Does Asking AI About Your Brand Every Day Train It to Recommend You?

How do assistants actually produce answers?

An answer comes from two places:

• The model's training, a snapshot of what it learned from its training data, fixed until a new version is released.
• What it retrieves at the time, from web search or connected tools, for questions where it searches.

Neither is updated by your daily chats. The model does not rewrite itself after each conversation. Retrieval depends on search indexes, which reflect published web pages, not chat history.

Personal features sit on top. ChatGPT's memory and custom instructions, for example, shape answers for the user who set them. That is why a founder who asks about their own company every day may start seeing it mentioned more. The assistant has learned what that user cares about, not what is true for everyone else. Testing with memory on gives a misleading picture for the same reason.

Can repeatedly telling an LLM “you forgot this website” make it cite the site?

In that conversation, yes. For anyone else, no.

Language models trained with human feedback tend to agree with their users. Anthropic researchers found this tendency, called sycophancy, consistently across five leading assistants in a 2023 study. OpenAI rolled back a ChatGPT update in April 2025 after it made the model noticeably more sycophantic. So when you tell a model "you forgot [site]", it usually apologises and adds the site. That tells you about the model's eagerness to please, not about its view of your site. Start a new chat, or ask from another account, and the site is gone again.

Repeating it daily does not accumulate. Each conversation starts from the same model and the same search results.

Could my conversations end up in training data?

Possibly, but not in a way that helps:

• Consumer ChatGPT conversations may be used to improve future models unless the user turns that setting off. Business accounts and API use are not used for training by default.
• Temporary chats in ChatGPT are not used for training, are not saved to history and do not create memories. So "asking in incognito every day" rules out even the small chance of training use.
• Anthropic's consumer apps let users choose whether their chats are used for training. Its commercial and API services are excluded.
• Feedback buttons can matter more than chat text. OpenAI says that giving feedback on a response may lead to the whole conversation being used for training.

Even when conversations are used, they are one input among an enormous volume. Training data is filtered and processed, and new models are released months apart. Repeated prompts from one company, often phrased the same way, are unlikely to survive filtering as meaningful signal. There is no evidence they shift recommendations.

Why does it sometimes seem to work?

People who try the daily method often report success. There are usually ordinary explanations:

• Memory and personalisation. The assistant has learned your interest in the brand and brings it up for you.
• Conversation carry-over. Asking in the same thread keeps your earlier corrections in context.
• Real changes elsewhere. A new review, article or page may have been published in the same weeks, and search-based answers picked it up.
• Natural variation. Answers differ from run to run. A few good runs in a row can look like a trend.
• Model updates. A new model version may simply know more about your category.

The way to tell is a controlled check. Ask the same questions from a clean account with memory off, several times, and compare with a period before the method started. If the brand appears only in your own sessions, the method did not change anything that matters.

Is automating the prompts a good idea?

No. OpenAI's terms of use prohibit automatically or programmatically extracting data or output from its services. Scripts that hammer consumer chat interfaces with the same brand questions risk account suspension, and they achieve nothing the manual version does not. If you need to measure visibility at scale, use the official APIs within their terms, or a monitoring tool that does, and treat it as measurement, not influence.

What does change what AI says about you?

The sources assistants draw on:

• Your own pages: clear, specific, current facts about what you offer, for whom, at what price.
• Independent coverage: articles, reviews, comparisons and expert mentions that name you in context.
• Review platforms and directories your category's answers cite.
• Accurate profiles on business listings, marketplaces and professional networks.
• Crawler access: making sure AI search crawlers can reach your pages.

How assistants choose brands in the first place is covered in how AI systems decide which brands to recommend. Why they keep naming the same ones is covered in why AI assistants keep recommending the same brands.

How should you ask AI about your brand, then?

As measurement, done carefully:

1. Use clean sessions: memory off, no custom instructions, logged out or on a fresh account, or through an API.
2. Use a fixed set of real buyer questions, mostly without your brand name, since questions that name you inflate the results.
3. Run each question several times and across several assistants, because answers vary.
4. Record mentions, positions, descriptions and cited sources.
5. Repeat monthly, not daily. Changes in sources take weeks to show.

How to find prompts where your brand appears is covered in how to find prompts that mention your brand on AI platforms.

What might an illustrative case look like?

Consider an illustrative founder of a garden design studio who asked ChatGPT every morning for "the best garden designers in [town]" and corrected it when their studio was missing. After a few weeks, the studio appeared almost every time, and the founder believed the method had worked. A colleague then ran the same question from a fresh account with memory off. The studio was absent, and the answer cited a local directory where it had two reviews and a home improvement article that listed five competitors. The founder stopped the daily ritual, started asking every client for an honest review in their own words on that directory, which allows businesses to request reviews, and contacted the article's author with a new project. The studio began appearing in clean-session answers about two months later.

What do people ask about influencing AI through prompts?

Does giving a thumbs-up to answers that mention us help?

Not measurably. Feedback may be used in training, but one company's ratings are a negligible share and are likely filtered.

Can we add our brand to ChatGPT's memory for all users?

No. Memory applies only to the account that saved it.

Do Custom GPTs or projects that mention our brand help?

They shape answers inside that GPT or project. They do not change the main assistant's answers for others, and OpenAI plans to retire Custom GPTs on December 11, 2026, as covered in whether Custom GPTs influence public ChatGPT answers.

Is there any legitimate way to tell AI companies about our business?

Yes: publish accurate information they can find, use official merchant and publisher programmes where they exist, and report factual errors through their feedback channels.