What OpenAI and Google say about how their assistants choose products, what ORIZI Group’s own traffic data shows, and the practical steps a brand can take without paying for “AI ranking” folklore.
Disclosure: This article is published by ORIZI Group and may refer to our own manufacturing experience, services and capabilities. External factual claims are supported by cited sources.
Key takeaways
- AI assistants already send real visitors: ChatGPT-referred landings on orizigroup.com rose from 237 to 817 per 30 days by September 2026, and AI assistants made up about 20% of real traffic.
- OpenAI says ChatGPT’s product results are selected independently and are not ads; sellers are ranked on availability, price, quality and whether they are the maker or primary seller.
- Google says no special schema or llms.txt file is needed for AI Overviews or AI Mode; a page must be indexed and eligible to show a snippet.
- Most AI crawlers do not run JavaScript, so specs that appear only after a script loads may be invisible to them.
- No provider publishes ranking weights, so treat any promise of a paid or certain AI placement as folklore.
Is ChatGPT really the new Shopee search?
Not for every purchase: marketplaces still own checkout, reviews and delivery. But AI assistants have become a discovery step that comes before the marketplace. An AI shopping answer is a reply from an assistant such as ChatGPT, Gemini, Perplexity or Copilot that names specific products or sellers in response to a buying question. A citation is the link an assistant shows as the source of that answer.
OpenAI’s merchant page states that shopping with merchant product feeds is currently live for ChatGPT users in the U.S., with more merchants and regions planned. For a Malaysian brand outside that programme, the material an assistant can work with is what it finds on the open web: your website, articles and directories that mention you, and public marketplace pages.
| Metric (orizigroup.com) | Earlier | Later | Window and source |
|---|---|---|---|
| ChatGPT-referred landings | 237 | 817 | Previous 30 days vs last 30 days to 21 Sep 2026; first-party referrer log |
| Distinct pages reached from ChatGPT | about 55 | about 140 | September 2026 check; first-party referrer log |
| AI assistants’ share of real traffic | — | about 20% | GA4 and first-party log, September 2026 |
| Engagement rate: AI assistant vs Google organic sessions | — | 62.0% vs 63.4% | GA4, 24 Aug – 20 Sep 2026 |
The ten pages that ChatGPT sent the most visitors to were all product or category pages — perfume, car and home fragrance, the product catalogue, lipstick and instant coffee among them — not the homepage or news posts. The assistant had already matched a specific need to a specific page before the visitor clicked. ORIZI Group is a B2B manufacturer, not a consumer shop, so read these figures as direction, not as a benchmark for retail brands.
How do AI assistants decide which products to recommend?
The providers’ own documentation is the firmest evidence:
| Source | What the provider says | What it means for a brand |
|---|---|---|
| OpenAI: Shopping with ChatGPT Search | Product results are selected independently and are not ads. ChatGPT considers structured metadata from first-party and third-party providers (such as price and product description) and other third-party content; its own example weighs price, reviews and ease of use. | Price, specs and reviews must be visible, and consistent wherever the product is listed. |
| OpenAI: how merchants are selected (same page) | Merchants are ranked on availability, price, quality and whether they are the maker or primary seller of the item. | Keep stock status current and make it obvious that you own the brand. |
| OpenAI: Overview of OpenAI Crawlers | OAI-SearchBot surfaces websites in ChatGPT search; sites that opt out will not be shown in ChatGPT search answers. | Check that robots.txt and firewall rules do not block it. |
| Google: AI features and your website | Standard SEO fundamentals still apply; a page must be indexed and snippet-eligible; no special schema is needed; Merchant Center and Business Profile data should be up to date. | Ordinary search hygiene is the entry ticket. |
| Google: guide to generative AI search | Unique, first-hand, non-commodity content matters most; AI responses can include product listings from Merchant Center feeds; Google Search ignores llms.txt files. | Original specs, data and photos are worth more than rewritten copy. |
What patterns can you see in AI product answers?
On 26 September 2026, ORIZI Group typed the same question, “Top 10 Cosmetics OEM Manufacturer in Malaysia”, once into seven AI engines: Perplexity, Google AI Mode, ChatGPT, Gemini, Grok, DeepSeek and Bing Copilot Search. Five patterns stood out:
- List articles are cited heavily. The most-cited sources were list-style “Top 10” articles and business directories. Manufacturers’ own websites appeared less often; orizigroup.com itself was cited by two of the seven engines.
- Specific facts travel; gaps get filled in. Where sources stated concrete facts — minimum order from 50 units, halal certification, ISO 22716 — the engines repeated them consistently. Where an engine relied on sources without those specifics, its description turned vague, and one described ORIZI as a dermatology-focused formulator, which is not how ORIZI Group positions itself.
- Statements are repeated almost word for word. Whatever a cited page says, positive or critical, can reappear in the answer, so every claim on a listing should be one you can support.
- Reviews and price are named inputs. OpenAI’s own shopping example weighs price, reviews and ease of use.
- Text beats scripts. Vercel’s study of AI crawler traffic (December 2024) found that none of the major AI crawlers it tracked — including OpenAI’s OAI-SearchBot and GPTBot, Anthropic’s ClaudeBot and PerplexityBot — rendered JavaScript, while Gemini uses Googlebot’s rendering. Specs that appear only after a script runs may never be read.
Limits: one sample per engine, answers change from run to run, and several engines were used while logged in, so personalisation may have shaped the results.
How do I get my product recommended by ChatGPT and other AI assistants?
- Put the facts in plain text. Product name, size, hero ingredient, price, who it is for, how to use it, where it is made, and any registration or certificate with the holder company’s name.
- Make the facts match everywhere. Website, Shopee, Lazada, TikTok Shop and any directory listing should show the same size, price and claims.
