GPT-6 Is Now the Default in ChatGPT: What Changed in Shopping Answers for Gulf and Egyptian Buyers (Before/After Test)

GPT-6 made ChatGPT’s shopping answers more structured and more Noon-heavy. We re-ran our 4 October 2026 prompts on 11 October and added 13 more. All 20 answers named one labelled top pick with a price in SAR, AED or EGP. Fifteen ended in clickable follow-up choices, and Noon appeared in all 7 repeat prompts, up from 5.

Some things did not move. Amazon and Salla or Zid stores appeared in none of the 20 answers, as on 4 October. Brand-owned stores now get linked where none were before, 7 answers warned that their own prices were possibly out of date, and two Gulf Arabic prompts got Egyptian dialect back. The test used one personalised Plus account and one run per prompt, so read it as a dated snapshot.

What did OpenAI change in ChatGPT on 7 October?

OpenAI made GPT-6 the model behind ChatGPT’s Chat tab on 7 October 2026, together with a feature it calls Intelligent UI. Answers can now mix text, comparisons, buttons, forms and small tools, and start streaming while the model is still thinking or searching. Paid plans get GPT-6 Sol, and Free and Go get GPT-6 Luna.

The ChatGPT release notes say the rollout started globally on 7 October for Plus, Pro, Business and Enterprise, and reached Free and Go from the next day. OpenAI’s launch post describes responses with “graphics, tappable buttons, forms, charts, and interactive experiences”. Plain text remains an option when it serves better.

A day earlier, Search Engine Roundtable reported that OpenAI was testing labels on ChatGPT product results, such as “best all-rounder” and “best for beginners”. For a retailer, those two changes matter more than the model name. They frame the product, and they let a shopper narrow the choice with one tap.

How we ran the before and after test

A VOCTOS team member ran 20 shopping prompts on 11 October 2026, between about 22:20 and 22:55 Cairo time, on the same ChatGPT Plus account used on 4 October. Seven prompts repeat our 4 October ChatGPT shopping test, and 13 are new. Each prompt ran once, in Temporary Chat, on desktop Chrome from Cairo.

Eleven prompts were in English, 6 in Gulf Arabic and 3 in Egyptian Arabic. By market, 10 asked about Saudi Arabia, 7 about the UAE and 3 about Egypt, for everyday purchases from air fryers and phones to abayas, dates and air conditioners.

ChatGPT’s web app recorded the model as gpt-6-thinking in all 20 conversations, with the reasoning setting left at its default, Medium. OpenAI’s notes say Plus runs GPT-6 Sol, but no answer showed the word Sol, so we describe it as the GPT-6 model the Plus app served. Reported thinking time ran from 8 to 33 seconds.

For each answer we counted product images, comparison tables, local-currency mentions, retailers and linked domains, labels on picks, interactive follow-ups and any “Sponsored” unit. Image counts are a floor, because images lower in an answer load late. The first prompt’s wording was reconstructed from the 4 October log.

What changed in ChatGPT’s shopping answers?

The format changed most. On 11 October every answer named one top pick with a label such as “Best overall”. Fifteen ended in clickable choices or a short form, and 11 included a comparison table. Product images and local-currency prices appeared in every answer, as they had in all 7 answers on 4 October.

The labels match what Search Engine Roundtable reported OpenAI testing a week earlier. English answers used “Best overall”, “Best value”, “Best for mopping” and “Best for Android”. Arabic answers used labels meaning “my first choice”, “cheapest” and “best budget pick”. The robot vacuum answer split its picks, with a Roborock Q10 VF+ at SAR 1,057 as “Best overall for vacuuming” and a Eufy model as “Best for mopping”.

Most endings were choice chips, such as household size for an air fryer or budget and style for abayas. Three closed with a small form and a “Find my best match” or “Find my best gift” button. The smartwatch answer had a selector that changed its pick by the shopper’s phone. Five still ended with a plain typed question.

On 11 October a shopper could narrow ChatGPT’s pick with one tap, and every pick arrived already labelled.

Did Noon gain ground on the repeat prompts?

Yes. On the 7 prompts asked on both dates, Noon went from 5 answers to all 7. The answers also named more sellers, rising from 6 distinct retailers on 4 October to 17 distinct sellers on 11 October, so Noon gained inside a busier answer. Faces, the perfume retailer named twice on 4 October, dropped out of both perfume answers.

