---
title: "Why ChatGPT and Perplexity Cite Different Pages: Follow the Fetch"
description: "A citation starts with a prompt that triggers retrieval. Follow the search, fetch and selection steps to see why ChatGPT and Perplexity surface different pages."
url: "https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/09fc092a-e30d-4ee6-82a2-1213d4ce5412.png"
published: "2026-11-01T06:12:19.772Z"
modified: "2026-11-01T06:12:19.824Z"
---

November 1, 2026 · 10 min read

# Why ChatGPT and Perplexity Cite Different Pages: Follow the Fetch

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![Two dogs run back across a lawn after the same throw, one carrying a dated sheet and one an old book, while a polished binder lies ignored in the grass.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/09fc092a-e30d-4ee6-82a2-1213d4ce5412.png)

In this article

1. [Step 1: Find out whether the prompt sends ChatGPT to the web](#step-1-find-out-whether-the-prompt-sends-chatgpt-to-the-web)
2. [Step 2: Match the search the assistant actually makes](#step-2-match-the-search-the-assistant-actually-makes)
3. [Step 3: Account for the engines’ different retrieval habits](#step-3-account-for-the-engines-different-retrieval-habits)
4. [Step 4: Use logs to separate indexing from live fetches](#step-4-use-logs-to-separate-indexing-from-live-fetches)
5. [Step 5: Give a fetched page a passage worth citing](#step-5-give-a-fetched-page-a-passage-worth-citing)
6. [Choose buyer prompts before you commission pages](#choose-buyer-prompts-before-you-commission-pages)
7. [Run a new question through the pipeline](#run-a-new-question-through-the-pipeline)

Most agencies won’t tell you this: ChatGPT and Perplexity can cite different pages for the same question because they ran different errands, not because one awarded your content a higher score. **A citation is the last step of a pipeline.** The prompt influences whether an assistant searches the web, retrieval determines which pages it fetches, and only a fetched page can be cited in that answer. Skip the fetch and your page was never in the race.

I came to this by watching our server logs instead of reading another list of tips for getting cited. One dated Google Ads changelog page received 569 live-answer fetches from ChatGPT-User in 30 days. That number does not tell me how often the page was quoted. It tells me something more basic: ChatGPT repeatedly came looking for it while answering prompts. The distinction matters when [ChatGPT searches live on only about 34.5% of queries](https://www.geoptimizer.io/blog/how-often-does-chatgpt-search-the-web), while [Perplexity runs fresh retrieval on every query](http://ziptie.ai/blog/how-perplexity-ai-answers-work/). Before you ask how to earn a citation, ask whether the assistant will look at a page at all.

## Step 1: Find out whether the prompt sends ChatGPT to the web

ChatGPT can answer from its existing knowledge without links, or it can use live web retrieval and produce citations. That second path is optional. [When it browses, the documented pipeline includes query rewriting, retrieval, reranking and extraction](https://thepromptinsider.com/aeo/how-does-chatgpt-choose-sources/). The choice to search depends on the prompt; [the prompt-content decision is described here](https://www.geoptimizer.io/blog/how-often-does-chatgpt-search-the-web). **No live search means no live-source citation.** The quality of your page cannot change a decision made before the engine encounters it.

That choice is lopsided. Discovery prompts trigger a search about 73% of the time, review-seeking prompts about 70%, comparison prompts about 65%, and purely informational prompts about 10%. These are patterns, not guarantees for any individual question. But they change what I would commission. I used to tell clients that a great definition page would earn AI mentions. I was wrong to treat that as a citation plan. A page answering _what is X_ competes in a prompt category ChatGPT usually answers without searching. A page answering _best X for Y right now_ gives it a reason to seek current evidence.

Think about our Google Ads changelog. A timeless explanation of what a setting does and a question about what changed in that setting are not the same retrieval opportunity. The second question needs a current record. **Write for questions that make fresh evidence useful**, not simply for topics buyers care about.

