ChatGPT search vs Perplexity for brand visibility

The short answer

OpenAI and Perplexity both let you decide whether your pages can be used in their answers, and both publish the mechanism. What they publish differs in a way that matters more than any ranking theory: OpenAI gives you two switches and tells you what each one turns off, while Perplexity gives you one crawler whose stated job is already the thing you want, plus a user-triggered fetcher that will fetch anyway.

If you are deciding where to spend limited attention, the split is roughly this. ChatGPT search is where a documented control surface exists and where mistakes are cheap to reverse. Perplexity is where the linked source is the product rather than an ornament on it. Neither publishes what actually decides which of the retrieved pages ends up in the answer, and that silence is the most important thing on this page.

How each platform decides your pages are usable

ChatGPT search

OpenAI's crawler documentation names OAI-SearchBot as the agent "used to surface websites in search results in ChatGPT's search features."[^oai-searchbot-purpose] That is a narrow, checkable statement, and it comes with the consequence attached: sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers."[^oai-searchbot-optout]

The important structural fact is that this is not the same agent as the one that gathers training data. GPTBot is documented separately, and disallowing it "indicates a site's content should not be used in training generative AI foundation models."[^oai-gptbot-training] Two agents, two decisions, no bundling. A site can refuse to be training data and still be eligible to appear in ChatGPT search, which is exactly the position most marketing teams say they want and many accidentally give up with a single blanket rule.

A third agent, ChatGPT-User, covers certain user actions in ChatGPT and Custom GPTs, and OpenAI adds the qualification that "because these actions are initiated by a user, robots.txt rules may not apply."[^oai-chatgpt-user-robots] That is a hedge rather than a promise in either direction, and it is worth reading precisely: it does not say the fetcher ignores robots.txt, it says the rules may not apply.

Perplexity

Perplexity's crawler documentation describes PerplexityBot as "designed to surface and link websites in search results on Perplexity. It is not used to crawl content for AI foundation models."[^ppx-bot-purpose] Both halves of that sentence carry weight. The first is the same job OAI-SearchBot does. The second removes a decision you have to make with OpenAI: there is no separate Perplexity training crawler to reason about, because the operator states this one is not doing that work.

The documentation then does something OpenAI's does not: it recommends allowing PerplexityBot in robots.txt so a site can be included in search results.[^ppx-allow-recommendation] That is a positive instruction, not just a description of an opt-out.

The second agent, Perplexity-User, is where the honest reading gets less comfortable. Perplexity states that this fetcher "generally ignores robots.txt rules" because the request came from a person.[^ppx-user-robots] It also states that when a user asks a question, Perplexity-User "might visit a web page to help provide an accurate answer and include a link to the page in its response."[^ppx-user-links]

So a robots.txt block on Perplexity is not a block on Perplexity. It is a block on the crawler that builds the index, while the fetcher that answers a live question may still arrive. OpenAI's equivalent hedge is softer, but the practical shape is the same on both platforms, and neither is a wall.

The comparison table, and what it will not tell you

The table above compares what each operator has committed to in writing. Every cell points at a claim with a source URL and the date it was fetched, which is a deliberate constraint: it is the only way to write a page like this without quietly asserting things nobody published.

Notice what is missing from it. There is no row for "which platform cites more sources per answer", no row for "average citation position", and no row for "how much content freshness matters". Those rows would sell better and they would all be invented. Neither OpenAI nor Perplexity publishes the ranking or selection logic that turns a set of retrieved pages into the two or three that get named in an answer. Anyone presenting that logic as documented fact is presenting an inference from samples, and inferences from samples of a non-deterministic system have a short shelf life.

What you can act on is the retrieval layer, and the retrieval layer is fully documented on both sides.

Verification, and why the user-agent string is not enough

Both operators publish IP ranges. OpenAI lists OAI-SearchBot addresses at openai.com/searchbot.json[^oai-ip-ranges] and Perplexity lists PerplexityBot addresses at perplexity.com/perplexitybot.json.[^ppx-ip-ranges]

This matters because a user-agent header is a string anyone can send. If your analytics show a surge of AI-crawler traffic and you are counting user-agent strings, you are counting claims rather than visits. Resolve against the published ranges before you draw a conclusion, and before you make a robots.txt change on the strength of it.

