Skip to main content

Short answer

AI recommends suppliers by searching the web first, then summarising what it finds into a shortlist. Whether you appear depends on four stages: can your pages be crawled, understood and trusted, and will buyers choose you. Basic SEO on Google and Bing is the prerequisite, but nobody can guarantee a recommendation.

The short answer: AI does not score and rank suppliers

With search on, an AI answers a supplier question like this: it decides whether to search, writes queries, reads the results, then summarises what it found into an answer with sources. Google’s documentation for Gemini describes it in exactly those terms. You appear on the list because your material did not drop out at any step of that chain.

We split the chain into four stages:

StageWhat the AI is doingWhat your material must doMatching bottleneck in our diagnosis
CrawledRetrieving and reading pagesBe accessible to search engines and AI crawlersInvisible
UnderstoodTurning page content into factsClear body text, complete parameters, natural languageHard to read
TrustedJudging whether sources are reliable and consistentVerifiable third-party sources, consistent details across platformsNot trusted
ChosenSummarising candidates into a shortlistSpecifications, certifications and terms that match the buyer’s questionThe outcome, not a separate action

Technical facts: where each AI gets its material

The platforms do not publish full retrieval details. We only state what they have said officially and mark the rest "not published".

PlatformWhat the vendor saysWhat it means for you
GeminiGoogle’s developer documentation says that with the Google Search tool on, the model decides whether to search, runs the queries, then answers from the results and returns sourcesYour pages need to be indexed and rank on Google first. That documentation covers the Gemini API; for the Gemini app and AI Overviews, follow Google’s own explanation
ChatGPT searchOpenAI’s help article says it uses third-party search providers and partner content, and it runs its own crawler, OAI-SearchBot (see the bot documentation)Which providers, and in what proportion, is not published. Outside observers often mention Bing, but that is observation, not an OpenAI list. So do basic SEO for both Google and Bing, and allow OAI-SearchBot
PerplexityRuns its own crawler, PerplexityBotAllow it in robots.txt as well

The practical conclusion: when you cannot confirm every platform’s sources, make the foundation you control complete. That means both Google and Bing can index you, AI crawlers can read you, and the body text is visible without JavaScript. See AI crawlers and GEO vs SEO.

Stage 1: being crawled

If a page never enters retrieval, nothing later matters. Common problems for manufacturers:

  • robots.txt or firewall blocks: many sites block scrapers and catch AI crawlers in the same rule. Check OAI-SearchBot, GPTBot, PerplexityBot, ClaudeBot and Google-Extended one by one.
  • Servers in mainland China: overseas crawlers may meet slow responses or network blocks, and a page that fails to fetch never reaches retrieval. Use overseas nodes or a CDN, and open your site from an overseas network to check.
  • Body text rendered by JavaScript: some AI crawlers do not run JavaScript and see an empty shell. Render the main content on the server.

Stage 2: being understood

A fetched page still has to be read and restated. Common problems:

  • Machine-translated English: terms are not the industry’s own, units are inconsistent, and the AI struggles to match the page to a buyer’s question.
  • Key parameters locked in PDF catalogues or images: the page body has no model numbers, dimensions, materials, tolerances or certificate numbers.
  • No clear product page: dozens of models on one page, so questions and answers do not line up.

For how to write a product page, see the 6-point checklist.

Stage 3: being trusted

AI leans on sources that confirm one another. The deductions cluster in three places:

  • Inconsistent entity details: your site uses one English name while LinkedIn, trade directories and B2B platforms each use another, so the AI cannot tell they are the same company. Align the full English name, short name, address and contact details; see entity.
  • Only a +86 landline: overseas buyers cannot reach you directly, and there is no signal that the company is real and can be talked to. Add a mobile or WhatsApp number with the international prefix, a company-domain email and a full address.
  • No verifiable third-party sources: certification bodies’ public databases, trade association directories, trade media and exhibitor lists carry more weight than your own claims. This is what the "trust" in E-E-A-T means.

Stage 4: being chosen

When an AI lists a few candidates, buyers look at whether specifications match, certifications are complete, minimum order quantity and lead time are clear, and cases are real. Pages that state these clearly are easier for an AI to use in its answer. But the final ordering and how many suppliers are listed are generated by the platform, and you do not control them.

A sensible order of work for manufacturers

  1. Self-test first: use the free audit to see your site’s technical foundation and how each platform answers, and record the baseline.
  2. Fix the foundation: allow the crawlers, render body text on the server, confirm the site opens overseas.
  3. Align entity details: company name, address, contacts and certificate numbers consistent everywhere.
  4. Rewrite the core product pages: complete parameters with sources, English reviewed by someone who knows the industry, technical content confirmed by your engineers.
  5. Build third-party sources: verifiable records in certification databases, trade associations and trade media.
  6. Re-test monthly: with a fixed question set; the definitions are in how to measure GEO results.

We do not promise rankings

We do not promise "guaranteed #1" or "guaranteed AI recommendation", for three reasons:

  1. Answers are generated by the platform from its own retrieval and model, which we cannot see or control.
  2. The same question gets a different answer at a different time, in a different region, or on a different account.
  3. Google’s guidance on hiring an SEO also says no one can guarantee a #1 ranking and warns against anyone who claims otherwise.

What we can promise is process: what we do, how we re-test, and honest reporting, including the months with no improvement. For how to judge any agency, see how to choose a GEO agency. Our export GEO service follows these four stages. To find out which stage you are stuck at, run a free audit, or read the glossary entry on GEO.

Frequently asked questions

What does ChatGPT base supplier recommendations on?
With search on, ChatGPT retrieves web pages and summarises them. OpenAI says it uses third-party search providers and partner content and has its own crawler, OAI-SearchBot; the proportions are not published. So do basic SEO for both Google and Bing, and allow OAI-SearchBot.
Does hosting in mainland China hurt AI visibility?
It can. Overseas crawlers may meet slow responses or blocks, and pages that fail to fetch never reach retrieval. Use overseas nodes or a CDN and check the site from an overseas network.
Is good SEO enough?
SEO is the foundation but not enough. AI also has to understand and trust your material, which needs complete parameters, consistent entity details and third-party sources. See GEO vs SEO.
Can you guarantee that ChatGPT or Gemini will recommend us?
No. Answers are generated by the platform and change with time, region and account. What we can do is make the four stages solid, re-test monthly and report honestly.

See how AI describes you today

The free AI visibility audit asks ChatGPT, Perplexity and Gemini questions as an overseas buyer would, and checks your website’s technical foundation.

Start the audit

More articles