Skip to content
DataMentari
All insights

Is Your Store in the AI Answer? A Repeatable Way to Check

Ask ChatGPT, Gemini and Perplexity one fixed prompt three times each, signed out, with your city named. Then check the named stores against what is true.

In this article
A single wooden chair lit by a spotlight on an otherwise dark stage
Photo: Allec Gomes on Unsplash

In the early hours of 20 November we asked ChatGPT where to buy an office chair. We were signed out, in a private Chrome window, on a connection in Malaysia. The question was “I want to buy an office chair online from a store in Kuala Lumpur, Malaysia. Which stores would you recommend? Please name specific stores and their websites.” We asked three times, in three fresh windows. The same two stores came out on top every time. Merryfair first, Ergoworks second.

Gemini got the same question three times and gave us a different list each time. Its top pick was SPACE My twice, then Alterseat. Only two stores, Alterseat and OfficePro, made all three Gemini answers. Merryfair, ChatGPT’s first choice in every run, never appeared in Gemini. Perplexity would not answer at all. Both times it stopped at “Sign in to continue using Perplexity”.

Sell chairs in KL and check only ChatGPT, and the shortlist looks settled. Check only Gemini and it looks like a lottery. One question typed once into one tool tells you very little.

And people are asking. The e-Conomy SEA 2026 report from Google, Temasek and Bain says “69% of SEA-6 consumers use AI to compare product options”. That comes from a July 2026 Google/YouGov survey of 9,006 people who already use AI, across six markets that include Malaysia. The same survey found that “More than 90% of users conduct further research or consult reviews following an AI recommendation”. So the answer builds the shortlist, and the shopper checks the shortlist. A store that is left out, or placed in the wrong city, may never get that visit.

How to run a check you can trust

Fix your prompts and never reword them. Put the city or Malaysia in the question. Run each prompt three times on each tool, a new chat every time, signed out, in a private window. Write down the country your connection appears to come from. Then check what each answer says about you against your own site.

Each part is there for a reason. OpenAI’s help page on searching the web with ChatGPT says it “may use your IP address to estimate your general location, such as your country, state, or city”, and that “A VPN or network location may affect the approximate location inferred from your IP address.” Its advice when local results look wrong is short. “Include your city, neighborhood, or postal code in your question.” Google’s Gemini Apps Privacy Hub says Gemini uses “the general area from your IP address” by default. Sit in PJ on a VPN that exits in Singapore and you may be testing Singapore.

Signing out matters because “If memory is enabled, ChatGPT may use relevant saved memories when rewriting a search query.” Your own account has probably talked about your store before. A stranger’s has not. OpenAI also says “People who are not signed in can also use web search”, so a signed-out check still gets the web results.

Three runs is a floor. It is the smallest number where you can see whether a name repeats, and nothing more scientific than that.

The last step is always your own site, because the same OpenAI page says “Search results and citations can be incomplete, outdated, or incorrect.” It is also worth pasting the English prompt into Google Search and noting whether an AI Overview or AI Mode answer names or links to you, three times, like the rest.

Our office chair run missed this method in three places, and we would rather say so. We ran the English prompt only. We skipped the Google Search step. And Perplexity hit its sign-in wall twice, so we stopped there and have no Perplexity answer to score. Write a wall down when it happens. It says nothing about whether your store would have been named, so it is a result to record and never a miss.

Four prompts to copy

Change the brackets and nothing else. Start with the category prompt, in English and in Bahasa Malaysia, because it shows whether you come up when the shopper has not named you. The brand prompt shows how the tool describes you once they have. Our test was the English category prompt, with an office chair and Kuala Lumpur in the brackets.

Category prompt (English): I want to buy [product category] online from a store in [city], Malaysia. Which stores would you recommend? Please name specific stores and their websites.

Category prompt (Bahasa Malaysia): Saya ingin membeli [kategori produk] secara dalam talian daripada kedai di [bandar], Malaysia. Kedai mana yang anda cadangkan? Sila namakan kedai tertentu dan laman web mereka.

We ran only the English version. In Bahasa Malaysia, our test prompt would read: Saya ingin membeli kerusi pejabat secara dalam talian daripada kedai di Kuala Lumpur, Malaysia. Kedai mana yang anda cadangkan? Sila namakan kedai tertentu dan laman web mereka.

Brand prompt (English): What can you tell me about [store name] in [city], Malaysia? Please say where the store is based and what it sells.

Brand prompt (Bahasa Malaysia): Apakah yang boleh anda beritahu saya tentang [nama kedai] di [bandar], Malaysia? Sila nyatakan di mana kedai ini berpangkalan dan apa yang dijualnya.

If you also sell into Singapore, write a second pair that says Singapore and keep those runs in their own tally.

What our office chair log shows

Here is our log, counted per tool. Count each tool and each language on its own and never add them into one score.

Store, as the answers named itChatGPT, named inGemini, named in
Merryfair3 of 30 of 3
Ergoworks3 of 31 of 3
SPACE My3 of 32 of 3
MD Store (M Degree)3 of 30 of 3
Aitsulab3 of 30 of 3
IKEA Malaysia2 of 32 of 3
TTRacing Malaysia1 of 30 of 3
Alterseat0 of 33 of 3
OfficePro0 of 33 of 3
EVIS0 of 32 of 3
AM Office0 of 32 of 3
Haus0 of 31 of 3
CK Home Furnishing0 of 31 of 3

Perplexity: sign-in wall on both attempts, no answers recorded.

