Key Takeaways
- Treat Perplexity as a reader, not a memory. It searches the live web for almost every question, then writes from the pages it just read. What it says about you today depends on what is published today.
- Get retrieved before you worry about being quoted. A page that never comes back in the search step cannot be cited no matter how good it is. Matching the way buyers phrase the question is the first gate.
- Write the sentence you want lifted. Citations come from passages that answer the question on their own. An answer assembled across four paragraphs is harder to quote than one clean sentence.
- Expect corroboration to matter more here. Because the answer carries visible links, a claim that only your own site makes is a risk the engine does not need to take.
- Check Perplexity separately from ChatGPT. They source answers differently, so being named by one tells you very little about the other. Two engines, two results, two lists worth keeping.
Perplexity is the assistant that shows its work. Every answer arrives with a row of numbered links underneath it, and those links are the whole story of how the answer got made.
That makes it the easiest engine to learn from. You cannot see why ChatGPT named three coaches, but you can see exactly which pages Perplexity read before it named them, which means you can read those pages yourself.
This post explains how Perplexity chooses its sources, what a page has to survive to end up in that row of links, and why the answer it gives about your niche is often different from the one ChatGPT gives.
How does Perplexity choose its sources?
It searches the live web for the question it was just asked, collects the pages that come back, and writes the answer out of the passages that address the question most directly. The pages those passages came from become the citations.
The order matters more than it sounds. Retrieval happens first and writing happens second, so a page that does not come back in the search step is not in the running at all. Quality never gets considered, because the page never arrives.
Perplexity's own help center describes the product as an answer engine that searches the web and cites the sources it used. Read plainly, that is a description of a reading machine rather than a remembering machine, and everything below follows from the difference.
The four questions a page has to survive
Four things have to go right in order. Each one removes pages for a different reason, and the reason your page fails decides which fix is worth your afternoon.
In writing, because nothing may live only inside a figure. Fork one asks whether the page came back for that exact wording, and pages that rank for a different phrasing of the same topic drop out. Fork two asks whether the page answers the question directly, and pages that circle the topic drop out. Fork three asks whether anything else says the same thing, and lone assertions nothing else supports drop out. Fork four asks whether a short passage can be lifted whole, and answers spread across four paragraphs drop out. What survives all four gets cited under the answer with a visible link.
Fork one is where most small businesses lose, and it is the least interesting failure. The page is fine. It simply answers "personal training for people with back pain" while the buyer typed "trainer who works with back problems", and nothing in the search step connected the two. The same problem shows up across engines, which is why the pages assistants quote most tend to use the buyer's phrasing rather than the industry's.
Why the answer differs from ChatGPT
Because they get their material from different places. Perplexity reads the web at the moment you ask. ChatGPT answers a great many questions out of what it absorbed during training, and only searches when it decides it needs to.
That single difference produces most of the confusion coaches report. A business that published a good page last month can be cited by Perplexity and unknown to ChatGPT at the same time, and neither result is wrong. They are answers to the same question drawn from two different pictures of the world.
It also changes what a win looks like. Getting into Perplexity is closer to getting a page found and quoted. Getting into ChatGPT is closer to becoming part of what the model absorbed, which is slower and depends more on being written about elsewhere, as the way ChatGPT decides who to recommend sets out in detail.
| Question | Perplexity | ChatGPT | Google AI Overviews |
|---|---|---|---|
| Where the answer comes from | Pages retrieved at the moment you ask | Mostly what the model learned in training, plus search when it chooses to | Pages already in Google's index |
| Are sources shown | Yes, numbered links under the answer | Sometimes, when it searched | Yes, links beside the answer block |
| How quickly new pages can appear | As soon as they are findable | Slowly, and often not at all without a search step | Once indexed |
| What it rewards most | A passage that answers the exact question | Being described consistently in many places | Pages that already do well in search |
| Your first move | Publish the answer in the buyer's wording | Get written about somewhere else | Fix the ordinary search basics first |
What actually wins a citation
A quotable passage on a findable page that something else in the world agrees with. That is the whole recipe, and each third of it fails for a different reason.
Quotable means the answer survives being cut out. If the sentence only makes sense after two paragraphs of setup, it is not a passage, it is a section, and sections do not get quoted.
Findable means the page uses the words the question uses. Not keyword stuffing. Simply writing "trainer for people with back problems" somewhere on the page if that is the phrase people type, alongside whatever you call it professionally.
Agreed with means the claim exists somewhere other than your own marketing. A profile, a directory listing, an interview, a review that mentions the same specialty. Princeton's study of generative engine optimization found that citing sources and quoting authorities improved how often content was used in generated answers, which is the same instinct pointed at your own pages. The source families that keep turning up in coaching answers are the realistic places to put that corroboration.
What to do about it this week
First, ask Perplexity ten questions your buyers actually ask and read the links rather than the prose. The prose is a summary of those links. The links are the market.
Second, open the pages it cited and note their shape. In our own runs the cited page is usually short, specific, and organized as questions with direct answers rather than as an essay. That is a pattern worth copying rather than a rule anyone published.
Third, write one page that answers the single question your buyers ask most, in their words, with the answer in the first two sentences under each heading. Google's guidance on creating helpful content is still the clearest public description of what that looks like, and it happens to describe exactly what a retrieval step can use.
Fourth, put one corroborating mention somewhere that is not your website. One is enough to change a lone assertion into a claim with a second witness, and the second witness is what fork three is looking for.
Then repeat the ten questions a month later with the wording unchanged. Two dated lists of domains tell you more about your position than any single impressive answer, and asking once tells you almost nothing.
Our free check runs twenty-five buyer questions from your niche at $0 and emails a plain-English readout within the hour, which is the fastest way to see where you currently stand. It runs as soon as you ask for it.
Questions we hear the most
How does Perplexity choose its sources?
It searches the live web for the question it was asked, gathers the pages that come back, and writes its answer from the passages that answer the question most directly. The pages it quotes become the numbered links under the answer.
Does Perplexity use the same sources as ChatGPT?
Often not. Perplexity retrieves fresh pages for almost every question, while ChatGPT answers a great many questions from what it already learned during training. The same question can produce two different sets of names.
Why does Perplexity cite a small site over a big one?
Because it is choosing a passage, not a publisher. A small page that answers the exact question in one clear paragraph is easier to quote than a large site that covers the topic across a dozen sections.
How fresh does my page need to be?
Fresh enough to be findable and current enough to be true. Because the search step happens at question time, a page published this month can be cited this month, which is not reliably the case with a model answering from memory.
Does schema markup make Perplexity cite me?
Not on its own. Markup helps machines read what a page is about, but the citation goes to a passage that answers the question. Clear writing does more here than any tag, and the two work best together.
How do I find out whether Perplexity cites me today?
Ask it the questions your buyers ask, in their wording, and read the links under each answer rather than the prose. Write down every domain that appears. Repeat the same list monthly with the wording unchanged, or the comparison means nothing.
Does AnswerHalo check Perplexity as well as ChatGPT?
Yes. The paid audit runs your 25 questions across five assistants including Perplexity and reports each one separately, because a business named by one engine is regularly missing from another.
What does the AnswerHalo free check include?
It costs $0. We ask ChatGPT 25 real buyer questions from your niche, then email a plain-English readout within the hour: how many answers named you, who was named instead, and the first fix. It runs as soon as you ask for it.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →