When a buyer asks an AI assistant, does your company come up?

Start with the buyer. Gartner surveyed nearly 650 B2B buyers between August and September 2025. On a recent purchase they used an average of seven information sources, and 45% said they used generative AI, mainly to gather information on vendors and products ¹³. In the same survey, 69% said they prefer to validate AI-generated insights with a sales rep, and 51% said they are more likely to run into misleading information from generative AI ¹³. AI is now part of procurement research, but buyers do not fully trust it. For a Chinese factory that means two jobs: get onto the shortlist the assistant produces, and make sure your website and public facts hold up when the buyer checks the AI's answer against them.

Assistants search before they answer

As of 2026, the major assistants look things up on the web before answering a question that needs current information, then compose the reply from the pages they find. Google explains that AI Overviews and AI Mode can split one question into many related searches (it calls this query fan-out), looking at different subtopics and sources before choosing the links that back up the answer ¹. OpenAI's help page describes a similar loop: ChatGPT turns the user's question into its own search queries, sends them to search providers, reads what comes back and may search again with narrower terms ⁹. Whether an assistant mentions you depends heavily on whether those searches can find your pages.

That is why ordinary search visibility still matters. Google's documentation sets the same bar as ordinary search: a page has to be in Google's index and able to show with a snippet before it can be linked from AI Overviews or AI Mode. Google adds that nothing extra is required, no special AI file and no special schema ¹. Google's May 2025 guidance adds the rest: unique content that helps people, a good page experience, pages Googlebot can crawl, structured data that matches what is visible on the page, and up-to-date Merchant Center and Business Profile information ².

Each platform runs its own crawlers

OpenAI runs separate bots for separate jobs. OAI-SearchBot is the one that lets ChatGPT search find and link your site; block it and your pages drop out of those answers. GPTBot gathers material OpenAI may use for model training, and blocking it tells OpenAI to leave your content out of training. ChatGPT-User only visits when a person in ChatGPT asks it to open a page ⁸. Perplexity divides its crawlers along similar lines: PerplexityBot finds and links sites for Perplexity's answers and, according to Perplexity, is not used for training foundation models, while Perplexity-User covers visits a user triggers ¹⁰. On Google's side, Google-Extended is a separate token that decides whether your content may be used for Gemini training and grounding, and Google says it has no effect on whether or where you appear in Google Search ³. The distinction matters. Companies that block every AI bot to protect their content often give up the chance to be cited in AI search as well.

A robots.txt file only tells well-behaved crawlers which URLs they may visit. It is not a way to keep a page out of search, and not every crawler obeys it ⁴. In practice, what actually locks AI crawlers out is more often a firewall rule, a CDN bot filter, or a server that is slow or times out for visitors outside China.

How an assistant decides who you are

When an assistant describes a company, it draws on several sources: your website, your structured data, industry directories, social profiles, press coverage and buyer reviews. Google says Organization structured data on your home page helps it understand your name, URL, logo, address, contact points, and the other official profiles you list with sameAs ⁵. For products, Google distinguishes product snippets from merchant listings, and says providing both structured data and a Merchant Center feed gives you the widest eligibility ⁶. Google also states that it does not guarantee features using structured data will appear ⁵.

Google's quality guidance names trust as the most important part of E-E-A-T, and asks whether content shows who wrote it, what it is based on, and whether it adds original information ⁷. Original information is where a factory has a real advantage: actual production lines, inspection steps, certificate numbers and application examples are things a trading company or a content farm cannot write.

Why the answer changes from run to run

Ask the same question today and tomorrow, or from Shanghai and from Chicago, and you can get different answers. ChatGPT estimates a general location from the IP address and, when memory is on, may use what the user said earlier to rewrite the search ⁹. Google's fan-out sends different searches for different phrasings ¹. For that reason our reports never quote a single position. The same list of buyer questions goes to the main assistants again at regular intervals, and what we follow is the share of answers that name you and the pages they link to, summarized in the monthly report.