TECH & TOOLS

CHATGPT AGENCY,
WHAT AI SAYS ABOUT YOU.

More and more people ask an AI rather than a search engine. We read what ChatGPT answers about your brand and your category, and where it gets it from.

ChatGPT official site
Official site, captured chatgpt.com
ChatGPT
✓ Read by a Licter consultant

CHATGPT AT A GLANCE

What is ChatGPT?

ChatGPT is OpenAI's conversational assistant, launched in November 2022. Since late 2024 it can search the web and cite its sources. It is not a listening tool: it is a source of opinion in its own right, to be audited.

Related expertise AI listening

Related offer Social Insights

The other tools Talkwalker, Visibrain, YouScan, SoPrism, Radarly, Brandwatch, Sprinklr, Semrush, Google Trends, AnswerThePublic, GEO, Meta Ads, Google News, Social Blade, Claude, Gemini, Perplexity, Grok

HOW CHATGPT ANSWERS

What shapes ChatGPT's answer.

  • Where do its answers come from?

    Without search, ChatGPT answers from its training data, frozen at a date. With web search, open to all its users since late 2024, it reads pages live and cites its links.

  • What sets it apart

    It is the most used assistant: it is often where a customer first asks "which brand should I choose?". Its answer mixes what it learned and what it finds, and the two do not always agree.

  • What we check

    We ask each question with and without web search. The gap between the two shows what is outdated in its memory, and which pages it reads today to correct itself.

WHAT IT DOES

What ChatGPT lets you analyse.

What AI recommends

The brands cited on your category's questions, and in which order.

The sources cited

The sites and articles the answer relies on.

Errors

False or outdated information about your brand.

Change over time

The same questions asked over time, to follow the answers.

CHATGPT AGENCY

Why work with a ChatGPT agency?

ChatGPT answers your customers. A ChatGPT agency like Licter questions it the way they do, has an analyst read its answers and turns them into a decision, presented to those who decide. Here is how we work.

ChatGPT on its own

  • A question asked once, by one person
  • Answers that change from one try to the next
  • No record of the sources cited
  • No comparison with your competitors

ChatGPT with Licter

  • The same questions asked several times, every month
  • The sources cited, collected and sorted
  • Your place against your competitors
  • What to correct, and where to act

You know what the AI says about you, and why.

  1. 01

    Framing

    We start from the decision to make, not from a keyword list: the perimeter follows the question.

  2. 02

    Collection

    Queries, sources, languages and periods, set by the consultants who will read the output themselves.

  3. 03

    Reading

    An analysis crossed with our other sources, and a recommendation your teams can act on.

WHAT WE DELIVER

What we deliver with ChatGPT.

01 / 03

AI answer audit

  • Your category's key questions
  • What the AI answers
  • Your place against competitors
  • Errors to correct
02 / 03

Regular tracking

  • The same questions every month
  • How answers change
  • Newly cited brands
  • A note for communication
03 / 03

Action plan

  • Missing content
  • Sources to strengthen
  • Messages to clarify
  • The link with GEO

USE CASES

ChatGPT: the use cases where it counts.

Each card leads to the use case it contributes to.

LIMITS AND COMPLEMENTS

The limits of ChatGPT, and how we make up for them.

Limit

Answers that vary

The same question does not always get the same answer: we ask each question several times and compare.

ChatGPT What we make up for

Limit

It is not listening

ChatGPT has no access to the whole social conversation: it completes social listening, it does not replace it.

ChatGPT What we make up for

We cross it with

GEO

On your category's questions, how often you are cited.

GEO See the page

We cross it with

Claude

The brands cited on your category's questions, and in which order.

Claude See the page

We cross it with

Gemini

The brands cited on your questions, and in which order.

Gemini See the page

YOUR BRAND IN CHATGPT, IN DETAIL

Your brand in ChatGPT: an audit of its answers

Written by Antoine Khaitrine and Adrien Krebs, co-founders of Licter · updated 12 October 2026

How do I find out what ChatGPT says about my brand?

To find out what ChatGPT says about your brand, you have to ask it the questions your customers ask, not just your name. "What do you think of this brand?" matters, but "which brand should I choose for this need?" matters more: that is where it recommends, compares and rules out. A question asked only once supports no conclusion.

You then need to record what comes back: the brands cited and their order, the strengths and weaknesses attributed, the sources shown when it searches the web. This record, repeated and compared, is what lets you track your brand's mentions in ChatGPT. A single screenshot shows one possible answer, not what most of your customers see.

Where ChatGPT gets what it knows about a brand

What ChatGPT knows about a brand comes from two places. First its training data: a large body of public texts, frozen at a date. Then, when it searches the web, the pages it reads at the moment of answering. The first source explains what it thinks it knows, the second what it checks or corrects.

OpenAI distinguishes two crawlers. GPTBot collects content that may be used to train its models; OAI-SearchBot is used to show sites in ChatGPT's search. The two are set separately in a site's robots.txt file: blocking one does not block the other.

In both cases, ChatGPT does not read an official profile of your brand. It picks up what others have written: your site, the press, comparison articles, forums, reviews. A poorly documented brand will be described from very little, sometimes from a single old page nobody has updated.