Across all 20 answers, Noon was named or linked in 14, including Noon Minutes for perfume and dates, for Saudi, UAE and Egyptian prompts alike. The seller totals are not like for like. The 25 mentions on 11 October include brand stores, Apple’s own price and dubizzle; the 13 on 4 October were all retailers.

The final top pick changed on 3 of the 7. The English air fryer answer moved from a Philips 3000 Series NA332/00 at about SAR 427 to a Philips 2000 Series NA230/00 at SAR 439, and both perfume answers switched from Montblanc Explorer to YSL Y. The Gulf Arabic iPhone answer led with Noon before correcting itself back to Carrefour, the same retailer it picked on 4 October. The English iPhone answer kept Carrefour as its first retailer, at AED 4,099.

Why are Amazon and Salla or Zid stores still missing?

We do not know, and OpenAI does not publish how ChatGPT picks retailers. What the data shows is a repeat result: Amazon in none of the 20 answers across Saudi Arabia, the UAE and Egypt, and no Salla or Zid store either. Meanwhile, brand-owned stores or a brand’s own price appeared in 7 answers, against none on 4 October.

The brand stores were Tefal Saudi for air fryers, Bosch Egypt for vacuums, Swiss Arabian Saudi for oud, Bateel for dates and Samsung Saudi for smartwatches. Louzan, KaafMeem and Leem appeared for abayas. The English iPhone answer also cited Apple’s own price.

Specialist chains carried several answers alone. Five of the 20 named a single seller. Noon had the robot vacuum and the Egyptian earbuds, eXtra the washing machine, Jarir every smartwatch price, and Al Manea the split air conditioner. Only the smartwatch answer also linked a brand’s own site, Samsung Saudi, next to its single seller.

Does ChatGPT now warn when its own prices are stale?

Sometimes. Seven of the 20 answers flagged one of their own prices as old, unverified or unavailable. Nearly all added that advertised prices are not guaranteed at checkout. One answer corrected its cheapest price partway through, and another caught a retailer’s homepage contradicting its own product page.

The warnings were specific. One price came from “an older Noon listing”, and the robot vacuum answer said its indexed prices were from earlier listings. The oud answer said the exact Saudi price was not verified, and two answers said an LG washer was not available at eXtra and an Acer laptop not at the observed price.

The Gulf Arabic iPhone answer first named Noon at AED 4,129 as cheapest. Later in the same answer it posted what it called an important update: Carrefour at AED 4,095.90. That fits OpenAI’s description of GPT-6 adding findings after it has started answering.

The sharpest catch came in the English air fryer answer. ChatGPT noted that eXtra’s homepage showed a Philips 6.2-litre air fryer at SAR 295 while the product listing said SAR 519. It called this “a promotion or pricing discrepancy worth checking”. The same evening, the Philips NA230/00 was quoted at SAR 439 in the English answer and at 329 to 415 riyal in the Gulf Arabic one. Our price-markup audit of 40 Arab store pages shows how gaps like these form.

Selling into these markets this season? Show me what ChatGPT says about my products and prices.

Where did dialect and personalisation slip?

In two of the 6 Gulf Arabic answers. One answered entirely in Egyptian Arabic and one opened with an Egyptian phrase before switching to Gulf wording. On 4 October every Arabic answer matched the dialect of its question. An English answer about Dubai also brought up Egypt, which the prompt never mentioned.

Both slips came on repeat prompts, the Gulf Arabic perfume and iPhone questions, which had drawn Gulf answers on the same account a week earlier. The account is used from Cairo and personalisation was on, which is the likeliest source, but one run per prompt cannot prove it. The English iPhone answer for Dubai told the user to check SIM compatibility “if you’re planning to use the phone in Egypt”.

Digits were inconsistent too: two Arabic answers wrote prices in Arabic-Indic digits and six used Western digits. For a Gulf brand the lesson is narrow. Your Arabic pages are one input among several, and the shopper’s account history can change the voice of the answer.

What should Arab retailers and brands fix before White Friday?

Make every product readable without a click: a clear model name, a price in SAR, AED or EGP, stock status and a price that is current. Then build brand pages that answer the comparison questions ChatGPT now labels, in Arabic and English, and keep marketplace listings in order, since Noon appeared in 14 of 20 answers.