Timing: the one operational difference that will bite you

OpenAI documents that "it can take ~24 hours from a site's robots.txt update for our systems to adjust."[^oai-robots-lag] That is a small, unglamorous fact and it is the difference between "the fix did not work" and "the fix has not propagated yet."

Perplexity's crawler documentation, as retrieved on 3 September 2026, publishes no equivalent window.[^ppx-no-propagation-window] That is not a criticism; it is a gap you should plan around. If you unblock PerplexityBot, you have no documented time-to-effect, which means the only honest procedure is to re-check on a cadence rather than to declare a date by which it should have worked.

Citations as an obligation, not a favour

There is one more asymmetry worth understanding, because it shapes how your brand appears in products built on top of these platforms rather than in the consumer apps themselves.

OpenAI's web search guide tells developers that when displaying web results to end users, "inline citations must be made clearly visible and clickable in your user interface."[^oai-citation-display] That is a requirement placed on everyone building on OpenAI's web search, not a feature of ChatGPT alone. Every application built that way is a surface where a link to your page is expected to be present and clickable.

Perplexity's developer documentation splits the difference across two products: the Agent API is described as delivering "web-grounded answers with built-in citations in one call", while the Search API is described as returning raw, ranked web search results.[^ppx-agent-api-citations] A developer choosing the second one gets results without the citation scaffolding attached.

The practical consequence for a brand is that "appearing in Perplexity" and "appearing in something built on Perplexity" are different questions with different answers, and the same is true of OpenAI in the other direction.

What to actually do, in order

1. Read your robots.txt for the four tokens that matter here: OAI-SearchBot,

GPTBot, PerplexityBot and Perplexity-User. Many sites that believe they are open to AI search are blocking one of them by accident, usually through a wildcard rule written years ago for a different reason.

1. Decide the training question separately from the search question. OpenAI

makes that separation available; take it. Blocking GPTBot costs you nothing in ChatGPT search, because OpenAI documents that appearance as governed by a different agent.

1. Verify crawler traffic by IP, using the published ranges, before you draw

any conclusion from your logs.

1. Re-check after 24 hours on OpenAI, and on a repeating schedule for

Perplexity, where no propagation window is published.

1. Stop there. Everything beyond the retrieval layer — which page gets

quoted, in what order, how often — is unpublished on both platforms, and optimising against an unpublished mechanism is how teams spend a quarter on a theory.

How this pairing differs from the neighbouring ones

It is tempting to treat all the big AI surfaces as one problem. They are not, and the differences are documented rather than aesthetic. Google's AI features are governed by ordinary Search indexing and snippet controls rather than by a dedicated answer crawler, which is why Google AI Overviews versus ChatGPT search turns on a completely different control surface from this page. Anthropic's crawlers split along yet another axis, which is the subject of ChatGPT versus Claude for brand visibility. And if your interest is specifically in how a cited source is represented back to a developer, that mechanism differs sharply between vendors, as Claude versus Gemini on citations sets out.

The reason to keep them separate is practical: a robots.txt rule that fixes one of these platforms has no effect on another, because the tokens are different and the eligibility rules are different. One change, four verifications.

Where the honest limit of this page is

This comparison is built entirely from what OpenAI and Perplexity have published about their own crawlers and APIs, fetched on 3 September 2026. It tells you who is allowed to read your pages and what each operator says will follow from that. It does not tell you who wins, because neither operator has published the information that would let anyone say so, and a page that answered that question anyway would be guessing in a confident font.

If your next question is whether any of this changed anything for your brand specifically, that is a measurement problem rather than a documentation problem, and it needs repeated sampling of real prompts on both platforms rather than another table.

[^oai-searchbot-purpose]: OpenAI crawler documentation, retrieved 2026-09-03. [^ppx-bot-purpose]: Perplexity crawler documentation, retrieved 2026-09-03.