frequently asked questions

Which models do you question?

The assistants your customers use most, set with you, always with the same questions so the change can be followed over time.

Can you influence what AI answers?

Indirectly: the models rely on public sources. We identify the ones they cite, and what should be corrected or added there.

How often do you measure?

A first measure, then every month or quarter, with the same questions so it can be compared.

Is it different from SEO?

Yes. AI assistants do not rank pages, they synthesise sources. We read those syntheses, and the sources feeding them.

What do you receive at the end?

An audit of AI answers, about a week after framing: what each model recommends, cites or gets wrong on your questions, your share of answers against your competitors, and the plan of sources to correct or to feed.

How much does an audit of what AI says cost?

On quotation: the price depends on the scope (brands, markets, languages, sources), the duration and how often you receive deliverables. After a first thirty-minute call, we send a costed proposal.

Let's talk about what AI says about you.

Leave your email or phone number. A consultant contacts you within 30 minutes, on weekdays from 9am to 7pm, and tells you which model fits.

A consultant, not a sales team. Weekdays, 9am to 7pm.

Already running social listening? Request a diagnostic instead →

  • 50+ organisations served
  • 160+ projects since 2022
  • 20+ languages monitored