1. Complete product data. Put model numbers, sizes and capacities in the title and specification table. ChatGPT named exact models such as the Philips NA230/00, and a page without one gives it nothing to match.
2. Local-currency prices in plain text. All 20 answers quoted the shopper’s own currency. Show the SAR, AED or EGP price on the page itself, one current price per product.
3. Price and stock freshness. Seven answers flagged their own prices as stale or unverified. Update price and stock on the page and in your feed on the same schedule, with sale start and end dates.
4. Brand store pages that answer comparison questions. Write the “best for” page yourself: which watch suits an Android phone, which oud stays under SAR 300. Our guide to how ChatGPT chooses what to cite covers the page structure.
5. Arabic product pages. Nine of our 20 prompts were in Arabic. Publish full Arabic pages on their own URLs, with the same model numbers and prices as the English pages.
6. Marketplace and chain listings. Your Noon listing is part of your AI visibility. If eXtra, Jarir or Al Manea stocks you, check their page carries your exact model and price.
7. Your own prompt log. Ask your buyers’ questions in each dialect, on Free and signed out as well as Plus, and log the answers weekly until White Friday in late November.

Our read

GPT-6 changed the shape of a shopping answer more than its sources. The same large retailers win, Noon most of all. What changed is that each answer arrives as a labelled shortlist with a button to narrow it, so the shopper makes fewer choices alone.

We think the opening for brands is the labelled pick. ChatGPT linked brand-owned stores or a brand’s own price in 7 of 20 answers, up from none. Each of those pages offered a product and a local price it could quote. A brand that publishes its own “best for” comparisons, in Arabic and English, hands the model the frame it is already using. ChatGPT still sends most AI chatbot clicks in Gulf markets, per our review of AI assistant use in the Gulf, so shape this answer first.

The risk is price trust. Seven answers in 20 told the shopper a price was old, unverified or unavailable. In White Friday week, when prices move daily, that warning will land on whichever store’s page is out of date.

What this test does not show

One signed-in Plus account with personalisation active cannot show what most Arab shoppers see. Each prompt ran once, so we cannot separate a GPT-6 effect from ordinary run-to-run variation, and the dialect slips point at the account as much as at the model.

We ran no Free or signed-out session, so we did not see GPT-6 Luna or any ad; Free and Go are the tiers that carry ChatGPT ads. Image counts are a floor. Labels, tables and follow-up chips were not scored on 4 October, so for those we report what GPT-6 showed, not a measured change. Prices are what ChatGPT reported, and we checked none of them at checkout.

FAQ

Is GPT-6 now the default ChatGPT model for shoppers in Saudi Arabia, the UAE and Egypt?

OpenAI’s release notes say GPT-6 began rolling out globally on 7 October 2026 for paid plans and from 8 October for Free and Go, with Sol on paid plans and Luna on Free and Go. Our Plus account, used from Cairo, was served the gpt-6-thinking model for all 20 shopping prompts on 11 October.

Why does ChatGPT keep recommending Noon instead of my store?

OpenAI does not publish how ChatGPT chooses retailers. In our test Noon appeared in 14 of 20 answers and in all 7 repeat prompts, up from 5 on 4 October. Brand-owned stores or a brand’s own price appeared in 7 answers, so a clear product page with a local price can sit alongside the marketplace.

Does ChatGPT show Amazon or Salla stores in shopping answers?

Not in our two tests. Amazon was named in none of 7 answers on 4 October and none of 20 on 11 October, across Saudi, UAE and Egyptian prompts. No Salla or Zid store appeared either. One account and one run per prompt cannot explain why, so test your own category and dialects.

ChatGPT showed an old price for my product. What should I fix?

Start with the page ChatGPT linked. In our test, 7 of 20 answers flagged their own prices as stale or unverified, and one caught a homepage price of SAR 295 against a product page price of SAR 519. Keep one current price per product, in local currency, and update page and feed together.

Do Arabic prompts get the same ChatGPT shopping answer as English ones?

Close, but not identical. On 11 October the air fryer and perfume prompts picked the same brand in English and Gulf Arabic, but prices and runners-up differed. Two of 6 Gulf Arabic answers slipped into Egyptian dialect on a Cairo-based account, so check Arabic answers separately for each market you sell in.

The better question

The useful question is not which model ChatGPT runs this month. It is whether your product page gives any model a name, a local price, current stock and a reason to pick it, in the language your buyer typed. Model names will keep changing, and those four facts will still decide what a shopping answer can quote.

If your products already carry clean model names, local prices and daily stock updates, spend the coming weeks on Arabic and English comparison pages and a weekly prompt log. If they do not, fix the product data first, because a labelled pick needs a price it can trust. Either way, help me get my products into ChatGPT’s shopping answers before White Friday.