Which questions to ask, and how many times

The questions to ask ChatGPT are built from your customers' questions, not from your internal keywords. There are three families: choice questions, such as "which brand for this need?", questions about the brand itself (reviews, price, reliability, commitments) and comparisons with a named competitor.

Customers' vocabulary is read elsewhere: in the searches they type, in what they post on social networks, in reviews. That is where the audit meets search listening and social listening. A question nobody asks is of no interest, even if the assistant's answer is flattering.

Each question is asked several times, then the answers are compared: what comes back on every try counts more than what appears once. The same panel is then asked again every month, with the same wording, so that the differences come from the assistant and not from the way of asking.

Tracking your competitors in the answers

Tracking your competitors in ChatGPT's answers means looking at whom it recommends when your name is not in the question. On choice questions, it usually cites a few brands, in a certain order, with a reason for each. These reasons say how it sums up your category.

The comparison covers how often each brand is cited, the order, the qualities attached to each brand and the sources used. A competitor cited more often than you rarely is by chance: it is better documented, more present in the press or in comparison articles. Monthly tracking also shows the new brands that appear.

What an audit actually delivers

An audit of ChatGPT's answers first delivers an inventory: your category's key questions, what the assistant answers, your place against competitors and the errors to correct. Each finding is tied to the answers recorded and, when they exist, to the sources the assistant cited.

It then delivers an action plan: missing content, sources to strengthen, messages to clarify. Licter does not produce this content; your teams or your agencies do. Regular tracking asks the same questions every month to see what has moved, with no promise of results: you act on the sources, and measure the effect.

An audit does not tell everything. It shows what the assistant answers, not how many people ask it these questions: to estimate that, it is crossed with search listening. And what ChatGPT says does not replace what your customers say themselves: the conversation on social networks is read through social listening.

Why ChatGPT does not always give the same answer

ChatGPT does not always give the same answer, because it writes each answer at the moment it is asked. Two tries on the same question can cite other brands, in another order, with other nuances. Wording matters too: one word more or less is sometimes enough to change the recommendation.

The answer also depends on who asks. When memory is turned on, ChatGPT takes saved information and past conversations into account to personalise what it answers. To know what a customer discovering your category sees, it is better to question the assistant with no history and no memory.

Finally, the assistant itself changes: new models, new features, new pages read. An answer recorded in spring does not say what it answers in autumn. That is why you measure a trend over several tries and several months, rather than a sentence recorded once and discussed as a verdict.

With or without web search: two different answers

With or without web search, ChatGPT does not answer the same way. Search, launched at the end of October 2024 for subscribers, was opened to logged-in free users in December of the same year. The assistant decides on its own to search when the question calls for it, and users can also trigger it themselves.

When it searches, it shows its sources, as links in the text or in a separate panel. You can then see which pages weigh on the answer. Without search, there is no source to read: you have to infer where the brand's image comes from, based on what it says and the mistakes it makes.

That is why we ask each question in both modes. The gap between the two answers shows what is outdated in its memory: an old name, a discontinued range, a price that has changed. It also shows which pages it goes to today to correct, or confirm, what it thinks it knows.

What to do about false or outdated information

False or outdated information in ChatGPT is not fixed by rewording the question. Nobody controls an AI's answer, and it cannot be changed directly. You act upstream, on the public sources that feed it, then measure whether the answer changes in the weeks and months that follow.

The first step is to find where the error comes from. When ChatGPT searches the web, its sources often tell you: an old article, a comparison never updated, a wrong listing. Without search, you trace back to the public pages that repeat the same information, and look at which ones carry authority.

You then need to correct or complete these sources: your site first, with clear and current information, then the press and reference pages. OpenAI states that a public site can appear in ChatGPT search if it does not block OAI-SearchBot: a setting to check with your technical teams. This is the ground of GEO.

ChatGPT and the other assistants

ChatGPT is not the only assistant your customers ask. Gemini relies on Google Search, Perplexity searches the web for every question and shows its sources, Claude answers first from its training data, Grok reads posts on X. On the same question, they do not always cite the same brands.

An audit limited to ChatGPT therefore gives a partial view. Depending on your market and your customers, we define with you which assistants to question, always with the same questions, to compare their answers and spot the sources they share. A source picked up by several assistants deserves to be handled first.

Sources OpenAI · Memory FAQ, OpenAI · Overview of OpenAI crawlers, TechCrunch · OpenAI brings search to all users, Search Engine Roundtable · ChatGPT search open to all (2024), OpenAI · ChatGPT search, OpenAI · Publishers and developers FAQ

FAQ

Frequently asked questions about ChatGPT.

Is Licter a ChatGPT agency?

Yes: ChatGPT is a tool anyone can use; what we bring is the method, the reading and the cross-check with other sources. Licter is not tied to its publisher; we are an independent consultancy.

Why track what an AI answers?

Because part of your customers get their information there before buying, and the answer does not come from you.

Can you change what the AI answers?

Not directly: we identify the content and sources that feed the answer, which is where you act (see GEO).

Do you send our data to the AI?

No. We ask public questions; no client data is sent.

TIME TO ACT

Looking for a ChatGPT agency?

Send us the question. If another tool answers it better, we will tell you: we use 19.