Why one ChatGPT mention now keeps recommending you to the same customer
OpenAI's June 4 ChatGPT memory upgrade hit 82.8 percent factual recall. The first AI mention now keeps recommending you to the same customer.
Co-Founder, Vyzz
ChatGPT now remembers which business it recommended to which customer
When a customer asks ChatGPT for "best pediatric dentist near me" and ChatGPT names three local practices in its answer, that customer now carries a piece of the answer with them in their ChatGPT account. The same customer asks a related question a week later. The earlier names get pulled forward into the new answer. A practice named in the first answer is likelier to be named again. Practices missed on the first ask tend to stay out of the loop for that customer.
OpenAI rolled out the memory system that makes this possible on June 4. The launch post calls it Dreaming V3. The number worth knowing for an operator: factual recall on the new system reached 82.8 percent. The 2024 saved-memory generation had hit 41.5 percent. The 2025 generation had hit 67.9 percent. ChatGPT correctly remembered user preferences 71.3 percent of the time in OpenAI's evaluation. Time-sensitive accuracy reached 75.1 percent. The compute cost of running the dreaming process at scale dropped roughly 5 times. That is why OpenAI is rolling the memory feature out to the free tier in the coming weeks.
What the new memory feature actually does
The new memory feature replaces the old saved-memory list. A background process now reads across years of past conversations. It updates what ChatGPT remembers about a user on its own. OpenAI says the system updates itself as the conversation goes on. Memory updates happen in the background. What ChatGPT remembers about a user updates automatically as their life or business changes.
For the customer, the change is quiet. ChatGPT continues to answer questions. The customer rarely notices ChatGPT's memory working in the background. Inside the account, ChatGPT now keeps a customer-specific record of what it has said and what the customer has asked. The customer's next conversation starts with that record loaded in.
For a service-business operator, the change matters a lot. Since the June 4 update, follow-up questions are anchored to whatever ChatGPT told the customer the first time. The competitor named in the morning answer now keeps an advantage in the same customer's afternoon follow-up. That advantage compounds over the customer's later questions.
The customer base is about to get larger
The June 4 rollout reached Plus and Pro subscribers in the United States first. The OpenAI launch post says the rollout reaches Free and Go users in the coming weeks. International users follow after that. Coverage from Dataconomy confirms the same schedule. Plus and Pro users get roughly twice the memory capacity of the prior generation. Memory sources are visible to the user across all consumer plans. The customer can see which past conversations fed a given recommendation.
The free-tier rollout matters more for SMBs than the Plus-tier rollout. Free ChatGPT users far outnumber Plus subscribers. Free users are the account type most service-business customers actually use to ask "best dentist for kids near me" or "do I need a new water heater." The compounding effect of an early mention is about to reach the broader customer base within a matter of weeks.
The work shifts toward the first question a customer asks
If the first ChatGPT mention of your business now pulls forward into every related question that same customer asks, the best work for an SMB owner shifts toward the very first question. The right question is the one a brand-new customer would ask ChatGPT before they have ever heard of you. That is the question your visibility now hinges on.
Test this in 20 minutes. Open ChatGPT in a private window. Ask the question a first-time customer would ask before booking your category. Write down the three or four business names that come back. Then ask a follow-up the way a real prospect would. Write down the names that come back the second time. Repeat in Claude and Perplexity. The customer base does not all live on ChatGPT.
Those first-time-customer questions are your AI visibility set. The names that come back on the first ask are the names ChatGPT now stores in that customer's account.
Why the memory upgrade is harder to reverse than a regular ranking drop
A regular search-ranking drop is reversible. A Google ranking change moves the same customer's results for everyone who runs the same query, and a recovery shows up across the user base. The new ChatGPT memory feature works on one customer's private account. The memory record sits there and travels with that customer. Reversing a single early loss against a competitor for a single customer is much harder than reversing a category-wide ranking drop.
OpenAI's launch post calls this "per-conversation memory updates." Each conversation a customer has with ChatGPT feeds that customer's record. The record is private to that account. Your competitor's early advantage with one customer does not show up in your dashboard. Recovery has to happen for that customer specifically.
For an SMB, getting cited the first time has always mattered. It now matters more. The cost of being missed on the first ask compounds across the same customer's later questions. There is no easy reset.
What to fix on your end this week
Three moves this week shift what ChatGPT says about your business the first time a real customer asks.
First, rewrite your Google Business Profile description. Make it answer the question a first-time customer would ask before booking. Most service businesses describe what they sell. The version that gets picked up writes the question first, then answers it in the customer's words.
Second, name the three customer questions your front desk hears every week. Make sure your website's pages name those questions. Answer them in plain language in the first paragraph. ChatGPT and the other assistants pull confirmation text from the first paragraph of the page; a buried FAQ at the bottom rarely gets read.
Third, run the first-time-customer test described above on a weekly basis. The set of names that come back will shift as the assistants update their sources. A weekly check catches a drop in the first answer before a competitor's advantage locks in for that customer.
You can do this loop in 30 minutes per week. Open ChatGPT in a private window, ask the question your front desk hears every Monday morning, write down the names, and compare to what came back the week before. If your name dropped, the rewrite work this week is on the page or profile that lost the citation. That is the work that matters now.
The June 4 memory rollout turns the first ChatGPT answer about your business into the start of a per-customer record that grows with every later question. The names in that first answer are the names ChatGPT pulls forward across the rest of the customer's questions for months. The hour you spend this week on the first-time-customer question pays back across every follow-up the same customer asks.