ChatGPT, Claude and Perplexity each decide who to recommend a different way — but none of them uses a fixed ranking, and none can be paid for placement. ChatGPT builds a shortlist from its training data plus a live search pass. Claude runs web search as a live tool call and attaches an inline citation to the exact passage it draws from. Perplexity lists every source it used in a visible Sources panel, so getting cited is something you can literally count. The mechanics differ enough that a business can pass one engine's bar and fail another's for the same buying question — here's what actually controls each.
ChatGPT combines what it already knows from training with a live search pass, then names a short list — usually two or three companies — with no paid placement and no guaranteed position. Two separate bots feed it: GPTBot supplies the training-time knowledge, OAI-SearchBot powers the live search. The full five-step breakdown of what controls this lives on the get recommended by ChatGPT guide.
Proof: OpenAI states there is no paid placement and no guaranteed position in ChatGPT search. OpenAI — ChatGPT Search.
Bottom line: if either GPTBot or OAI-SearchBot can't reach a page, that page can't feed the shortlist — access is checked before content quality ever gets a vote.
Claude runs its web search as a live tool call and attaches an inline citation — source URL, title and the exact cited text — to every claim it draws from the web, and Anthropic runs three separate bots (ClaudeBot for training, Claude-User for live fetches, Claude-SearchBot for search quality) that the company says all honor robots.txt. That third bot, and the claim that all three respect the same rule, is what sets Claude apart from the other two — see the full breakdown on the get recommended by Claude guide.
Proof: Search Engine Land — Anthropic clarifies how Claude bots crawl sites and how to block them.
Good news: Claude is the one engine of the three where a correct robots.txt is, by Anthropic's own description, a complete answer — no separate real-time exception to plan around.
Perplexity shows a numbered, inline citation for nearly every claim in its answers, with a Sources panel listing exactly which pages it used — unlike ChatGPT's or Claude's shortlist reasoning, that makes citation something you can literally count for a given buying question. Perplexity runs two crawlers: PerplexityBot, the indexing crawler that Perplexity's documentation says respects robots.txt, and Perplexity-User, a real-time agent that Perplexity has described as acting on a user's explicit request rather than as a bot — the one genuine asymmetry among the three engines. Full detail on the get cited by Perplexity guide.
Proof: Perplexity — crawler documentation (PerplexityBot & Perplexity-User).
Check it now: ask Perplexity a real buying question about your category and open the Sources panel — your citation rate is right there, not a guess.
Different mechanics, same four controllable levers: crawler access, plain-HTML facts, unambiguous identity, and a page that answers the buying question directly. A page invisible to JavaScript-blind crawlers, an inconsistent company name across profiles, or a buying question nobody on the site actually answers costs a business the same way on ChatGPT, Claude and Perplexity — covered in full on the entity clarity and answer-first writing guides.
Do this: fix these four levers once, site-wide, rather than chasing a separate playbook per engine.
The direct test is asking each engine the exact buying question a customer would ask, in that engine's own interface, and reading what comes back. Reflexa runs a repeatable version of that same check across ChatGPT, Claude and Perplexity — tracking Mentions, Recommendations and Citations per engine, per Buyer Question, over time — described on the tools overview.
The free check reads your crawler access and shows real citation data across ChatGPT, Claude and Perplexity — 3 minutes, evidence included.