A customer asks ChatGPT who to call. Does your business get named?

Three things, in order: whether the assistant's crawler can reach and read your site at all, whether your basic facts (name, address, phone, hours, services) are stated in plain unambiguous text somewhere it can find, and whether other sources — directories, review sites, local press — say the same facts back. Design and traffic rank do not matter to this process the way they matter to Google's blue links.

Every major answer engine works the same way at a mechanical level: a program requests your page, reads the text, and either uses it to answer a live question or folds it into a training run. If that program is blocked — by a robots.txt rule, by a site that requires JavaScript to render any visible content, by a page with no text describing what the business actually does — nothing downstream happens. You cannot be cited from a page that was never read.

Assuming access, the next filter is clarity. These systems are extracting facts, not vibes. A page that states 'Licensed electrician serving Port St. Lucie, Fort Pierce, and Vero Beach, open Monday through Saturday, emergency calls after hours' gives a model something to quote. A hero image with the same information baked into a graphic gives it nothing — no crawler reads text off a JPEG.

The last filter is corroboration. Answer engines lean on aggregate signal the same way people do: if your Google Business Profile, your own site, an Angi listing, and a local news mention all agree you're an electrical contractor at the same address with the same phone number, that agreement is itself a ranking input. One glossy site with no outside confirmation reads as less trustworthy than five plain, matching listings.

Perplexity runs PerplexityBot for indexing and Perplexity-User for live answer fetches — its own documentation states neither is used to train foundation models. Anthropic runs ClaudeBot, Claude-User, and Claude-SearchBot for the same three purposes. Google's AI Overviews use standard Googlebot indexing plus Google-Extended, a separate toggle that controls AI training and grounding use without affecting normal Search ranking.

The naming is consistent across vendors once you see the pattern: one crawler for background indexing, one for live per-query fetches, and — where the vendor also trains models — a third, separate one just for training data. That separation exists specifically so a site can allow being cited in a live answer while opting out of training, or vice versa. Read each vendor's bot page directly rather than guessing; user-agent strings and behavior have changed before and will again.

Google is the case worth being precise about, because it's the one people misconfigure hardest. Google-Extended has nothing to do with whether Googlebot can crawl and index you for regular Search — it only governs whether crawled content feeds Gemini training and certain grounding uses. Blocking it does not hurt your ranking. Google's own AI-features documentation is also blunt that there is no special schema or markup requirement to appear in AI Overviews — the standard SEO fundamentals (crawlable, indexed, factually clear) are what it evaluates.

Open robots.txt and read it. Confirm nothing is blocking GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, ClaudeBot, Claude-User, Claude-SearchBot, or Googlebot unless you have a specific reason to. Most accidental blocks come from a platform default, not a decision anyone made.

Write your core facts in plain text somewhere a crawler can read without executing JavaScript: business name, service area, phone, hours, and what you actually do, in sentences — not only inside a hero graphic or a contact form. Add LocalBusiness schema with the same facts. It won't force a citation, but it removes any ambiguity for whatever does the reading.

Match your NAP — name, address, phone — exactly across your site, Google Business Profile, and every directory listing you control. Mismatched suite numbers or a swapped phone number are the single most common reason corroboration fails silently.

Write a handful of pages that answer a real question in the first two sentences, the way you'd answer it out loud to a customer on the phone. Assistants quote clear, self-contained answers far more readily than marketing copy that takes three paragraphs to say what you do.

Get named somewhere you don't control — a local news mention, an industry directory, a chamber listing. Third-party corroboration is the input you cannot manufacture solely on your own site, and it's the one that separates a business an AI trusts from one it merely found.