ChatGPT Gives Everyone Different Advice. Does Your Business Still Need AI Visibility?

TL;DR

ChatGPT, Claude, Gemini and Perplexity already remember users and tailor advice to them. But memory changes the question the assistant sends to search, not the list of companies it chooses from. AI takes its candidates from the web, reviews and training data, so a company that is not there will not be recommended to anyone. Meanwhile, "where to buy" and "near me" queries are growing: 45% of US consumers have already used AI to find a local business, up from 6% a year earlier (BrightLocal, 2026). Personalization hits the mid-market hardest, which makes AI visibility work more precise, not unnecessary.

Key facts

• Every major assistant now has memory. Claude offers it even on the free plan since March 2, 2026, and Google added it to AI Mode in January 2026.
• OpenAI says memory helps ChatGPT write a better search query. Its shopping research feature skips sites that block bots.
• 45% of US consumers have used AI to find a local business, up from 6% a year earlier (BrightLocal, 1,002 respondents, February 2026).
• The share of ChatGPT prompts that look like ordinary search queries grew from 18.9% to 34.9% (Semrush, October 2025 to February 2026).
• Category leaders keep about 80% of their recommendations across buyer types, while mid-market brands lose up to 75% (arXiv preprint, 2,000 runs, May 2026).
• Personalized Google changed only 11.7% of search results (Northeastern University, 2013).

Does AI really give everyone different advice now?

Yes, and it is no longer an experiment. ChatGPT remembers past conversations, saved facts and connected apps such as Gmail. Claude has memory even on the free plan since March 2, 2026. Perplexity launched assistants with memory in November 2025 and explicitly lists choosing running shoes, trips and gifts among its examples.

Google personalizes search too. In January 2026 the company added Personal Intelligence to AI Mode: an answer can take the user’s email and photos into account if they allow it. So if a customer asks which car to buy, two people really will get different advice. The assistant will factor in that one drives three kids around and the other commutes on the motorway every day.

What exactly does personalization change in an answer?

The question, not the catalogue. OpenAI says so directly in its help center: when memory is on, ChatGPT may use details it knows about a person to write a more useful search query. The company’s example is location and dietary preferences when searching for a restaurant.

Google describes AI Mode in a similar way. A question is broken into subtopics, and the system runs many searches at once on the user’s behalf. This is called query fan-out. Personal context shapes which sub-queries the assistant sends. But the answers to them still come from the shared search index. None of the four major developers describes personalization as a separate list of companies for each user.

Where does AI get the companies it recommends?

From the same open web. Before recommending anything, the assistant gathers candidates: pages, profiles, reviews, roundups, product feeds. Only then does memory help pick the ones that suit a specific person. OpenAI says shopping research results are organic and based on publicly available retail sites. Sites that block automated access are skipped, and the data comes from other sites instead.

The second source is the model’s training data. Brands that are often mentioned in reviews and comparisons are the ones the model “remembers” first. The takeaway for a business is simple: if your company is not in the candidate set, even the most precise personalization will not help. An assistant cannot recommend what it does not know about. How to remove the technical barriers that keep you out of that set is covered in our step-by-step checklist for getting into ChatGPT answers.

Which queries does personalization barely touch?

The ones where the answer is tied to a place, product availability or a specific company. If someone is looking for a dentist in Camden, memory won’t move a clinic in Croydon any closer. If they ask where to buy a washing machine with same-day delivery, stock and price decide, not chat history. Personalization is strongest in “what suits me” questions, but even there AI takes its candidates from the web.

The fewer the options, the more stable the answer. In a SparkToro study (January 2026, nearly 3,000 runs) recommendations were most consistent where the choice was limited. There are fewer than a dozen Volvo dealerships in Los Angeles but thousands of new science fiction novels, and in the first case assistants named the same ones much more often.

How many queries like that are there?

A lot, and their share is growing. A BrightLocal survey (February 2026, 1,002 US adults) found that 45% of consumers had already asked ChatGPT or another AI about a local business. A year earlier the figure was 6%. According to Semrush, the share of ChatGPT prompts phrased like ordinary search grew from 18.9% in October 2025 to 34.9% in February 2026. Most of them are navigational or transactional.

Commercial intent is already visible. Profound analyzed more than 50 million ChatGPT prompts (June 2025) and counted 9.5% commercial and 6.1% transactional. Semrush and Profound sell AI visibility tracking tools, as do we, so read their data with that in mind. In Ukraine, for example, 27% of people use AI regularly, up from 17% a year earlier (KIIS, January 2026). The audience asking “where” and “who” keeps growing. How AI answers are changing who gets the clicks in Google is covered in our piece on the AI Overviews update.

Didn’t the same thing happen with Google in 2009?

It did, and SEO survived. In 2009 Google switched on personalized search for everyone, and marketers started talking about the end of optimization. Measurements by Northeastern University researchers showed otherwise: only 11.7% of Google results turned out to be personalized. What changed them was mainly being logged in and location. A 2019 study in EPJ Data Science reached a similar conclusion: there is very little room for personalization in search results.

Personalization reshuffled results from a shared index. It did not create a separate internet for each person. The difference in 2026 is that an assistant’s memory knows far more about a person than Google did in 2009. So the analogy points the way, but it does not guarantee everything will repeat in the same way.

Where does personalization really hurt a business?

In the mid-market. In an arXiv preprint (May 2026) researchers ran 2,000 tests: 10 buyer types, 8 prompts, models from OpenAI and Anthropic. A short description of who was asking was added to each prompt. Category leaders kept about 80% of their recommendations regardless of the buyer. Mid-market brands swapped up to 75% of the list as soon as the buyer type changed.

This data has limits. It is a preprint without peer review, and the buyers in it are synthetic. No one has yet published a study comparing real accounts with memory switched on. The takeaway for the mid-market: simply being online is not enough. People need to write about you in the context of your specific customers. More data on why AI recommends different brands to two different customers is in our piece on GEO trends in fall 2026.

The second limit is advertising. OpenAI launched ads in ChatGPT on February 9, 2026 and says they do not influence answers. Ads Manager is available in 63 countries, so check whether yours is on the list. Ukraine, for example, is not, and there organic visibility is currently the only way into ChatGPT answers.

Where to start: four steps

1. List your queries. Write down 10-15 queries people use to find you and sort them by the types in the table above. Pay the most attention to local, price and company-specific queries.
2. Check the answers several times. Run each query in ChatGPT, Gemini and Perplexity 3-5 times, in a temporary chat without memory and from different accounts. Record how often you are named, not in which position.
3. Open up to bots. Check robots.txt, your Google Maps profile and whether your price is visible on the page without JavaScript.
4. Describe who you are a fit for. For each customer type, find 2-3 platforms where people could write about you: roundups, reviews, industry media. The Haivoronskyi Framework shows how these external factors fit into one picture.

Free AI visibility mini-audit. We run your key queries through ChatGPT, Perplexity, Gemini, Claude and Google AI and show you who gets recommended instead of you. Get a mini-audit

Key takeaways

• Personalization changes the question sent to search, not the list of companies AI chooses from.
• The assistant takes its candidates from the web, reviews, product feeds and training data.
• Queries about place, price and a specific company are the least personalized, and there are more of them every year.
• Personalized Google in 2009 did not kill SEO because it reshuffled results from a shared index.
• Personalization hits the mid-market hardest: up to 75% of recommendations change with the buyer.
• ChatGPT ads are available in only 63 countries, so in many markets organic visibility remains the main channel.
• Measure visibility by how often you are mentioned, separately for each customer type.