Engines

ChatGPT

ChatGPT Search · ChatGPT citations

Definition

ChatGPT is OpenAI’s conversational product. For earned media and SEO teams it is not a chatbot novelty. It is a retrieval-and-synthesis surface people use to ask category questions that used to go to Google. Some answers show source lists. Some do not. Compared with Perplexity, public crawls often find fewer citations per ChatGPT answer — HubSpot, summarizing Fan Out data, put ChatGPT around 3.3 sources per query against Perplexity around 10.8. Treat the exact averages as study-bound. The operational fact is stable: ChatGPT is a sparse citer.

Memory, custom GPTs, browsing mode, and logged-in personalization change what one staff member sees. A tracking program that uses a single Plus account with memory on is measuring that account, not “ChatGPT.”

How it works

People ask in sentences. The system may answer from model weights, from retrieved pages, or from a mix. When sources appear, they are the clip. When the brand is named with no URL, log a mention. When a URL is shown, open it and check the sentence matches. Hallucinated citations happen: a plausible title, a 404, or a page that does not support the claim.

Example: 40 prompts about “SOC 2 audit timeline for a 40-person SaaS firm” are run monthly in a logged-out browser session. In April, 11 answers show a source list. The firm’s methods page appears in 3. Two trade write-ups of the firm’s study appear in 4. The firm is named with no link in 6 more answers. Report those rows separately. Do not convert them into “25% visibility” and compare the figure to Perplexity’s 25% on a different citation base.

Earned journalism still feeds this surface. PR Daily’s citation series found news domains holding roughly a quarter of links across ChatGPT, Claude, and Gemini in three editions. Owned pages still matter when they carry the table the model wants. Paid native posts are a weak bet if the engine can see better independent sources.

How it differs

Perplexity is built to show its work and usually returns a longer source list. Google AI Overviews live inside Search and require snippet-eligible indexation on Google’s side. Claude’s public product often names organizations with fewer outbound links. GEO is the practice. ChatGPT is one engine the practice is measured on.

Common errors

Prompting “recommend us” and calling the answer a tracking hit. Optimizing only for phrases that appear in your ads. Ignoring trade coverage because “ChatGPT reads Reddit.” Changing three product facts on the website and leaving the old facts in a popular review the model still retrieves.

Sources