Perplexity vs ChatGPT for Brand Visibility: How They Cite, and How to Win Each

They differ in which sources they favor, how visibly they cite, and how much access publishers grant their bots. Both are retrieval-augmented generation systems, meaning they retrieve outside pages and write an answer from them.

Our position: run a shared plan plus two engine-specific ones. Access and clear answers help in both. ChatGPT's links skewed to third-party pages, while Perplexity also drew on brand, video and community sources.

Each engine has its own search crawler and a separate fetcher that runs when a person asks. You control the first with robots.txt; the second is only partly governed by it.

• OAI-SearchBot (ChatGPT): OpenAI says it surfaces sites in ChatGPT search, and opted-out sites "will not be shown in ChatGPT search answers," though they can still appear as navigational links.
• GPTBot (ChatGPT): used for training only. OpenAI says each robots.txt setting is independent, so blocking it is a training choice, not a search choice.
• ChatGPT-User: runs for user actions; OpenAI says robots.txt rules may not apply, and it is not used to decide Search appearance.
• PerplexityBot: Perplexity says it surfaces and links sites in Perplexity results and is not used for foundation-model training.
• Perplexity-User: fetches pages for a person's question and, per Perplexity, generally ignores robots.txt.

Both vendors say robots.txt changes take about 24 hours to apply.

One question can trigger several searches: OpenAI describes agentic search that keeps searching, and Perplexity's Agent API guide describes a bounded multi-turn loop. Your page may be retrieved for a sub-question such as price.

In the 2025 preprint, ChatGPT's cited links were almost all earned media, meaning reviews, news and other independent sites, while Perplexity mixed in brand sites and social platforms. The paper is an unreviewed arXiv preprint that queried sonar-pro and gpt-4o-search-preview through their APIs in 2025, so read it as one dataset. For ranking-style brand queries it reports these link shares:

• Well-known brands: ChatGPT 93.5% earned, 6.5% brand, no social. Perplexity 67.4% earned, 8.8% brand, 23.8% social.
• Niche brands: ChatGPT 95.1% earned, 4.9% brand, no social. Perplexity 73.4% earned, 9.1% brand, 17.5% social.

In car queries, YouTube was Perplexity's most frequent source; ChatGPT leaned on Wikipedia, AP News and review sites.

The takeaway: brand-owned domains were only 4.9% to 6.5% of ChatGPT's links here. Reviews, roundups and encyclopedic mentions did the work, so being named there can matter more than polishing your homepage.

Both attach links to answers, but their API documentation exposes different details. Consumer apps may differ, so check the interface yourself.

OpenAI's web search guide says inline citations carry the URL, title and location of the cited source. Its sources list shows every URL the model consulted, and the number of sources is often greater than the number of citations. Being read is not the same as being cited.

Perplexity's web search documentation lists url, title, snippet, date and last_updated for each result, with recency filters from the past hour to the past year.

Citations are not proof of accuracy. The Tow Center's March 2025 test found Perplexity answered 37 percent of its queries incorrectly. It found ChatGPT biased toward wrong answers over no answers, and several tools cited syndicated copies instead of originals. Check which URL is cited for your brand and whether the facts match your pricing page.

Yes, in our sample. In AEOeye's crawler access study (October 2026, 384 sites), 35 sites (9.1%) disallowed PerplexityBot on their homepage, against 16 (4.2%) for OAI-SearchBot. Among 38 news publishers it was 26 (68.4%) against 11 (28.9%).

The block sets are nested. All 16 sites that blocked OAI-SearchBot also blocked PerplexityBot, and 19 blocked only PerplexityBot. For brands the numbers are tiny: 1 of 193 SaaS sites blocked either bot.

Our reading: your own robots.txt is rarely the problem. We have not measured whether blocked third-party sources change what each engine cites.

Limits: a convenience sample, one fetch per site, main host only. A block is a request. RFC 9309 states that its rules are "not a form of access authorization," and the Tow Center found Perplexity identified all ten paywalled National Geographic excerpts it was given although the publisher had disallowed its crawlers. Per-site data is in our CSV; our crawler directory explains each bot.

Make your page fetchable by PerplexityBot, put a checkable answer in the raw HTML, keep real dates, and earn presence where Perplexity's broader source mix looks, including video and community threads. Our Perplexity playbook has all seven steps; the short version:

1. Allow PerplexityBot in robots.txt and permit its published IP ranges in your WAF or CDN.
2. Check the raw HTML. View source and search for your key sentence; if it is missing, render it on the server.
3. Lead each section with a 40 to 60 word answer that names your brand, because its search results carry snippets, not whole pages.
4. Show honest dates. Update the facts first, then the date. No Perplexity document we found states how freshness is weighted.
5. Be present on YouTube and in community threads your buyers read, with specifics instead of promotion.

Verify by asking the question and reading which sources are cited.

Allow OAI-SearchBot, then concentrate on being named in the independent pages ChatGPT cites for your buyer questions. Your own site is necessary but rarely enough. Our ChatGPT SEO guide covers the on-site work.

1. Allow OAI-SearchBot and OpenAI's published IP ranges. Keep GPTBot as a separate training decision.
2. List the domains cited when you ask ChatGPT your unbranded buyer questions. Those roundups, reviews and encyclopedic pages are your targets.
3. Get accurate coverage there with facts reviewers can verify. You cannot buy a slot.
4. Check bot protection. In our study 24 of 88 e-commerce and consumer homepages challenged or blocked our self-identified crawler. A WAF that cannot tell a search bot from a scraper can quietly hide you.
5. Answer directly on your own page.

Four habits help in both engines: let the search bots in, answer in the first lines, keep facts consistent across the web, and test the same questions repeatedly.

• Consistent facts. Use the same name, pricing and claims everywhere, so retrieval finds one story.
• Answer-first pages. One question per section, the answer in the first two sentences, one sourced number per claim.
• Repeat tests. Run 10 to 15 unbranded buyer questions in both engines on different days and record who is named and which domains are cited. Your share of voice is the percentage of answers that name you.
• Track clicks separately. A citation is not a visit; track AI referrals in GA4.

AEOeye's free audit asks ChatGPT three buyer questions and shows the full answers and the competitors named. The $29 full report adds Perplexity, Claude, Gemini and Google AI across ten questions. For daily monitoring of hundreds of prompts, trackers such as Profound, Peec AI and Otterly fit better.