# Why ChatGPT tells you one thing and your customers another | Illia BizTech

> Your own account remembers what you ask it. The unfiltered version of your firm shows up only in a temporary chat started without personalisation.

Source: https://illia-biztech.de/en/forum/chatgpt-lobt-uns-kunden-bekommen-offenbar-andere-antworten

1. [Home](/en)
2. [Forum](/en/forum)
3. ChatGPT praises us, customers apparently get other answers

# ChatGPT praises us, customers apparently get other answers

In my own account it all sounds good. A customer read me something else.

* ##### Asked by  
TiredCircuit714
* ##### Asked on  
September 20, 2026

I ask ChatGPT about our firm regularly, and the answer comes back friendly every time. Last week a customer read me, over the phone, the answer he had got when he asked the same thing. His was much drier and only put us third.

We are a small electrical firm. How do two people asking the same question end up with different answers?

#### Answers

![Portrait of Illia Belichenko, technical specialist at Illia BizTech](/_ipx/f_webp,q_75/https://cms.illia-biztech.de/uploads/small_Illia_Belichenko_author_81f88864f0.png)

##### Illia Belichenko · Founder & Technical Lead

✓ Accepted answer September 23, 2026 

The two answers came out of the same product and out of two different histories. OpenAI's help pages list several sources that feed into later answers. Two of them matter here. One is a memory feature that stores details from earlier conversations, and the other is your own custom instructions. Ask about your own firm for months and the account builds up material about it. The answer you then read as proof is shaped by that material.

None of that is stored for your customer. He starts from zero, with no history and no saved details, and sees the version that counts before the first call.

You can look at that unfiltered version yourself. OpenAI calls the feature temporary chat. Whether it is personalised is decided when you open it and cannot be changed afterwards. The non-personalised option runs with no memory, no custom instructions and no history. Ask your question there the way a customer would ask it, and never say your company name.

One run is not enough, because a single answer can come down to chance. That is why the [AI visibility report](https://illia-biztech.de/en/ki-sichtbarkeitsbericht) holds several conversations and puts every transcript on the table.

##### 

#### [The booking button on our Google profile disappeared overnight, without us changing anything](/en/forum/der-buchen-button-an-unserem-google-profil-ist-uber-nacht-verschwunden-ohne-dass-wir-etwas-geandert-haben)

November 13, 2026

[](/en/forum/der-buchen-button-an-unserem-google-profil-ist-uber-nacht-verschwunden-ohne-dass-wir-etwas-geandert-haben)

#### [A second listing with a competitor's phone number shows up under our practice's name on Google](/en/forum/unter-dem-namen-unserer-praxis-taucht-bei-google-ein-zweiter-eintrag-mit-einer-fremden-telefonnummer-auf)

November 13, 2026

[](/en/forum/unter-dem-namen-unserer-praxis-taucht-bei-google-ein-zweiter-eintrag-mit-einer-fremden-telefonnummer-auf)

#### [We put real effort into replies on TripAdvisor and Booking.com, but our Google review count in Markkleeberg doesn't budge](/en/forum/wir-pflegen-unsere-antworten-flei-ig-auf-tripadvisor-und-booking-com-unsere-google-bewertungszahl-bei-markkleeberg-bewegt-sich-trotzdem-nicht)

November 12, 2026

[](/en/forum/wir-pflegen-unsere-antworten-flei-ig-auf-tripadvisor-und-booking-com-unsere-google-bewertungszahl-bei-markkleeberg-bewegt-sich-trotzdem-nicht)

#### [ChatGPT and Perplexity show different information about our business in Ingolstadt, Gemini gets it all right, why?](/en/forum/bei-chatgpt-und-perplexity-stehen-unterschiedliche-angaben-zu-unserem-betrieb-in-ingolstadt-bei-gemini-stimmt-alles-warum)

November 12, 2026

[](/en/forum/bei-chatgpt-und-perplexity-stehen-unterschiedliche-angaben-zu-unserem-betrieb-in-ingolstadt-bei-gemini-stimmt-alles-warum)

#### [Our ad shows a phone number we never entered anywhere](/en/forum/in-unserer-anzeige-steht-eine-telefonnummer-die-wir-nirgends-eingegeben-haben)

November 11, 2026

[](/en/forum/in-unserer-anzeige-steht-eine-telefonnummer-die-wir-nirgends-eingegeben-haben)

**Share** 

**Topics** 
* KI-Sichtbarkeit
* ChatGPT

An owner asks ChatGPT about his firm in his own account and reads something friendly back. OpenAI lists several sources that feed into later answers, among them a memory function and the custom instructions the owner set himself. A customer without that history sees something else.

Whether a temporary chat is personalised is settled when you open it, and the option without it runs with no memory and no history. You ask the question the way a customer would and leave your company name out of it. One answer is still one sample, because the same question in two separate chats can come back differently.