ChatGPT named five stores in every run. None of those five made every Gemini answer, and three of them never showed up in Gemini at all. Alterseat and OfficePro went the other way, in every Gemini answer and in none from ChatGPT. Neither tool showed a model name, so we wrote “not shown” in the sheet.

ChatGPT showed its sources in every run, with links out to the stores’ sites, and in runs 2 and 3 those links carried utm_source=chatgpt.com. Gemini was patchy. Its first answer had no source chips, only a website column in a table. The second showed chips on some stores. The third showed none. Gemini’s help page notes that “Not all responses include related links or sources”, so record a missing source list as missing.

Then we checked a few things the answers said against the stores’ own sites, on 20 November. ChatGPT’s second answer said Merryfair has “a five-year warranty on specified components. Free delivery and professional installation in Klang Valley.” Merryfair’s online store page lists a 5-year warranty on the mechanism, arms, frame and seat foam, and free delivery and professional installation within Kuala Lumpur and the Klang Valley. The showroom ChatGPT placed on Jalan Genting Kelang is at the address Merryfair gives. Gemini said SPACE My has “a showroom along Jalan Klang Lama in KL”, and SPACE My’s contact page gives Medan 28, Jln Klang Lama. Gemini said Alterseat offers “free delivery and assembly within the Klang Valley”, and Alterseat’s home page says Klang Valley and Penang. Those five held up. We could not confirm one price range Gemini gave from the store’s home page, and we did not check the rest.

So in what we checked, the facts were not the problem. Being named at all was. For your own store, check every run anyway. Compare the city, the range, the prices and any “open” or “closed” claim with your own site that day. If the answer cites a page, open it and see whether it supports the line.

The fair sentence from our log is “Merryfair was named in 3 of 3 ChatGPT runs and 0 of 3 Gemini runs, in English, signed out, from a Petaling Jaya connection, on 20 November 2026.” Calling Merryfair “the AI’s favourite chair store” would be wrong. And none of it says which chair is better to sit on.

The log also cannot tell you how many sales an answer caused. It tells you where you stand on a given day, and what to fix.

If you are missing, check whether the tools can read you

A clean miss across three runs is a reason to check whether the tools are allowed to use your pages. This can rule a barrier in or out. It will not explain why someone else got named.

For ChatGPT search, OpenAI’s crawler documentation says “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.” So allow OAI-SearchBot in robots.txt. GPTBot, which OpenAI uses for training, is a separate setting, and blocking it does not opt you out of search. OpenAI says a robots.txt change can take about 24 hours to show up in search, and its help page adds “Placement is not guaranteed.”

For Perplexity, the crawler documentation says PerplexityBot “is designed to surface and link websites in search results on Perplexity” and “is not used to crawl content for AI foundation models.” Perplexity recommends allowing it and says “it may take up to 24 hours for our systems to reflect changes.”

For AI Overviews and AI Mode, Google’s page on AI features and your website says “To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet”, and “There are no additional technical requirements.” Watch for a nosnippet rule. Google’s robots meta tag documentation says it “will also prevent the content from being used as a direct input for AI Overviews and AI Mode.” Search Console also has a Search generative AI control, available to websites worldwide as of 31 August 2026. It includes your site by default. Set it to exclude and your links and content will not appear in those features.

For the Gemini app, Google’s common crawlers documentation says the Google-Extended token in robots.txt controls whether crawled content may be used to train Gemini models and for grounding in Gemini Apps. Make sure you are not disallowing it.

All of this only makes a page eligible. It does not put your store in the answer.

Copy this recording sheet

One row per run. Keep the sheet even when the answer flatters you, because next month’s rerun only means something if this month was logged the same way.

FieldWhat to write
Date and timeMalaysia time, to the minute
ToolChatGPT, Gemini, Perplexity or Google Search
Model name shownCopy it, or write “not shown”
Signed out?Yes or no
Private window?Yes or no
Connection countryWhere the connection appears to come from, VPN exit included
PromptThe exact wording, city and country included
LanguageEnglish or Bahasa Malaysia
Stores namedThe names exactly as written, one per line
How your store was describedPaste the sentence, or write “not named”
Sources or labels shownPage titles, URLs, or “none shown”
Checked against your siteMatch, mismatch with the error quoted, or not applicable
NotesSign-in wall, refused answer, anything that broke the run

When the runs are done, write four lines per tool and language. Named in n of 3. Misdescribed in n of 3. Sources shown in n of 3. Connection country. Those four lines are what you can stand behind.

Is your store on the shortlist?

When a shopper asks an AI tool where to buy, the answer is the shortlist. If your store is not on it, or it lists the wrong city or range, that shopper starts with someone else. Send us your filled-in sheet. We will go through it with you and help you fix the pages and business listings these answers draw on.

Let’s talk.

Tell us what you’re working on and what you want to improve.

Get in touch