## Step 2: Match the search the assistant actually makes

Even when ChatGPT decides to browse, it does not necessarily search for the user’s sentence verbatim. Ask for the best project management tools in 2026 and it may split the request into backend queries along the lines of _best project management tools 2026 comparison_ and _top-rated reviews 2026_. That query-rewriting behavior is part of [documented breakdowns of ChatGPT source selection](https://thepromptinsider.com/aeo/how-does-chatgpt-choose-sources/).

**Retrieval runs on those rewrites, not just on the wording you see in the chat.** A page built around one clever version of the user’s question can miss the category, comparison and date terms the assistant goes looking for. I used to load titles with clever phrasing. Now I put the plain-language version near the top, where both retrieval and extraction have a better chance of finding it. The copy still has to be worth reading. It just has to be findable first.

This helps explain why a dated changelog is a plausible match for questions about a recent Google Ads change. Its date and change notes give retrieval something specific to match. That does not prove which prompts produced our 569 fetches; server logs do not show me the user’s question. It tells me why I would test a page like that before betting on a broad, undated explainer. **Match the likely sub-queries in your headings and early copy.**

## Step 3: Account for the engines’ different retrieval habits

Perplexity is search-first. It [runs fresh retrieval on every query and pulls 60-plus sources on a standard search](http://ziptie.ai/blog/how-perplexity-ai-answers-work/), using hybrid keyword and embedding retrieval before reranking. Citations are embedded into the prompt before generation rather than added as decoration afterward. That is why [numbered sources appear by default](https://www.xcodefix.com/perplexity-vs-chatgpt-seo): the answer is built with retrieved material in view.

ChatGPT is assistant-first. It [cites live sources when web search is used](https://www.xcodefix.com/perplexity-vs-chatgpt-seo), but many prompts never reach retrieval. Once it does search, the pages it selects can differ from Perplexity’s. One analysis of 680 million citations found Wikipedia prominent in ChatGPT citations and Reddit prominent in Perplexity citations. **Getting searched is one difference; getting selected from the search is another.** You cannot solve both by putting “AI-optimised” in a content brief. What the deck calls a strategy, I call a heading with no plan behind it.

The changelog example keeps the distinction honest. Perplexity’s search-first approach gives a current change note an opportunity to enter retrieval whenever the question fits. In ChatGPT, the question first has to send the assistant to the web. Neither path promises a citation. The practical choice is to build a page that clearly identifies the change, answers the likely question and lets each engine assess a usable passage.

## Step 4: Use logs to separate indexing from live fetches

Here is the step you can actually observe on your own site. OAI-SearchBot and PerplexityBot crawl to build indexes in the background. ChatGPT-User and Perplexity-User fetch pages live while answering prompts. A [small-site server-log breakdown](https://dev.to/elirany/what-ai-crawler-logs-showed-us-bots-vs-fetches-4g5f) makes the split visible: in one eight-day sample, it recorded 75 ChatGPT-User fetches against 345 OAI-SearchBot hits. **An index visit says your page may be available. A live fetch says it was pulled into an answer workflow.** Neither, on its own, proves the answer cited it.

Our logs showed one dated Google Ads changelog page receiving 569 ChatGPT-User live-answer fetches in 30 days. A second changelog page received 402 in the same window. Both sat on the same domain, used the same structure and had no new link campaign behind them. I cannot read the prompts from those counts, and I will not pretend every fetch became a citation. I can say the pages attracted repeated live retrieval, which is a stronger signal for this question than a background crawler visit.

That is why I separate the two bot categories before judging a page. If a URL gets plenty of index hits and almost no live fetches, rewriting its opening paragraph may not be the first fix. I would first ask whether the prompts it targets cause the assistant to search and whether its terms match the likely retrieval queries. **If ChatGPT-User never fetches a URL for a prompt set, that URL cannot be cited from a live fetch in that set.** Start your diagnosis there.