- Earn real reviews and real coverage. Google notes that seeking inauthentic “mentions” is not as helpful as it might seem.
- Let search crawlers in. Allow OAI-SearchBot, keep pages indexable for Google, and serve key content in the page HTML rather than through scripts.
- Use structured data as a mirror, not a trick. Google says it is not required for generative AI search but still helps with rich results; never put a fact in markup that is not on the page.
- Make the product easy to describe. One hero claim, a visible spec, a clear format and price. That is a manufacturing decision as much as a marketing one, and the same brief behind our viral and TikTok product manufacturing work; our Malaysia skincare trend index for TikTok Shop shows which formats sell, and what makes a product go viral online covers the selling points.
- Measure it. Segment analytics by referrer (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com) and watch which pages those visitors land on.
Which AI optimisation tactics are folklore?
| Claim you will hear | What the evidence says |
|---|---|
| “Add an llms.txt file to rank in AI.” | Google says Google Search ignores llms.txt files; they neither help nor harm there. |
| “Chunk content and write every synonym.” | Google says neither is required; its systems understand synonyms and multi-topic pages. |
| “Schema is the AI ranking factor.” | Google says structured data is not required for generative AI search. |
| “Pay to be recommended in ChatGPT.” | OpenAI says product results are not ads and are not influenced by its partnerships. |
| “Exact AI ranking factors, in percentages.” | No provider publishes weights. The academic GEO study (KDD 2024) found that adding citations, quotations and statistics raised visibility by up to 40% in its research setup, not in a live assistant. |
ORIZI perspective
For ORIZI Group, the AI shift appeared first in analytics: ChatGPT-referred landings more than tripled within about a month, at a time when most Google clicks to orizigroup.com were still brand searches. We cannot yet tie enquiries to AI traffic, because enquiry events were not yet marked as key events in GA4 in September 2026; that is the next measurement step.
The spot check made one lesson concrete: the answers that described ORIZI accurately drew on pages stating plain facts. It matches an editorial rule ORIZI Group already follows: name the holder company and start year for every credential instead of writing “certified” in general — for example, Dr Orizi (M) Sdn. Bhd. holds NPRA cosmetic GMP (since 2024) and ISO 22716 (since 2022), and D&O Nutraceutical Manufacturing Sdn. Bhd. holds SIRIM GMP MS 1514 (since 2024), as listed on our certifications page. A vague credential is exactly what an assistant fills in wrongly. Limits: this is one B2B website over one month, and no one outside OpenAI, Google or the other providers can see how their systems rank.
Risks, limitations and common mistakes
- Skipping marketplace basics. Listing quality, reviews and stock still decide the sale after an assistant sends the shopper.
- Over-claiming. An assistant may repeat your claim verbatim; a claim it repeats is still your claim, so keep cosmetic claims within cosmetic scope.
- Inconsistent facts. Different sizes or prices across channels can give an assistant a reason to pick another seller.
- Blocking crawlers by accident. Aggressive bot-protection settings can shut out search crawlers along with scrapers.
- Judging visibility on one prompt. Answers vary by engine, wording, login and day; track trends, not single screenshots.
FAQ
How do I get my product recommended by ChatGPT?
Make the product easy to find, read and verify: plain-text specs and price on a crawlable page, the same facts on every marketplace listing, genuine reviews and third-party coverage, and no robots.txt block on OAI-SearchBot. OpenAI says ChatGPT weighs structured product data, third-party content and reviews.
Are ChatGPT product recommendations paid ads?
No, according to OpenAI. Its help centre states that product results are selected independently by ChatGPT, are not ads and are not influenced by OpenAI partnerships; ads are shown separately from product results.
Do I need schema markup or an llms.txt file for AI search?
Not for Google. Google says no special schema.org markup is needed for AI Overviews or AI Mode and that Google Search ignores llms.txt files. Structured data is still useful for rich results, as long as it matches the visible page.
Is ChatGPT shopping available for Malaysian brands?
The merchant product-feed programme is U.S.-first: OpenAI’s merchant page says shopping is currently live for U.S. users, with more regions planned. ChatGPT search can still cite and link to websites, so Malaysian brands should keep their pages crawlable and their facts consistent.
How can I tell whether AI assistants are sending visitors to my site?
Check referrers in your analytics for chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, and note the landing pages. On orizigroup.com, ChatGPT-referred landings rose from 237 to 817 per 30 days by September 2026.
Sources & references
- OpenAI Help Center — Shopping with ChatGPT Search — product results not ads; structured metadata, third-party content and reviews; merchant ranking factors
- OpenAI — Power product discovery in ChatGPT (merchants) — product feeds, Agentic Commerce Protocol; shopping live for U.S. users
- OpenAI — Overview of OpenAI Crawlers — OAI-SearchBot and opting out of ChatGPT search answers
- Google Search Central — AI features and your website — no additional requirements; indexed and snippet-eligible; no special schema
- Google Search Central — Guide to optimizing for generative AI features — non-commodity content; Merchant Center feeds; llms.txt ignored; inauthentic mentions
- Vercel — The rise of the AI crawler (December 2024) — major AI crawlers do not render JavaScript
- Aggarwal et al. — GEO: Generative Engine Optimization (KDD 2024) — citations, quotations and statistics raised visibility by up to 40% in a research setup
- ORIZI Group first-party data (internal) — GA4 and referrer log for orizigroup.com, September 2026; seven-engine AI answer check, 26 September 2026
Update history: October 2026 — first published.
Building a product that shoppers and AI assistants can both describe in one line? See how ORIZI Group approaches viral and social-commerce product manufacturing, or send us a short product brief.
