How ChatGPT, Gemini and Perplexity describe your company

TL;DR

• AI assistants describe your company from two sources: what the model learned during training, and what it finds in a live web search at answer time. The second is where you have the most influence.
• Each assistant searches differently, but all of them build answers from pages they can crawl, index and trust — your own site and what others publish about you.
• To change what they say, fix the sources: clear facts on your site, consistent third-party listings, crawlers allowed. Then measure with a fixed prompt set and the reports Google now provides.

Why this matters for B2B companies

Buyers now ask assistants before they ask you. In a Gartner survey of 645 B2B buyers (August–September 2025), 45% said they used generative AI during a recent purchase, "primarily to gather information on vendors and products." The same survey found buyers split on trust: asked where they are more likely to encounter misleading information, 51% said generative AI and 49% said a sales rep. And 69% prefer to validate AI-generated insights with sales reps.

That gives you two jobs: make sure the assistant's first answer about you is right, and make your own site the place where buyers can check it.

Two sources of what an assistant "knows"

A large language model starts with what it learned in training: text collected up to a cutoff date, including pages fetched by training crawlers such as OpenAI's GPTBot or Anthropic's ClaudeBot. That memory is broad, but it is frozen and blurry on details. Ask a model about a mid-sized B2B company from memory alone and you may get an outdated plan name, a feature from a competitor, or a confident invention — a hallucination.

So assistants also look things up. They run a web search, read a handful of pages and write an answer from them, usually with citations. This is grounding, and it is the path you can influence, because it runs on pages you publish and pages about you that others publish.

How each assistant looks things up

ChatGPT

OpenAI's help center says ChatGPT "may search the web automatically when your question would benefit from current information," and users can also start a search by hand (OpenAI). When it searches, it "typically rewrites your query into one or more targeted queries" and may send them to partner search providers; the privacy policies OpenAI links for those providers are Microsoft's and Shopify's. The pages it can show and cite come through OAI-SearchBot, and OpenAI says sites that opt out of that crawler "will not be shown in ChatGPT search answers" (OpenAI crawler docs).

Gemini and Google's AI features

Google's Grounding with Google Search documentation describes the loop: the model "determines if a Google Search can improve the answer," generates and runs one or more queries if needed, then answers with citations. For AI Overviews and AI Mode inside Google Search, Google says the features are "rooted in our core Search ranking and quality systems" (optimization guide) and may use "query fan-out" — "issuing multiple related searches across subtopics and data sources" (AI features documentation).

One control is easy to miss. The Google-Extended robots.txt token governs use of your content for Gemini training and for grounding in Gemini apps and Vertex AI, but it "does not impact a site's inclusion in Google Search" (Google). Blocking it keeps you in AI Overviews but can keep you out of Gemini app answers.

Perplexity

Perplexity says it searches "the internet in real-time" and that "each answer includes numbered citations linking to the original sources" (Perplexity Help Center). Its index is built by PerplexityBot, which Perplexity says is used "to surface and link websites in search results" and "not used to crawl content for AI foundation models" (Perplexity docs).

Side by side

Blocking these crawlers is covered in robots.txt for AI bots. The short version: block training if you like, but blocking search crawlers removes you from these answers.

What decides which pages get used

No assistant publishes its full ranking logic, and Google's guide to generative AI features says plainly that "no third-party tool has access to our internal ranking or AI systems." What the documentation and research do support:

• Eligibility first. For Google's AI features, a page "must be indexed and eligible to be shown in Google Search with a snippet." For ChatGPT and Perplexity, the page must be reachable by their search crawlers.
• Relevance to many sub-questions. Fan-out and query rewriting mean an answer is built from several narrow searches. A page that clearly answers "Does X integrate with HubSpot?" can be picked even if your homepage never ranks for the broad question.
• Specific, original content. Google's guide puts it directly: "Don't just recycle what others on the internet have already said, or could easily be produced by a generative AI model."
• Evidence in the text. In the GEO study (KDD 2024), adding quotations, statistics and cited sources to pages raised their visibility in generative-engine answers by up to about 40% in the authors' benchmark, while keyword stuffing gave little or no improvement. It was a lab benchmark run on 2023–2024 systems, so treat it as direction, not a formula.

Third-party sources count too

Look at the citations under answers about your company. Besides your own site, you may find review platforms, comparison articles, directories, forum threads and partner pages. Assistants that search the web describe you partly the way the web describes you. If a directory still lists a product you discontinued, or a three-year-old review quotes old pricing, expect to see it repeated.

Why assistants get companies wrong

The usual causes, in rough order of how often they are fixable:

1. The fact isn't on your site in plain text. Pricing in an image, integrations behind a script, or "contact us for details" on everything.
2. Your own pages disagree. Gartner found 69% of B2B buyers saw inconsistencies between a vendor's website and what its sellers said. Inconsistent pages confuse models the same way.
3. Stale third-party pages. Old reviews and listings outnumber your current pages.
4. Name collisions. A similarly named company, product or open-source project gets blended into yours.
5. The answer came from memory. No search was run, so the model filled gaps from training data.

OpenAI's own help page warns that "search results and citations can be incomplete, outdated, or incorrect." Plan for errors; do not assume they are rare.

A routine to audit and fix your AI presence

What doesn't work

• AI-only files and rewrites for Google. Google's guide says you don't need machine-readable AI files, special markup or Markdown, and that "there's no special schema.org markup you need to add." An llms.txt file may help some agents, but not Google Search.
• Manufactured mentions. Google says pursuing inauthentic mentions "isn't as helpful as it might seem."
• Keyword stuffing. The GEO study found it gives little or no improvement in generative answers.
• Blocking everything, then asking why you're invisible. No crawl, no citation.

How to measure it

• Google Search Console now has a Generative AI performance report showing impressions in AI Overviews and AI Mode by page, country, device and date; Google says it was rolled out to all websites as of August 31, 2026.
• Google Analytics groups visits referred by assistants such as ChatGPT, Gemini, Deepseek, Copilot or Grok into an AI Assistant channel. AI Overviews and AI Mode clicks stay in Organic Search.
• Server logs show OAI-SearchBot, ChatGPT-User, PerplexityBot and similar fetches of your key pages — a leading indicator that your pages are being used.
• Your prompt set: percentage of correct answers and of answers citing your site, month over month.

The site-side groundwork — crawl access, facts in HTML, labeled forms — is in the agent-ready website checklist.

Practitioners now call this work generative engine optimization, while Google calls it "still SEO." Either way, the job is the same: publish clear, consistent facts where assistants look, and check what they say about you on a schedule.