## Step 5: Give a fetched page a passage worth citing

Retrieval gets a page fetched. It does not get the page quoted. Selection still depends on whether a passage matches the claim, can stand on its own and fits the engine’s assessment of the source. [One analysis found 44% of citations came from the first third of a webpage](https://thepromptinsider.com/aeo/how-does-chatgpt-choose-sources/). That makes the opening of a page more than an introduction. It is where a short, defensible answer can meet an engine looking for one.

Perplexity puts similar pressure on the shape of the answer. About [70% of its top citations were updated in the previous 12 to 18 months, and 90% placed the answer in the first 100 words](http://ziptie.ai/blog/how-perplexity-ai-answers-work/). The same breakdown found cited pages with few referring domains, so a small site is not automatically out. A useful shorthand from [work on citation signals](https://www.omnibound.ai/blog/signals-that-make-ai-cite-a-webpage) is _findable, relevant, extractable, trusted, fresh_. Miss _extractable_ and the other qualities may not rescue a buried answer.

For a changelog, that means putting the dated change and its practical meaning together near the top, not making the reader work through a long preamble before learning what changed. For a comparison, it means stating the distinction before the extended explanation. **Give the engine a self-contained passage without making the human reader dig for it.** Those are not competing requirements.

## Choose buyer prompts before you commission pages

Start from the fetch, not the keyword list. I sort candidate prompts into three groups before drafting:

- **Time-sensitive prompts:** Questions about current changes, dated pricing or other details that make fresh evidence useful.
- **Comparison and review prompts:** Questions where a buyer needs to distinguish options for a particular use case.
- **Memory-friendly prompts:** Definitions and timeless how-it-works questions that ChatGPT often answers without browsing.

The first two are stronger citation targets. The third may still deserve a page for human readers, but I would not sell it as a reliable way to earn live citations. That is a planning distinction, not an instruction to delete every explainer on your site.

Our logs put a concrete page behind the argument. Dated change notes, pricing breakdowns with a year attached and direct comparisons drew live fetches; timeless explainers did not show the same pattern. That is why [our most-fetched ChatGPT page was a changelog](https://groas.com/post/stop-hunting-for-the-right-prompts-ai-assistants-cite-pages) rather than one of the pillar guides we spent longer writing. Work hours are not a retrieval signal. Unfortunate for anyone selling work hours.

I would make the writing brief specific to the engine without pretending the engines require entirely separate sites:

- **For Perplexity:** Target a specific, current question that you can answer in one passage. A prompt such as _best appointment software for two-chair dental clinics in 2026_ calls for a direct answer near the top.
- **For ChatGPT:** Prioritise comparison and review questions that give it a reason to browse. For _X vs Y for \[use case]_, make the dated evidence, relevant settings and limits easy to find.
- **For neither as a citation-first bet:** Avoid commissioning a pure definition page solely because you expect live-source citations from a prompt ChatGPT is likely to answer from memory.

The page has to serve the question once the fetch happens. But first, **pick a question likely to cause the fetch**.

## Run a new question through the pipeline

Before I approve a page, I list the buyer questions it is meant to answer and check them in order. Is each question likely to trigger a live search in ChatGPT, Perplexity or both? What terms might the assistant search after rewriting it? Can the page give a dated, self-contained answer near the top? Then I watch the live-fetch logs by URL and date. That sequence keeps me from mistaking an index crawl for progress or [a fetch for a citation](https://groas.com/post/250-bot-visits-zero-citations-a-postmort).

Say you run a home services company and want to reach someone asking _best CRM for small HVAC companies with same-day dispatch in 2026_. Apply the model rather than starting with a content template. The year and dispatch requirement make current evidence useful; the word _best_ makes this a comparison. Perplexity will retrieve, while ChatGPT may search because comparison prompts trigger browsing about 65% of the time. Its decision is not guaranteed.

Next, consider the rewrites. A search for current HVAC CRM comparisons may not contain the user’s full sentence, so a page needs plain category and use-case language, not just an exact-match headline. Then comes the fetch: a bot visit in your logs tells you the page entered that workflow, not which tool the buyer chose or whether your passage was quoted. Finally comes selection. A first paragraph that names the relevant options and states a specific dispatch limit gives the engine more to work with than a page that waits until section four to address dispatch. Keep the dated evidence and details consistent further down for a reader who needs to check the answer.

Now check the logs. If the page draws index crawls but no live fetches, revisit the prompt and retrieval fit. If it draws live fetches, work on the passage rather than celebrating a citation you have not confirmed. The broader [reason some pages get seen and others never do](https://groas.com/post/how-chatgpt-and-perplexity-actually-choo) often starts earlier in this pipeline than the writing team expects.

I do this work inside [groas](https://groas.com/) now because the manual version stopped scaling years ago. Watching fetch logs, improving passages and keeping dated pages current across Perplexity, ChatGPT, Gemini and Google is daily execution, not a quarterly audit. It runs alongside paid search for one flat fee per service. For the HVAC question, I would not begin by ordering ten pages. I would begin with the prompt, follow it to the fetch, and make sure the page waiting there answers it. **Start with the fetch.**

## Frequently Asked Questions

### Why do ChatGPT and Perplexity cite different pages for the same question?

The difference lies in the specific search, retrieval, and selection pipeline of each assistant. A citation is only the last step, meaning content must first be fetched by the engine to ever appear in an answer. The quality of a page cannot change decisions made before the engine encounters it.

### What types of questions cause ChatGPT to search the live web?

The frequency of live search depends on the prompt type: discovery prompts trigger a search about 73% of the time, and review-seeking prompts trigger it about 70%. Prompts that require fresh evidence, such as comparisons or current details, are more likely to cause retrieval.

### How does Perplexity AI differ from ChatGPT in its search habits?

Perplexity is search-first, running fresh retrieval on every single query and pulling more than 60 sources during a standard search. In contrast, ChatGPT is assistant-first, meaning many prompts never even reach the retrieval stage.

### Why might a page not be found even if it matches the user's prompt?

The assistant often rewrites the user's sentence into backend queries during the retrieval phase. If a page is only built around the user's clever phrasing, it may miss the specific category, date, or comparison terms the assistant is actually looking for.

### What criteria determine if a fetched page is ultimately selected for a citation?

Retrieval only gets the page fetched, but selection depends on the passage matching the claim, standing alone, and fitting the engine's assessment. Effective content must be findable, relevant, extractable, trusted, and fresh to ensure the answer uses the provided text.

### How can I tell if my content is being used in an AI's answer?

Server logs should be reviewed to distinguish between bots that crawl indexes (like OAI-SearchBot) and those that fetch pages live while generating prompts. A live fetch indicates the page entered the answer workflow, which is a stronger signal than a background index visit.

### What kind of content is best suited for earning AI citations?

The most effective content targets time-sensitive prompts, comparison queries, and review-seeking questions. Pure definition pages or timeless how-it-works explanations are less likely to earn live citations because the assistants often answer them from memory.

### Where should key information be placed on a page that has been retrieved?

Important information should be near the top of the webpage because both ChatGPT and Perplexity look heavily at initial sections for selection. Providing a short, defensible, self-contained answer immediately gives the engine material to use in its output.

## Paid Ads & SEO, Run For You

Businesses pay one flat monthly fee per service, agencies resell it under their own brand, and groas does the work either way.

[See Pricing](https://groas.com/pricing)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","description":"groas is a done-for-you marketing service, not software: elite people working with proprietary technology systems run search ads, Meta ads, SEO and AI search for businesses, and white label for agencies, with a dedicated senior strategist owning every account.","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same#webpage","url":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same","name":"Why ChatGPT and Perplexity Cite Different Pages: Follow the Fetch","description":"A citation starts with a prompt that triggers retrieval. Follow the search, fetch and selection steps to see why ChatGPT and Perplexity surface different pages.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/09fc092a-e30d-4ee6-82a2-1213d4ce5412.png"},"breadcrumb":{"@id":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same#breadcrumb"},"dateModified":"2026-11-01T06:12:19.824Z","mainEntity":{"@id":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"Why ChatGPT and Perplexity Cite Different Pages: Follow the Fetch","item":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same#article","url":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same","mainEntityOfPage":{"@id":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"Why ChatGPT and Perplexity Cite Different Pages: Follow the Fetch","description":"A citation starts with a prompt that triggers retrieval. Follow the search, fetch and selection steps to see why ChatGPT and Perplexity surface different pages.","datePublished":"2026-11-01T06:12:19.772Z","dateModified":"2026-11-01T06:12:19.824Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/09fc092a-e30d-4ee6-82a2-1213d4ce5412.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"chatgpt, page, retrieval, fetch, perplexity, pages, search, question","wordCount":2100,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"My background includes Goldman Sachs and Stanford Computer Science.","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/why-chatgpt-and-perplexity-cite-different-pages-for-the-same#faq","mainEntity":[{"@type":"Question","name":"Why do ChatGPT and Perplexity cite different pages for the same question?","acceptedAnswer":{"@type":"Answer","text":"The difference lies in the specific search, retrieval, and selection pipeline of each assistant. A citation is only the last step, meaning content must first be fetched by the engine to ever appear in an answer. The quality of a page cannot change decisions made before the engine encounters it."}},{"@type":"Question","name":"What types of questions cause ChatGPT to search the live web?","acceptedAnswer":{"@type":"Answer","text":"The frequency of live search depends on the prompt type: discovery prompts trigger a search about 73% of the time, and review-seeking prompts trigger it about 70%. Prompts that require fresh evidence, such as comparisons or current details, are more likely to cause retrieval."}},{"@type":"Question","name":"How does Perplexity AI differ from ChatGPT in its search habits?","acceptedAnswer":{"@type":"Answer","text":"Perplexity is search-first, running fresh retrieval on every single query and pulling more than 60 sources during a standard search. In contrast, ChatGPT is assistant-first, meaning many prompts never even reach the retrieval stage."}},{"@type":"Question","name":"Why might a page not be found even if it matches the user's prompt?","acceptedAnswer":{"@type":"Answer","text":"The assistant often rewrites the user's sentence into backend queries during the retrieval phase. If a page is only built around the user's clever phrasing, it may miss the specific category, date, or comparison terms the assistant is actually looking for."}},{"@type":"Question","name":"What criteria determine if a fetched page is ultimately selected for a citation?","acceptedAnswer":{"@type":"Answer","text":"Retrieval only gets the page fetched, but selection depends on the passage matching the claim, standing alone, and fitting the engine's assessment. Effective content must be findable, relevant, extractable, trusted, and fresh to ensure the answer uses the provided text."}},{"@type":"Question","name":"How can I tell if my content is being used in an AI's answer?","acceptedAnswer":{"@type":"Answer","text":"Server logs should be reviewed to distinguish between bots that crawl indexes (like OAI-SearchBot) and those that fetch pages live while generating prompts. A live fetch indicates the page entered the answer workflow, which is a stronger signal than a background index visit."}},{"@type":"Question","name":"What kind of content is best suited for earning AI citations?","acceptedAnswer":{"@type":"Answer","text":"The most effective content targets time-sensitive prompts, comparison queries, and review-seeking questions. Pure definition pages or timeless how-it-works explanations are less likely to earn live citations because the assistants often answer them from memory."}},{"@type":"Question","name":"Where should key information be placed on a page that has been retrieved?","acceptedAnswer":{"@type":"Answer","text":"Important information should be near the top of the webpage because both ChatGPT and Perplexity look heavily at initial sections for selection. Providing a short, defensible, self-contained answer immediately gives the engine material to use in its output."}}]}]}
```